Show HN: Virsma – Linux Remote Desktop Service

Virsma is a cloud hosted desktop service.

Get your pre-configured Linux desktop machine in the cloud for as little as 7€ per month with no hidden fees or per-usage costs.

Every desktop computer comes with pre-installed software and is fully dedicated to you.

Connect using Remote Desktop Connection clients in Windows, macOS, Linux, iOS and Android.

The first AI model that translates 100 languages without relying on English data

  • Facebook AI is introducing, M2M-100 the first multilingual machine translation (MMT) model that translates between any pair of 100 languages without relying on English data. It’s open sourced here.

  • When translating, say, Chinese to French, previous best multilingual models train on Chinese to English and English to French, because English training data is the most widely available. Our model directly trains on Chinese to French data to better preserve meaning. It outperforms English-centric systems by 10 points on the widely used BLEU metric for evaluating machine translations.

  • M2M-100 is trained on a total of 2,200 language directions — or 10x more than previous best, English-centric multilingual models. Deploying M2M-100 will improve the quality of translations for billions of people, especially those who speak low-resource languages.

  • This milestone is a culmination of years of Facebook AI’s foundational work in machine translation. Today, we’re sharing details on how we built a more diverse MMT training data set and model for 100 languages. We’re also releasing the model, training, and evaluation setup to help other researchers reproduce and further advance multilingual models.

Breaking language barriers through machine translation (MT) is one of the most important ways to bring people together, provide authoritative information on COVID, and keep them safe from harmful content. Today, we power an average of 20 billion translations every day on Facebook News Feed, thanks to our recent developments in low-resource machine translation and recent advances for evaluating translation quality.

Typical MT systems require building separate AI models for each language and each task, but this approach doesn’t scale effectively on Facebook, where people post content in more than 160 languages across billions of posts. Advanced multilingual systems can process multiple languages at once, but compromise on accuracy by relying on English data to bridge the gap between the source and target languages. We need one multilingual machine translation (MMT) model that can translate any language to better serve our community, nearly two-thirds of which use a language other than English.

In a culmination of many years of MT research at Facebook, we’re excited to announce a major milestone: the first single massive MMT model that can directly translate 100×100 languages in any direction without relying on only English-centric data. Our single multilingual model performs as well as traditional bilingual models and achieved a 10 BLEU point improvement over English-centric multilingual models.

Using novel mining strategies to create translation data, we built the first truly “many-to-many” data set with 7.5 billion sentences for 100 languages. We used several scaling techniques to build a universal model with 15 billion parameters, which captures information from related languages and reflects a more diverse script of languages and morphology. We’re open-sourcing this work here.

Something Went Wrong

We’re having trouble playing this video.

Mining hundreds of millions of sentences for thousands of language directions

One of the biggest hurdles of building a many-to-many MMT model is curating large volumes of quality sentence pairs (also known as parallel sentences) for arbitrary translation directions not involving English. It’s a lot easier to find translations for Chinese to English and English to French, than, say, French to Chinese. What’s more, the volume of data required for training grows quadratically with the number of languages that we support. For instance, if we need 10M sentence pairs for each direction, then we need to mine 1B sentence pairs for 10 languages and 100B sentence pairs for 100 languages.

We took on this ambitious challenge of building the most diverse many-to-many MMT data set to date: 7.5 billion sentence pairs across 100 languages. This was possible by combining complementary data mining resources that have been years in the making, including ccAligned, ccMatrix, and LASER. As part of this effort, we created a new LASER 2.0 and improved fastText language identification, which improves the quality of mining and includes open sourced training and evaluation scripts. All of our data mining resources leverage publicly available data and are open sourced.

Something Went Wrong

We’re having trouble playing this video.

Facebook AI’s new many-to-many multilingual model is a culmination of several years of pioneering work in MT across breakthrough models, data mining resources, and optimization techniques. This timeline highlights a few key noteworthy achievements. Additionally, we created our massive training data set by mining ccNET, which builds on fastText, our pioneering work on processing word representations; our LASER library for CCMatrix, which embeds sentences in a multilingual embedding space; and CCAligned, our method for aligning documents based on URL matches. As part of this effort, we created LASER 2.0, which improves upon previous results.

Still, even with advanced underlying technologies like LASER 2.0, mining large-scale training data for arbitrary pairs of 100 different languages (or 4,450 possible language pairs) is highly computationally intensive. To make this type of scale of mining more manageable, we focused first on languages with the most translation requests. Consequently, we prioritized mining directions with the highest quality data and largest quantity of data. We avoided directions for which translation need is statistically rare, like Icelandic-Nepali or Sinhala-Javanese.

Next, we introduced a new bridge mining strategy, in which we group languages into 14 language groups based on linguistic classification, geography, and cultural similarities. We did this because people living in countries with languages of the same family tend to communicate more often and would benefit from high-quality translations. For instance, one group would include languages spoken in India, like Bengali, Hindi, Marathi, Nepali, Tamil, and Urdu. We systematically mined all possible language pairs within each group.

To connect the languages of different groups, we identified a small number of bridge languages, which are usually one to three major languages of each group. In the example above, Hindi, Bengali, and Tamil would be bridge languages for Indo-Aryan languages. We then mined parallel training data for all possible combinations of these bridge languages. Using this technique, our training data set ended up with 7.5 billion parallel sentences of data, corresponding to 2,200 directions. Since the mined data can be used to train two directions of a given language pair (e.g., en->fr and fr->en), our mining strategy helps us effectively sparsely mine to best cover all 100×100 (a total of 9,900) directions in one model.

To supplement the parallel data for low-resource languages with low translation quality, we used the popular method of back-translation, which helped us win first place at the 2018 and 2019 WMT International Machine Translation competitions. If our goal is to train a Chinese-to-French translation model, for instance, we’d first train a model for French to Chinese and translate all of the monolingual French data to create synthetic, back-translated Chinese. We’ve found that this method is particularly effective at large scale, when translating hundreds of millions of monolingual sentences into parallel data sets. In our research setting, we used back-translation to supplement the training of directions we had already mined, adding the synthetic back-translated data to the mined parallel data. And we used back-translation to create data for previously unsupervised directions.

Overall, the combination of our bridge strategy and back-translated data improved performance on the 100 back-translated directions by 1.7 BLEU on average compared with training on mined data alone. With a more robust, efficient, high-quality training set, we were well equipped with a strong foundation for building and scaling our many-to-many model.

We also found impressive results on zero-shot settings, in which there’s no training data available for a pair of languages. For instance, if a model is trained on French-English and German-Swedish, we can zero-shot translate between French and Swedish. In settings where our many-to-many model must zero-shot the translation between non-English directions, it was substantially better than English-centric multilingual models.

Scaling our MMT model to 15 billion parameters with high speed and quality

One challenge in multilingual translation is that a singular model must capture information in many different languages and diverse scripts. To address this, we saw a clear benefit of scaling the capacity of our model and adding language-specific parameters. Scaling the model size is helpful particularly for high-resource language pairs because they have the most data to train the additional model capacity. Ultimately, we saw an average improvement of 1.2 BLEU averaged across all language directions when densely scaling the model size to 12 billion parameters, after which there were diminishing returns from densely scaling further. The combination of dense scaling and language-specific sparse parameters (3.2 billion) enabled us to create an even better model, with 15 billion parameters.

We compare our model with bilingual baselines and English-centric multilingual models. We start with a 1.2 billion parameter baseline with 24 encoder layers and 24 decoder layers and compare English-centric models with our M2M-100 model. Next, if we compare 12B parameters with 1.2 billion parameters, we gain 1.2 BLEU points of improvement.

To grow our model size, we increased the number of layers in our Transformer networks as well as the width of each layer. We found that large models converge quickly and train with high data efficiency. Notably, this many-to-many system is the first to leverage Fairscale, the new PyTorch library specifically designed to support pipeline and tensor parallelism. We built this general infrastructure to accommodate large-scale models that don’t fit on a single GPU through model parallelism into Fairscale. We built on top of the ZeRO optimizer, intra-layer model parallelism, and pipeline model parallelism to train large-scale models.

But it’s not enough to simply scale the models to billions of parameters. In order to be able to productionize this model in the future, we need to scale models as efficiently as possible with high-speed training. For example, much existing work uses multimodel ensembling, where multiple models are trained and applied to the same source sentence to produce a translation. To reduce complexity and compute required to train multiple models, we explored multisource self-ensembling, which translates a source sentence in multiple languages to improve translation quality. Also, we built on our work with LayerDrop and Depth-Adaptive, to jointly train a model with a common trunk and different sets of language-specific parameters. This approach is great for many-to-many models because it offers a natural way to split parts of a model by language pairs or language families. By combining dense scaling of model capacity with language-specific parameters (3B in total), we provide the benefits of large models as well as the ability to learn specialized layers for different languages.

On the path toward one multilingual model for all

For years, AI researchers have been working toward building a single universal model that can understand all languages across different tasks. A single model that supports all languages, dialects, and modalities will help us better serve more people, keep translations up to date, and create new experiences for billions of people equally. This work brings us closer to this goal.

As part of this effort, we’ve seen incredibly fast-paced progress in pretrained language models, fine-tuning, and self-supervision techniques. This collective research can further advance how our system understands text for low-resource languages using unlabeled data. For instance, XLM-R is our powerful multilingual model that can learn from data in one language and then execute a task in 100 languages with state-of-the-art accuracy. mBART is one of the first methods for pretraining a complete model for BART tasks across many languages. And most recently, our new self-supervised approach, CRISS, uses unlabeled data from many different languages to mine parallel sentences across languages and train new, better multilingual models in an iterative way.

We’ll continue to improve our model by incorporating such cutting-edge research, exploring ways to deploy MT systems responsibly, and creating the more specialized computation architectures necessary to bring this to production.

Get it on GitHub:

Read the paper:

Our Consensus Reality Has Shattered

We will remember 2020 as many things. The year we spent alone. The year we spent online. The year so many died. The year of protests. The year of QAnon. The year of domestic terrorism. The year of the election.

Most of all, perhaps, it is the year of not knowing. Is it safe to send my kids to school? Can I go to the store? Should I vote by mail? Do I still have a job? Is it safe to go to work? Can I afford to stay home? Is it safe to exercise? To fly? Do I still have to wipe down the mail? The groceries? What does the CDC say about that? Can I trust the CDC anymore?

A whirlwind of uncertainty landed on us this year, and it threatens to rip the country apart. We have been struck by an unexpected and little-understood disease, explained in wildly contradictory terms by doctors, politicians, pundits, friends, families, and internet weirdos. The pandemic is an enigma unfolding in real time, where yesterday’s certitudes become tomorrow’s grave mistakes.

All of this is taking place within a profoundly broken information ecosystem. We grope, blindly, forced to independently assess a bewildering barrage of seemingly factual claims that arrive on our doorstep daily, with the lives of our children, parents, lovers, and neighbors hanging in the balance.

All of this is bad enough on its face, but its secondary effects could be disastrous. When people don’t know what’s real, they turn to others for reassurance. But in a world overrun by social media, that process results in a smorgasbord of confusing and conflicting inputs, a problem deepened by the Trump administration’s relentless three-and-a-half-year assault on the very notion of truth.

More by this writer

When no clear, authoritative source of truth exists, when uncertainty rages, human nature will lead many people to seek a more stable reality by wrapping themselves in an ever-tighter cloak of political, religious, or racial identities. The more uncertainty rises, the more alluring that siren call becomes. And some Americans are responding by seeking out exclusive, all-encompassing identities that are toxic and fragile—and hold the seed of violent extremism.

We don’t like uncertainty. We’re wired that way. It’s a survival trait. We need to know. But our knowledge is incomplete, our senses fallible. We can’t always answer the important questions. When that happens, we seek a gut check from the people around us.

The gut check goes by various formal names—constructivism, the social construction of reality, or simply consensus reality.

The sociologist Anthony Giddens wrote that our perception of reality depends on feedback from people we trust. We have to check our facts against the perceptions of others. The more people who agree on a fact, the more we understand it to be real. “Knowledge resides in consensus, rather than in any transcendent or objective relationship between a knower and that which is to be known,” Giddens argued in 1991, although the idea goes back further still.

Objective reality is presumed to exist, and it may enforce its strictures tangibly—for instance, through COVID-19 death tolls and hospitalizations. But objective reality is apprehended through consensus. We do not set out, individually, to count the dead. We trust others to do it for us. When our enveloping social consensus agrees that 200,000 Americans have died, it becomes a fact. It becomes real.

Discerning the consensus has never been a perfect solution to uncertainty, because it’s never entirely clear who can be trusted, and even those we trust may let us down.

The consensus may be objectively wrong. Everyone may agree the world is flat, but that doesn’t mean it’s true. The consensus may be unstable. Many people once agreed that the world was flat, and now most people agree that it is not.

Perhaps most importantly, the nature of the consensus depends on who you know. Even today, surrounding yourself with people who believe that the world is flat is eminently possible. The more people you know who believe it, the more likely you will believe it as well. But if you move, or make new friends, the consensus may again change around you.

The instability of the consensus has always presented a challenge, but in today’s globally networked world, realities collide around us every day, sometimes dramatically—even violently—opposed in their verdicts on values, opinions, and facts.

For three years, the Trump administration’s drumbeat of lies and manipulations has eroded confidence in government data. The accuracy and truthfulness of the federal government on matters large and small came under constant assault, literally from day one, on issues such as national security, foreign diplomacy, and even the humble weather forecast.

The government is the custodian of a remarkably large amount of mundane information about the weather, public health, crime, and the economy. These data points normally tick along invisibly, underwriting a stable consensus and a consistent picture of living conditions in America.

Those days may be over. The fact that it happened in front of our eyes did not diminish, but rather accentuated, the impact. A president who was capable of drawing a new storm track on a weather map with a Sharpie was capable of anything. More than capable of inundating the nation with lies by the thousands. More than capable of training an entire nation to question everything and trust nothing.

But our deteriorating consensus reality didn’t start with Trump. The president is as much its product as its author.

The rise of the internet, and especially social media, had already created a volatile and unwelcoming environment for the idea of objective truth.

Communities of specialized interest and questionable intention sprang up like weeds, as they always had, but now they grew faster and more capable of bridging geographic divides, nourished by a stream of algorithmic rankings and automated recommendations. Credibility could be granted through a blue checkmark or earned through earnest prolixity, or if all else failed, purchased from retweet and follower farms.

Social media revolutionized the art of finding consensus. The numeric nature of such platforms lent itself to easy scoring, and the business incentive of the firms that operated them was always to boost every point of view, to give credibility to every opinion and theory and fantasy. These sites operated as judges with their fingers on the scales, inexorably biased toward indiscriminately promoting content, any content, all content. All clicks were created equal. All posts were entitled to a shot at virality.

They monetized the consumption of content, with business models built only to amplify. That amplification was predicated on engagement, and engagement was explicitly framed as evidence of an emerging consensus. Bookmarks became favorites and favorites became likes. And despite what anyone says, retweets are endorsements—when viewed statistically at scale. The social media companies made fortunes from virality and engagement, becoming both cultivators and arbitrators of consensus, and thus of knowledge itself.

How much engagement does it take to make an alternative fact credible? One hundred thousand retweets? Fifty thousand likes? Ten thousand shares? These numbers were within reach for virtually everyone, and even they are overkill. For some people, seeing 100, 50, or 20 is enough. In a small group—a chat room or a Telegram channel—affirmation from 10 people might be sufficient to tilt someone toward violence, because consensus is more powerful when it is found among others you trust. We listen most closely to chat members, friends, family, and colleagues. We value most dearly the opinions of people from the same neighborhood, or from the same religion, or from the same race.

The health of consensus reality was dire enough before the arrival of the novel coronavirus.

The pandemic would have presented a challenge in any information environment. COVID-19 was new and not well understood. Scientists tested and reported on hypotheses in real time, even as people started to die. Amateur epidemiologists sprang up by the thousands to “educate” their peers about the virus’s threat, or lack thereof, but even good information was bad.

News outlets swarmed toward important and necessary scientific research that came with caveats you could drive a truck through. These studies were crucial steps toward developing real knowledge through the scientific method, but most were miles away from being settled knowledge.

Many Americans consumed them voraciously, seizing on every hint of peril or glimmer of hope, only to face contradictory guidance a few days later. Amateurs and professionals alike argued over which studies were better, leveled by the online playing field.

But while good information was bad, bad information was much worse.

New COVID-specific conspiracy theories exploded across social media, even as the president of the United States suggested that people inject themselves with disinfectant. Nation-states and domestic political actors sought to exploit the chaos with disinformation campaigns, even as a wave of nationwide protests further complicated the information landscape.

Early concerns about the politicization of the Centers for Disease Control and Prevention matured into justifiable panic as the Trump administration actively undermined the work of government scientists in the middle of the crisis. Even less irresponsible figures contributed to the uncertainty—for instance, by telling Americans that masks would not help contain the virus, only to later reverse that advice.

In addition to the mystery of the disease itself, its secondary effects dramatically increased uncertainty for most of the country—suddenly and unexpectedly wiping out millions of jobs and setting the stage for a massive surge in evictions, poverty, and homelessness, distributed unequally across demographic groups.

In the context of mysterious lights seen in the sky, uncertainty can be fun—but most of the time, it’s not so abstract. Mysterious lights are rare, but we think every day about how we’re going to pay the rent.

COVID-19 pushed a teetering nation off a cliff of uncertainty, leaving Americans with a staggering number of questions, worries, and unknowns.

Consensus is a tool for reducing uncertainty, so it becomes much more important during times like these. But in the current information environment, the search for consensus is fraught. When we reach out to others for a gut check, we find a new level of chaos—multiple competing realities, often in violent conflict: Masks are good. Masks are tyranny. Vaccines will save us. Vaccines are dangerous. Trump is making the pandemic worse. Trump is saving the economy.

There are a million other points between these poles, and to the left and right of them. What the consensus looks like depends on whom you talk to.

Some people are better at living with uncertainty than others and can navigate a landscape of contradictions more comfortably. But most of us will seek to reduce uncertainty by turning to the people we trust the most: people who are like us, people with whom we can identify, what social scientists refer to as an in-group.

The in-group is not a designation of power or popularity. It’s simply your group. Anyone who’s not in your group is part of an out-group. In-group identities can be defined in any number of ways, but the most common involve politics, nations, religions, races, genders, and sexual orientations.

Scholars have long understood that people tend to have favorable feelings about others who are part of the same in-group. We identify with in-groups because we understand that they are filled with people like us—who hold similar opinions, listen to similar music, enjoy similar foods. Because they’re more like us, we relate to them more easily and agree with them more often than we do members of our out-groups.

A related effect is equally venerable, but less understood. People who associate with in-groups tend to develop negative attitudes about out-groups. We like our music and don’t like theirs. Our food is good; theirs is not as good. This often extends to the quality of the members: Our people are better than they are.

These are not universal reactions. They’re tendencies that become visible in aggregate, when examining masses of people, and they’re typically weak ones.

For example, in-groups are fluid and easy to shift. If you simply tell a group of people that they’re all on the same team, they will feel more positive about their fellow team members. And in-groups don’t necessarily develop negative feelings about out-groups, even when the groups are competing for resources or status. As the social psychologist John T. Jost has demonstrated, people usually favor maintaining the status quo over changes that might benefit their in-group, an effect called “system justification.”

When the status quo is upended, as in a civil war, people experience massive uncertainty. When the status quo collapses, there is no system to justify. But even short of societal collapse, the system-justification impulse can fail. What happens when the status quo is not just beset by uncertainty, but is itself the source of uncertainty?

That’s when things get ugly.

When we elect to join an in-group, we are also subscribing to its consensus reality, stabilizing our own lives within a communal understanding of what is true.

Identifying with an in-group is not simply a way to ascertain our place in the world; it selects and affirms the world itself. It makes the world real.

During times of great uncertainty, our need to make the world real and know what is true becomes much more urgent, and we can satisfy that need by immersing ourselves ever deeper in an in-group that offers a clear, authoritative consensus.

The social psychologist Michael A. Hogg found that feelings of uncertainty make people more likely to strongly identify with in-groups.

But Hogg’s findings go further. People who are experiencing uncertainty tend to assign a higher value to the in-group’s most distinctive traits, such as skin color or religious practice. They are attracted to in-groups with rigidly defined rules and boundaries, and to in-groups that are internally homogenous—filled with people who look, think, and act in similar ways.

More destructively, people who are experiencing uncertainty tend to develop hostile attitudes toward out-groups, seeing them as threats, and entertaining dark fantasies of hostile actions toward the hated other. Some in-group members may go beyond fantasy, engaging in acts of violence, terrorism, even genocide. They gravitate toward social movements that are bigoted, hateful, and authoritarian.

They become extremists.

During times of great uncertainty, the in-group consensus can become an overwhelmingly powerful anchor to stabilize reality. Conditions in the world might seem to be changing, but the in-group is portrayed as an island of constancy, a historical through line, a fortress. And woe to any threat perceived outside the gates.

In some circumstances, an out-group’s consensus is understood to threaten the very nature of reality. To members of the in-group, such threats seem existential in the broadest sense. They undermine everything, attacking reality itself. Such threats must be crushed.

During times of low uncertainty, an in-group’s consensus reality can often tolerate contact and even friction with conflicting out-group consensuses, especially when the stakes are low, as in disagreements over the proper way to dress, or the correct way to prepare a potato.

The center tends to hold, until it doesn’t. From time to time, things blow up, unleashing violent extremism in highly destructive waves. The policy crowd has favorite explanations for these waves—unemployment, poverty, lack of education. At best, these factors play out in specific local arenas, but when you try to apply them globally, causality mostly falls apart.

Unemployment and poverty do not drive extremism directly. People can live with deprivation if they know what’s expected, where they fit into the picture, and how they will survive, if only barely. They can live with adversity if they can plan for it.

But when unemployment and poverty surge unexpectedly, overturning the status quo, when hopes and dreams and long-laid plans fly out the window, extremism becomes much more attractive. When uncertainty overtakes the system itself, when the system is the source of uncertainty, things can really fall apart, and it becomes difficult to know which way society will turn.

We’re in such a moment now, as the world grapples with profound complexity. The inherent uncertainty about COVID-19 and the sudden decimation of national and global economies have created interlocking storms of misinformation, disinformation, and conspiracy.

In-groups have become vital to establishing what is real, but the normally overlapping circles of consensus have drifted apart, and the less they overlap, the more divergent our realities become.

Donald Trump is an obvious flash point in this multiverse, the biggest and loudest point of divergence in the path that leads from Universe A to Universe B (or C, or D, or E, or F).

On a near-daily basis, Trump mobilizes online armies to battle over what is real. The crowd was small. The crowd was big. Trump hates the troops. Trump loves the troops. Trump is corrupt. Trump is draining the swamp. Trump is a racist. Trump is the best president Black people have ever had. 

Most of us marvel at these incongruent realities, at the shocking ability of some Americans to disregard facts that seem objective and uncontested. But consensus reality rests on identity, and Trump has tripled down on calls to identity.

Even the refuge of the in-group is fraught with questions and division. Trump’s calls to identity are generally not explicit, but implicit—if only just below the surface. As a result, no clearly dominant in-group reality can be invoked. Instead, the process of defining group identity has become a life-and-death competition, fertile ground for extremist movements of every stripe.

This is clearly visible in the diversity of extremist in-groups and movements now on display in America. As the traditional anchors of stability fail, one after another, more and more people are publishing their own pitches for the new consensus reality, and their scripts are getting wilder by the day.

Among “siege culture” neo-Nazi groups, the uncertainty has strengthened identification with broad racial in-groups while dividing extremist subgroups, resulting in schism after schism. Each splinter group now tries to outdo the next in the violence of its rhetoric, seeking the magic recipe for recruitment and mobilization.

For groups such as the multifaceted “boogaloo” movement, the uncertainty serves as a beacon in itself. Uncertainty is their in-group. They share a commitment to tearing everything down and building something new, even if they don’t agree on what.

These are just two of a menagerie of groups and movements, including Incels, Proud Boys, Patriot Prayer, Oath Keepers, Three Percenters, whatever Ammon Bundy is this week, and many more. These groups are frequently in competition, always in evolution, and sometimes in open, escalatory conflict.

And that’s just the right wing. While left-wing extremist groups continue to maintain a comparatively minor level of activity and ideological development, that isn’t guaranteed to remain the case. Most scenarios for such a shift are still highly speculative. But the growing hurricane of uncertainty is likely to swamp all quarters of the American political landscape before it recedes, and it’s impossible to know exactly what’s coming next.

Perhaps the most striking example of the muddled battle to make the world seem real is QAnon, the sprawling, near-impenetrable conspiracy theory that claims President Trump is leading a secret war against an ill-defined cabal of elite pedophiles, among many other things.

Conspiracy theories tend to reduce uncertainty by explaining why the world is the way it is. They’re especially useful for extremist movements, since they often blame negative developments on secretive out-group activities.

Conspiracy theories engulf adherents in a robustly detailed reality that can be highly resistant to contradiction, in part because the conspiracy is understood to be concealed by design, and in part because adherents invest massive mental resources in what the historian Richard Hofstadter called “heroic strivings for evidence,” to render their beliefs rationalist, resulting in a version of reality that “is far more coherent than the real world, since it leaves no room for mistakes, failures or ambiguities.”

Conspiracy theories of all sorts are thriving in the age of COVID-19, but QAnon has some distinctive features. It’s especially contagious, partly as a result of adherents’ adept exploitation of social media. It’s especially immersive, because of its adherents’ “heroic strivings” to create a worldscape of staggering complexity. And QAnon is especially resistant to contradiction, as evidenced by the sad, steady stream of posts about true believers being disowned by their families.

QAnon adherents have become violent on a number of occasions, but the movement generally hovers around the indistinct border between conspiracy theory and full-blown extremist movement, in part because its in-groups and out-group are weakly defined.

Q may seem like a far-right movement because of its association with Trump, but polling suggests a more complicated complexion. The parameters of its out-group are equally vague, redolent of anti-Semitism but not precisely aligned. The out-group potentially includes anyone, even Tom Hanks,  the pope, and the employees of your local pizza joint.

A more important distinction between Q and traditional extremism is the fact that its adherents are not broadly committed to taking hostile or violent action against its satanic, cannibalistic, pedophiliac out-group—because they believe that Trump is already fighting a successful war on the movement’s behalf.

That could change if Trump loses in November, or succumbs to the coronavirus. Adherents may feel a need to take matters into their own hands. Even a Trump victory might not stop the movement from escalating. Q adherents have already made inroads into conventional politics. Imagining the apparatus of the state being deployed to support their cause requires no great leap.

If you’re outside the consensus, it’s tempting to dismiss QAnon and other fringe or extremist movements as outliers, oddities, or even part of a mental-health crisis. The beliefs of their adherents are so unmoored from the dominant consensus that outsiders find them difficult to understand and take seriously.

But in a recent poll, 32 percent of respondents said they believed that QAnon was at least partly true, and a third of all Republicans said that it was mostly true. The movement can almost certainly claim millions of full or partial adherents—a massive in-group, more than large enough to establish its own consensus reality. Belated moves to kick the movement off the major platforms may limit its growth, but the damage has long since been done.

People who believe in QAnon are following a predictable pattern of human behavior, finding certainty in a perspective shared by countless others. QAnon may be many things, including objectively false, but it’s not madness, at least not exactly. Instead, it’s a dramatic demonstration of the power of consensus, the power of knowing that other people see the world exactly the way you do.

Where do we go from here?

The November election is an obvious inflection point. Donald Trump is the chaos candidate, for whom uncertainty is not a bug, but a feature. Joe Biden is the candidate of the pre-Trump status quo, an improvement to be sure, but no panacea.

America’s descent into uncertainty preceded Trump, and neither COVID-19 nor social-media platforms will vanish if Biden is elected. The challenger might mitigate some uncertainty, but his election could mobilize some extremist in-groups to violence, with Trump poised to throw gasoline on the fire.

And those are best-case scenarios. Rather than save us from uncertainty, the election is all too likely to increase it, at least in the short term. Key states might be too close to call on Election Night, partisans may allege rampant fraud, and the incumbent might simply refuse to leave.

Worse still, the news that Trump has been infected with COVID-19, and that many more in the White House have been exposed, holds out the prospect of more urgent chaos, potentially fueled by the administration’s heavy-handed efforts to manage the situation, and the unreliability of the information it offers. The state of the president’s health is more uncertain than the status of Schrödinger’s cat, and the wisdom of his choices during this time will be debated forever, no matter the outcome.

Eventually, hopefully, the smoke will clear, and America will get down to dealing with the aftermath of 2020 in a serious way. When the time comes—if the time comes—we need to be ready with new solutions.

The next generation of initiatives to counter violent extremism must start at the level of stabilizing consensus. That’s a tall order in a society where dramatic change is needed in multiple areas of policy—most notably racial justice and health-care reform. The trick will be finding ways to cushion necessary changes, clearly communicating what the public should expect, and then delivering on those expectations.

Messaging initiatives may also help. Empirical research shows that certain kinds of messaging are effective at swaying opinions and preventing people from joining extremist movements. These messages—which must come from every level of a reasonably unified government and society—should focus carefully on reducing uncertainty, whether by rebuilding institutional trust or providing clear explanations of expected outcomes. Such messages may not always be obviously relevant to extremism, but they are relevant to the conditions that aggravate extremism.

Finally, we must set about the laborious and painful process of rebuilding a society in which overlapping consensus realities can coexist with less friction. There will never be a perfect outcome. Extremism has been with us for nearly as long as people have gathered in groups. But it doesn’t have to be this bad.

For these efforts to stand a chance, the obvious obstacles to rebuilding a common consensus must be overcome. Americans will need to tackle social-media platforms and internet companies, as well as their business models, which blindly and amorally prioritize engagement, virality, and faux consensus. Although the major platforms continue to improve their efforts to manage extremism, the pace of change has been glacial.

We need tech leaders who want to change the system at its roots, instead of plastering their products with barely effectual Band-Aids and wringing their hands when asked to confront political power. The tech sector’s capitulation to the Trump administration’s blizzard of lies and manipulation cannot simply be followed by a similar capitulation to a theoretically more beneficent Biden administration.

The companies that control the public debate must be willing to value truth over clicks, and to address their broken and destructive business models.

Donald Trump may or may not leave office in January 2021, but Mark Zuckerberg and Jack Dorsey will almost certainly remain. They helped bring us to the precipice. The best thing they can do now is step aside and let a new generation of leaders walk us back.

Social Media Marketing Basic Guide for Entrepreneurs

What is Social Media Marketing?

Social Media Marketing is a process of using social media platforms such as Instagram, Twitter and Facebook to connect with the audience. It also involves building brand, driving sales and increasing traffic to brands website.

The Social Media Strategy is the thought out plan of posting great content, interacting with the audience through comments and direct messages. It also includes using statistics to analyse results as well as running ads on different social media platforms.

As an entrepreneur to entrepreneur I can tell you that the best tip would be to start and focus on Social Media Marketing right now. Don’t give yourself excuses.

Start now. Get perfect later.

Little Story

I love social media and social media marketing. You fall in love with it after you grow your page from 0 followers to 100k followers just by doing hard work(with solid plan).

There has been more and more scams and scam-like activity around Social Media Marketing. As an Entrepreneur it has been harder and harder to find good sources of explanation and inspiration.

I want to tell you, only my personal opinion on how exactly to do it right and how not to fail.

This is a full and very professional guide on Social Media Marketing, make sure to get the most value out of this.

1 — The essentials.

2 — How to do it right.

3 — Industry’s best practices.

4 — How to sell.

1 — The essentials:

The building blocks of any Social Media Marketing strategy are the trends and patterns of user behaviour on a specific platform. We will be looking at Instagram here.

This can be applied to any social network or any website/app in general as these are the most important fundamentals.

The most important question people ask is — how do you get followers.

Value matters? Value matters.

It is easy. You simply give your followers value.

As soon as you open Instagram profile, you read introduction. You need to see what value you will get.

Then, you open one of the posts, you read it with interest and you send it to someone(maybe information is that useful). You save it for later, for reference.

All of this is amazing for Instagram algorith

When you open a story, you engage with them. After all the stories are finished, you think to yourself:

Dam, I learnt something. I will need to come back for some more.

This is value.

You can also think of it that you get an active reader, listener or just a follower in exchange for the work you put for the content on your page.

So, here is the deal:

You worked on that content, you put your time. In exchange you get a follower.

How direct ads affect your followage count:

If a potential follower sees an ad in your account, he will leave.
If your timeline is only ads, he will leave.
If your stories are only ads, he will leave.

All of us Entrepreneurs are going into Social Media Marketing for the same reason.

To sell.
To promote our product.
To show why the service we provide is the best

I have been trapped in the same thinking pattern as well. I had no results and I didn’t know what to do.

The change came when I started experimenting and thinking different. That is when the growing happened and when the sales started rolling in.

The thing I changed was:

I started giving value.

The hardest part is realising that just selling won’t work.

The standard on Instagram is increasing every month. Your content needs to be better and better, but even a small amount of value makes a huge difference.

This blog post is also sharing value. I am expecting that some of you will find this useful and will subscribe to the newsletter, follow me on Twitter and join my channel on telegram. 👍


How to be human:

One of the key principles is to show that you are also a human being.

You have to be human. If you behave like your profile has no soul and you are a corporate machine, you will have no way of appealing to potential followers.

Corporate = ignored.
Human = followed.

The Easiest way to show that you are human is to show yourself behaving like one.

be human

How does that help you?

You need to reach out and interact. Do some unsexy work that will not look good and will take time and effort.

It is not scalable. It is not something people brag about. But it is what matters.

Investing into ADs is a good way to scale and get new customers.
But it is not the way to do Social Media Marketing for Entrepreneur doesn’t matter how rich and successful your company is.

Every single successful company have organically scaled their following.

ADs are the best to boost your engagement with the following that is already established.


Be smart.

If you are in some niche which is taken by at least 10-200 other businesses for your target audience, YOU will have to go out and talk.

What do I mean by Talk to people:

  • Follow people
  • Make comments relatable to the posts(don’t make them spammy)
  • Send Direct Messages telling people what you like and what you appreciate

If you are doing all of this with correct people, that is at least 100 potential customers/hour.
Let’s say 5 hours per day, 5 days a week and you have:

2 500 customers per week
10 000 customers per month
120 000 customers per year

If you can afford on skipping this much potential clients, you do not need me to tell you what to do or how to do it. 😉

In my opinion, if you are starting out, you are not allowed to skip this part.

After you did all that, do not forget.

It will take time.

How much work it requires?

If you are searching for help you have probably realised that there is no such thing as free lunch. You need to work for everything.

If you are smart you can take shortcuts, work more efficiently and smarter and it will take 100x or 1000x less time.

But it will not be free.

If you are running a REAL business, it will be unsexy and most likely boring at first. But what will be coming after is amazing.

As a side note. Please don’t believe in success overnight. It doesn’t exist and you will only waste your time. If someone is selling a course on How to get rich quick it is already a dying niche and all the profits are taken(think fidget spinners).

phone with apps

2-How to do it right.

There are a couple of different ways to look at content on Instagram and the way to get results from it.

diagram of 2 Instagram accounts with 2 different strategies. Consistency is atomic habits and quality is something like IMB or Microsoft, actually chose Apple.

It is either you post consistently with decent content or you post less often but with content that blows your mind.

I am afraid that the second option doesn’t generate nearly enough engagement and impressions. The insane hard work doesn’t pay off by itself.

What matters now is consistency and quality all together.

What you need to do is post content that is interesting, engaging and relevant to your followers but as often as possible.

You wonder how often you should post?

The magic number has been 1 post per day. At least for every single SMM professional and me, I speak to. It allows the post to get traction as well as get promoted in the timeline to the relevant followers who have missed the post.

Any more than 1 post per day and your posts will be competing with each other in the timelines of the followers. That means some content will be buried and never seen or engaged with.


How about Stories?

For stories, there is also a magic number, but it changes depending on your content.

If you are an entrepreneur and you are running a blog where you personally show face, then the good number of stories would be 15-25 but not more.

If you are a corporate account showing deals, testimonials and photos and videos of your products than 5-10 is a good number per day.

Also, a great tip on Instagram Stories is to delete all old still alive photos and videos once you are starting a new day with a new set of stories to share.

What really matters with content

Every time a follower sees a post from a corporate account, a user is looking for an excuse to unfollow you.

Posting bad content creates a chance that user will unfollow you.

Your most important job is to not give followers excuses.

So here is the deal:

Great content is the main reason why people share posts. This will get you more followers.

Or they can get annoyed and just unfollow.

Every time you create great content for post or story you might get no traction and engagement. Especially if you do it using no planers for time and topics.

But if you put bad content. It is only a matter of time until there is a huge drop in engagement, impressions and ultimately followers.

You are probably wondering how does Instagram algorithm work:

The Secret of Instagram Algorithm is actually simple. It was said in one of the Instagrams personal blog posts for the Japanese community.

It looks at how many people got impressions of your content and how many of them engaged with the content. If the ratio is good, then Instagram promotes it to the rest of your followers, doing check every time it is shown to see if it is still relevant.

take more photos

In other words:

People saw your post 1000 times(impressions) and only 10 people pressed like. That is 1% engagement rate!

Do you have 100 000 followers? And you only got 10 likes? That is 0,01% followers/engagement rate. Instagram will never promote that.

That is why you should never use bots!

Bots are bad.

I get it. You just started out with Instagram. You have no budget for expensive ADs. Your page looks irrelevant with small followers to count.

Those are not excuses to get bots. Those are reasons to work harder and harder every day on your account.

There is also an incorrect understanding that your corporate page will be more strong, will be more trustworthy to the new customers and will convert better when you get bots to boost followers numbers. That is wrong and incorrect logic.

Any normal Instagram user checks posts before following as there is an overload of information and the users are very-very picky about what they choose to consume.

That is why if you have 5 thousand followers but 5-10 likes that shows you even worse than having no followers and being an underdog.

Content is key

To sum up:

Bad content is going to be shown less and less.

Good content will be shown more and more, getting more impression, reach and organic sharing and in the end – followers.

If you are struggling to post quality content every day, skip a day.

Don’t feed junk to your followers, organic followers don’t deserve it.

Paid followers are too expensive to be neglected with bad posts.

content is key

3 — Industry’s Best Practices.

Influencers are hot

Influencers have been a hot topic for the past 5 years, since the invention of this term.

If you want to get far on Instagram, you have to interact with influencers.

There are a couple of ways to do it.

I have written a full post about Nano Influencers and why I think they provide the best value.

The categories of Influencers:

Nano-influencers: 1,000 – 10,000 followers ✅ Micro-influencers: 10,000 – 50,000 followers 😉 Mid-tier influencers: 50,000 – 500,000 followers 🤤 Macro-influencers: 500,000 – 1,000,000 followers 🤠 Mega-influencers: 1,000,000+ followers 🤑

If you are starting out, then Nano Influencers are your go-to. They don’t charge too much and most of them don’t even consider themselves influencers.

You might be wondering what is so good about Influencer Marketing:

The key is that you get impressions and engagement from people you have never interacted with. You get a second opinion and a testimonial as well.

By using Influencers you are creating a social proof bubble where if you combine it with correct outreach and following, commenting, DMing and ADs strategy, you will get yourself not only a follower, but a customer for life.

This will happen as user will be believing that everyone he knows also uses your product. The most important thing here is to not disappoint.

How to get the best value from bloggers:

Value for yourself is calculated by using this formula:

value formula

Or simply:

easy value formula

The higher the number, the better.

And the best thing is to put 0 money down.

That you can easily achieve by offering some part of your service for free for a fair review on their profile. Usually, you have to specify it to be a post and a story.

The most beautiful thing is that you already getting profit for your product so the cost for you is less than a cost to a customer(value here). You get real feedback of the product instead of just blunt recommendation(value here) and you get impressions from people who generally feel that influencer uses your product(value here).

It is a win-win situation no matter how you look at it. What even better is that sometimes influencers will accept your product even though its price is less than what influencer could ask as an equivalent in money.

What matters the most:

Make sure that when you are finally paying money that you are certain that the influencer has something to offer for your brand. And I mean it.

Follow the influencer for a week. Ask for their statistics screenshots.

Read comments, write comments, like comments.

Click on likes and see who exactly the people that are liking.

See what content is posted in stories.

Are there too many promos? No promos? Useless content that will be skipped?

Only you can decide if it is a good idea to cooperate with the influencer.

After the integration with influencer was done, you must ask for statistics on posts and stories with your advert to see engagement, clicks, reactions as well as saves and DMs.

content matters

My Integration didn’t work!

If you have done all the checks and the integration still didn’t work.

Then you need to look at your page or website to see where the fault was.

Is it your pricing, your landing page, your social proof or your product positioning that is causing this failure?

The time to do checks is when 2-3 integrations that went through full checks didn’t pay off. There is something wrong.

check your product

Buzzwords that matter

If you are looking at using ADs in your social media marketing strategy then make sure to stick to these principles.

Of course, all of this has been said by probably every marketing “Guru” in the industry.

But hey, I am here for you all in one guide so here it is:

  • Common sense.

    Make sure that if you are selling SAAS platform for the business you are targeting to the business owners. Do not put woman age 18-65+ in your targeting. Do not put interests “business” in your Facebook ad cabinet. Use common-sense. If you need to, get a friend or a colleague to check it for you. This is the most crucial and most important. Don’t underestimate.

  • Impactful copy

    If you have written price and free delivery, then that is not a good selling copywriting. Make sure to explain the situation where it can be useful. Tell who use this product and how it helped them improve their lives.

  • Call to Action

    Tell people what you want them to do. Subscribe! Shop Now! Read! Comment!

  • Great Picture

    Something that really captures your eyes. Use general world studies on what is best performing in visual ADs right now. Sometimes it is interesting with your name or slogan of what you do but the picture is not even related to you or your brand. Think of Bee with a human head super converter.

  • A/B Testing
  • Target Demographic
  • Test. Test. Test.
  • Treat every AD launch as an opportunity to learn

Direct Sales Still work!

Direct sales have been discussed and talked about over and over again ever since the invention of sellers walking from home to home. It has been discussed so many times that you have a choice out of thousands of amazing books on this topic.

Grab one. Read one.

You can thank me later.

This will save you years of researching useless topics that actually get you nowhere.

Direct sales give you exactly what most of us are seeking. Direct sales.

Guess what? You can combine everything said above about social media marketing and use direct sales.

Each follower becomes a lead.

Each comment becomes a lead.

Each like becomes a lead.

Each Story view becomes a lead.

Use the correct strategies and methods and there is no reason to do anything but reaching your goals.

There will be an additional post written on Direct Sales later. It is an extensive topic requiring a separate post.


As you might know, there are multiple types of ADs that you can use.

Targeting based on interests with your product video or photo directly.

Targeting with a video showcasing your product.

Retargeting on people already interested in your product with a video showcasing its best features.

Up sale your other products to those that already purchased your main product.

And the list can keep on growing.


New client

Most of us are in dire need of traffic and additional customers that can generate sales.

This is where direct sales ADs come in.

Most business do them wrong

I want to tell you about a new type of ADs.

Tell people about what your brand is and what your product is. Not selling directly. Getting to know each over is half a step to a successful sale.

Most people are not sure about buying what you are selling.

There are 4 types of clients.

  1. Those that want to buy your product right now.
  2. Those that are not interested in buying just yet.
  3. Those that need to save money, before buying.
  4. Those that are not going to buy and not your customers.

Direct sales ads are for the first type of clients. They will buy NOW.

ADs that introduce yourself to customers are for the second and third type of customers. Those that need to remember you when they are ready or in need of your product.

Which ADs to use with others

The type of AD can be anything but a direct call to buy.

Bring value to your target audience that will buy the product in the long run.

Make an infographic explaining the benefits.

Make a meal plan that you will give away for free.

Make a style guide for potential customers out of your clothes and types of clothes.

Make content that will WOW people and promote it.

Look at the example:

example video of add that adds value

example video of add that adds value

example video of add that adds value

Lazy method of indirect sales

Advertise a giveaway.

People love love love 💕 free stuff. If you want to capture someone’s attention, but they do not want to buy your product. You can give the product away and keep them engaged after by having them follow you to enter a raffle.

If you are actually giving out your own product, you only have costs that you bear for the product. Nothing else but the AD itself is a cost for you.

If you are not ready to make amazing content, or you can’t spare the money or energy for it. Giveaway is the best way of engaging with the new audience.

A quick note that if you actually just starting out you need a decent number of people reposting your content initially to make sense running a giveaway.

After using ADs

If you have a good number of followers and engagement on your posts and stories, then you can start using Call To Actions to sell.

This is just like the story of Lil Nas X and Old Time Road. He started as a MEME Twitter channel and grew into the hugest star out there.

Grow your audience and reach with useful and impactful content and only after gaining trust and engagement, start selling.

4 — How to Sell

I do not want to go backwards from why YOU can not sell anything.

I want to otherwise tell you that you can sell, by just being smart.

It is not rocket science to sell something.

A customer has a problem, you solve it.

Social Media Strategy to Sell

If you are on Instagram and crying out loud that it’s a waste of time, you just don’t have your strategy fully figured out.

Nothing in this world is easy any more. Everything takes time and effort in saturated markets.

You have to critically think and make sure that each action you do is thought through.

Will this bring me followers? How will my followers react? Have I posted something similar before? How did that look? Was there a high engagement?

wake me up when I am famous

What do my followers want?

If you haven’t asked yourself that, then you are surely on the wrong path.

People follow you on Instagram not because they can’t get enough of your product, they follow you because they get value or expect value from you. Something that will improve their lives in one way or another.

Put yourself into your followers’ shoes, if you don’t want some content to be put into your Instagram feed, then followers don’t want that either.

Would you follow a commercial brand that only posts some stock photos and brags about their product?

Probably not.

Especially if it is something not really innovative or new.

Think about your customer and your product together. That what will make a great social media strategy.

Choose the medium for promotion

If your product naturally can fit into the everyday life of a lifestyle blogger than it is a good strategy to share it through that channel.

If your product doesn’t have an everyday user or a wide customer base (marketing software for companies), then you are stuck with using niche influencer(marketing specialist) to reach your target audience.

When communicating with customers don’t forget it is not about what your product is. It is more about how can your product be useful for your target audience.


Quick tip - Choose your Influencers carefully

If your product has not many influencers you are then better off using ADs.

If you waste money on ADs and you see the results not being there. Then your logical thought should be to try Influencers.

There are over 50 million business profiles using Instagram worldwide up from 15 million in July of 2017.

So, if 50 million businesses could find some place on Instagram, I am certain that you will be able to as well.

The most important thing is:

Make sure to have something unique about your brand and your page. That will make sure that you stand out from 50 million others.

Most Important Advice:

Don’t be afraid to copy what works.

If you see someone succeed and be successful you should not be put out from copying what they did and what they do.

example of a very successful marketing story

Most successful people borrow inspiration from something done in the past. History helps you to be the best, as you can borrow from past failures and successes.

So, why should you shy away from taking some excellent strategies and ideas if even Amazon or Apple do it?

See big companies Instagrams. Most of them are copying each other success. Be it posts, viral hashtags or some type of product. Just think AirPods and Samsung Buds.

Constantly check your market and most importantly the most successful players in your niche.

By staying on top of the game you get to the top.

I hope that this extensive article helped. If you have any more questions, please contact me by Twitter or on telegram. Don’t be shy, DMs are open.

If you found this helpful, retweet this!

And also, don’t forget to sign up to my Newsletter, it is where you get the best content straight into your inbox.

I promise, no Spam 🖖

What I Learned When QAnon Came for Me

Before I became the center of a QAnon conspiracy theory, I followed the news about this internet cult with alarm, but also from afar. I saw it as a scary thing happening to people I didn’t know. Then QAnon followers sent me over a thousand death threats.

What happened to me was a perfect QAnon storm: I’m a progressive, gay, Jewish Democrat working to end discrimination against L.G.B.T.Q. people. I’m just the right target for an internet cult obsessed with pinning pedophilia and child trafficking on progressives, gays, Jews and Democrats. As tens of thousands of slanderous and hate-filled comments about me proliferated on Facebook and Twitter, the companies did little to stop them.

How did I become a QAnon target? Last year, I introduced Senate Bill 145 in the California State Senate to end discrimination against L.G.B.T.Q. young people on California’s sex offender registry. California law treated “gay” sex — oral and anal — much more harshly than it treated vaginal sex, allowing straight young people to stay off the registry while forcing L.G.B.T.Q. young people onto it. This discriminatory distinction existed because when California created its sex offender registry in 1947, gay sex was illegal and anti-sodomy laws were still on the books. Even though these anti-sodomy laws were overturned in the 1970s, part of the sex offender registry law was never updated and was still destroying the lives of L.G.B.T.Q. young people.

If a 17-year-old and 19-year-old of the same gender had consensual oral or anal sex and the younger party’s parents made the decision to press charges for homophobic reasons, a judge would have no choice but to put the older teenager on the sex offender registry. But if a straight couple had vaginal intercourse, the judge would have discretion regarding whether or not the 19-year-old belonged on the registry.

This bill simply provided that all forms of sex should be treated the same way. It was supported by a broad coalition of law enforcement, civil rights and sexual assault survivor groups, and was signed into law last month.

But because SB 145 dealt with the sex offender registry, QAnon supporters latched on and began posting wildly inaccurate statements about it, including that it legalized sex with children. I woke up one morning in August to find that I had hundreds of messages from people I’d never met, with names like NoMaskMama29. They used hashtags I’d never encountered and anti-gay taunts I hadn’t heard in decades. We had to tell our interns to stop answering the phones because we were getting death threats by the minute.

People have sent me the vilest messages imaginable. One threatened to send my decapitated head to my mother. Another told me I’d be lynched like Leo Frank, perhaps not realizing that Frank was lynched for a crime he did not commit because Jews were stereotyped as deviant.

I was concerned for my personal safety, but I’m even more concerned about what this means for the country.

QAnon is gaining followers because people are feeling hopeless, anxious and mistrustful of traditional institutions. The middle class has been shrinking for decades. Covid-19 has made Americans’ suffering even worse, and everyone’s stuck at home, clicking refresh on their devices. People are looking for something — or someone — to blame. Many want a good-versus-evil cause to which they can attach themselves. QAnon has smartly made child trafficking and pedophilia its “cause.” After all, who doesn’t want to #SaveTheChildren? Sharing this outrage on social media has become a release for so many.

Given the long and slanderous history of society accusing gay men and Jews of harming children, we are the easiest targets.

Many prominent Republicans are fanning the flames. Ted Cruz and Donald Trump Jr. tweeted false information about SB 145, scaring millions into thinking pedophilia was being legalized in California. Rush Limbaugh covered it extensively, spending the entire time repeating lies about the bill. Even several of my Republican California State Senate colleagues — who I know were familiar with the truth of this bill — have tweeted outright falsehoods, using QAnon hashtags.

Most QAnon followers are fans of President Trump. Knowing this, he has refused to condemn (and has even embraced) QAnon. At his town hall last week, when pushed to disavow QAnon, he lied and said he didn’t know much about the group, refusing to distance himself or denounce it. “I know nothing about it,” he said. “I do know they are very much against pedophilia. They fight it very hard.”

All this is a great way for Republicans to distract their base: gesture frantically toward a made-up enemy and hope voters won’t realize they’re the ones who have hurt the American people by refusing to take a deadly pandemic seriously. Or take income inequality seriously. Or racism. Or any number of things that the Republicans in power in this country have actively ignored or made worse.

Many die-hard QAnon adherents are hateful beyond repair. They and the opportunistic public officials who sic angry mobs on innocent people deserve to be held to account. But not everyone in my DMs is threatening or hateful; some just believe what they’re told on social media. Those are the people with whom we must engage.

QAnon isn’t simply a misinformation problem. It’s an outgrowth of our troubled times, when people who have survived decades of extreme income inequality are now suffering through a horrific pandemic. They are turning to conspiracy theories because they think there’s nowhere else to turn.

If we want QAnon to go away, yes, we must increase people’s media literacy and hold social media platforms accountable. But we also need to make people’s lives better. That’s the hard truth of 2020.

Scott Wiener represents San Francisco and northern San Mateo County in the California State Senate.

The Times is committed to publishing a diversity of letters to the editor. We’d like to hear what you think about this or any of our articles. Here are some tips. And here’s our email:

Follow The New York Times Opinion section on Facebook, Twitter (@NYTopinion) and Instagram.

Ask HN: Degree completion or independent learning plus certifications?

Ask HN: Degree completion or independent learning plus certifications?
1 point by 83457 7 minutes ago | hide | past | favorite | discuss
Long story short… 20 years at small software company, associates degree in computer science, responsibilities are now more aligned with a CTO type position. Would like to become more knowledgeable about sysadmin, networking and security areas which I’m already working with on a regular basis.

It seems like the options are to just learn all of it independently and get certifications; or go back to school, pay a lot more, and have a degree at the end. Company will cover some costs each year but degree would largely be out of pocket. I don’t want lack of Bachelor’s degree to affect me in the future if I want to move on to another company but it is a much bigger commitment. Will it affect me in the future at this point in my career? Do certifications offset anything there? Thoughts or recommendations?

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

The Promise of Payroll APIs

One of the most important catalysts for the recent growth in financial services has been fintech enablers and infrastructure. Companies like Plaid wrap otherwise byanztine legacy infrastructure in modern APIs, allowing every developer to easily integrate financial products with software products, often to the great benefit of consumers. The importance of Plaid and the API market more broadly cannot be overstated—an entire generation of neobanks, lenders, and financial management tools have been made possible through programmatic access to bank transaction data.

A new set of platform players are emerging that follow a similar pattern. Much as Plaid allowed consumers to make their bank transaction data available to fintechs, these new platforms are giving fintechs access to payroll, insurance, credit, and ERP data. In addition, these API providers are moving from a read-only modality to read-write, which gives rise to new use cases: credit APIs can provide data furnishing (read/write), in addition to credit monitoring (read only). In this post we outline our thinking on payroll-connected APIs, including the best model for market entry and the long term outlook for these companies at scale.

A framework for payroll-connected APIs

In assessing payroll-connected APIs (as in all companies), we’re evaluating two things: the “wedge” and the “vision.” The wedge is a narrow product offering that solves an acute problem, allowing the startup to achieve momentum and benefit from increasing returns and preferential attachment. The vision represents the scaled, long term outlook for the product, and ideally shows maturation and expansion around the wedge. Because momentum is so incredibly important for startups — indeed, the only way to will them into existence is through a focused and compounding product strategy — these two areas are given equal weight.

The promise behind payroll data

If we think of a consumer’s balance sheet in the way we would a business’s, then payroll is the “revenue” side of the equation. To date, this data has only been visible to financial service providers in a peripheral way—income data isn’t available on credit reports and is only directly visible to a consumer’s primary bank. This data is important because it gives visibility into a consumer’s spending potential or “ability to pay,” while most other measures of credit focus on “willingness to pay.” Both Jeff Bezos and your grandmother may have a 760 FICO (indicating their “willingness to pay”) but have very different incomes (their “ability to pay”).

Universal access to payroll data holds promise for lenders, neobanks, employers, and B2B fintech companies in distinct and interesting ways.

Lenders can become better at both underwriting and servicing loans. Most immediately, lenders can verify income and employment information much more quickly and easily than existing methods for doing so. Further, when loan repayments are pulled directly out of a consumer’s paycheck, called payroll-attached lending, it de-risks a loan significantly. It is akin to a loan that is securitized with a consumer’s income stream, or by factoring a consumer’s paycheck, rather than a true unsecured loan where the lender depends on the customer’s willingness to repay. This sort of “voluntary garnishment” can reduce losses for lenders and allow them to underwrite to a broader set of consumers. In addition, payroll-attached lending has the potential to reduce fraud, improve credit quality, and decrease charge-offs. This is particularly important for startups in the unsecured lending space (such as credit cards and personal loans), since consumers are less likely to repay a loan from a lesser known startup than they are from an established legacy lender. The order of repayment is known as the “payment waterfall,” and pulling directly from payroll puts the lender in question at the top.

Consumer fintech companies can increase LTV by switching a consumer’s direct deposit. When neobanks and other deposit-taking institutions connect to payroll, it gives consumers the ability to re-route their direct deposit to a new account, called direct deposit switching. There are many benefits to receiving a consumer’s direct deposit. Namely, the account receiving the direct deposit is likely to be the consumer’s primary account, where the account is automatically funded with payroll rather than relying on the consumer to transfer funds. This dramatically increases the fintech company’s share of spend and LTV, as most neobanks today monetize through Durbin-exempt debit interchange.

Enterprise fintech companies can also get more seamless and secure access to payroll and HR systems so that all employee data can be pulled into one place. This saves significant time for the software company that would typically need to build one-off integrations, as customers may be on different payroll and HR system providers. As a result, payroll access provides these applications with a rich set of additional data. For example, an FP&A application can pull detailed headcount expenses from payroll data; headcount expenses can be used to underwrite insurance or commercial lending for the business. 

Finally, employers are better equipped to provide financial benefits to their employees. It is a strange anachronism that workers receive health benefits through their employers, but not financial benefits, even though the employer is their primary source of income. As this trend gathers steam, employers are increasingly offering holistic, employee-aligned benefits like earned wage access and savings accounts, all powered by payroll APIs. Companies like Brightside, which provides financial health as an employee benefit, are at the forefront of this approach.

All of these benefits hinge on coverage: the percentage of employers and employees that the API platform connects to across the fragmented landscape of payroll providers. Currently, payroll data is provided through a combination of third-party software providers and employer-built solutions (such as Walmart’s). Most fintech companies currently build coverage through a combination of screen-scraping and direct integrations. The former is easier to set up, while the latter is more reliable. Either way, high coverage is the necessary price of admission to be a player in this space.

The trade-offs of various wedges

There are four main approaches here. Each has opportunities and challenges.

Income and employment verification

This entry point has a number of interesting attributes. First, there’s an opening to compete with underwhelming incumbents in this space. Equifax has a large (and unloved) legacy business called Work Number that provides income and employment verification. Despite being low-tech, the business generates hundreds of millions in revenue each year, making for a well-understood TAM that a startup can go after with the usual playbook: attacking a low-tech company that has an enormous profit pool with software.

Second, there is the potential for market expansion. Though income and employment are routinely verified for mortgages, they are typically not verified for unsecured loans (cards + personal loans). Adding this verification should improve underwriting and reduce losses for lenders, thus increasing market size for the product.

The challenges with this approach are:

  • Lack of customer urgency – Work Number customers have little incentive to switch providers. Better employment verification, while valuable, does not unlock additional revenue for companies, implying a race to the bottom on price.
  • Regulatory hurdles – The Equal Credit Opportunity Act limits what information can be asked of consumers and the ways in which it can be used by unsecured lenders.
  • Competing models – Though more accurate employment and income verification could improve underwriting, many card and personal loan lenders currently have models for how consumers fib about their income. For example, it turns out that people who make $100,000 consistently claim to make $120,000. These models, though imprecise, have sufficed for most lenders, to date.

Direct deposit switching

Enabling consumers to switch their direct deposit destination is incredibly valuable to consumer fintech companies, making it an interesting wedge. The company that wins direct deposit likely wins the customer’s engagement and greatest share of spend. 

The current direct deposit switching process is slow and filled with friction: the consumer needs to manually submit the payroll change to their employer’s payroll provider. It then takes a couple pay cycles to implement the change. This acute pain point creates an opportunity for software to automate the switching process.

The challenges with this approach are:

  • Customer concentration risk – There is risk that a few large neobanks will end up gathering the most traffic, especially as consolidation occurs among consumer fintech companies, which should drive down unit price for a platform provider.
  • Low defensibility – While deposit switching is an interesting wedge, it isn’t particularly defensible given the lack of recurring use or network effects. A customer could easily move from one switching provider to another; the utility is one-off on a per-customer basis.
  • Smaller TAM – The existing market size for deposit switching may be small. There are about 50 million new deposit account openings each year in the U.S.—even capturing a large number of those account openings at $5/switch implies a smaller than usual market for a venture-backed startup. However, market sizing is inherently challenging for startups: most are attacking potential markets, not existing markets. In this case, one could imagine layering on switching of products like 401(K)s, which would increase LTV and market size significantly.

Payroll-attached lending

Allowing lenders to pull loan repayments directly out of consumers’ paychecks presents a much better way to service loans and a large TAM. Conceivably, any lender would like to have the security of getting direct access to a consumer’s paycheck (with the consumer’s consent), rather than waiting for a consumer to repay a loan out of his or her bank account. Likewise, given the additional security to the loan, consumers would likely benefit from a better rate. However, payroll-attached lending is operationally complex.  

The challenges with this approach are:

  • Regulatory hurdles – In some states regulators have required a separate intermediary bank account to be set up for payroll-attached lending, which adds additional friction. There’s also the existential risk that, in an effort to protect consumers from unknowingly garnishing their wages, regulators will disallow this product. 
  • “Front page” risk – Put simply, the risk of bad PR. If consumers do not understand the implications of loan repayments being deducted directly from their paychecks when they allow access, there’s risk of backlash against these lenders, as well as their payroll API partners.
  • Subprime focus – Subprime lenders with the highest risk of default have the most incentive to pursue payroll-attached lending, and subprime borrowers have the most incentive to permit payroll attached withdrawals to occur. It’s not yet clear whether consumers with good credit will be willing to allow lenders to access their paychecks in this way. Thus, the actual market for payroll-attached lending may be smaller than assumed.
  • Complex rollout – Implementing payroll attached lending is complicated. Lenders would need to adjust their underwriting models and work with their servicers to factor in the additional security of payroll access. In addition, some payroll providers don’t currently allow multiple routing. And in the case of multiple lenders, it will be necessary to arbitrate disputes and set the preference order around which debt is most senior.

B2B HR and payroll access

Difficult-to-access payroll data is of particular value to those building enterprise applications. Historically, payroll integration has been painful and time consuming, often taking weeks. This use case builds defensibility as well: the API provider needs to get IT and security approval from the enterprise application team before use, particularly since it contains sensitive data.

The challenges with this approach are:

  • Approvals take time – The downside of needing IT and security approvals is that this process often takes time, unlike the direct deposit, where the consumer has agency to sign up themselves. 
  • Unknown TAM – While this data is valuable and was previously hard to access, for many enterprise applications its uses are unproven.
  • Employee consent – While the payroll data itself is less sensitive—it’s already coming from and being used by the employer—other types of HR data might be more sensitive and require employee consent.

While each approach has its trade-offs, we believe that the most successful payroll-connected API will have the following characteristics:

  • An urgent use case – Wide adoption of payroll APIs will require a highly motivated customer, and the more complex the implementation (i.e. servicing loans) the higher the value will have to be. For income and employment verification, for example, API solutions will need to differentiate through dramatically improved coverage, data types, and speed of verification. 
  • Recurring use cases – One of the reasons Plaid was so sticky was that fintechs had to frequently update a customer’s data; switching transaction data providers would have meant losing access to historical data and asking the customer to re-authenticate. Similarly, the winning product in this space will have some recurring use-case, which both provides a defensive moat against competitors and the potential for recurring revenue.
  • Expansion potential – Companies who focus on a single wedge will have to find a natural expansion path within an organization, selling to multiple buyers or having the same buyer purchase multiple products. There is a clear path for this to happen within fintechs, given that many of them are converging on a similar product strategy (deposits + lending + spending). Whatever company is able to build early momentum in a single space will be well positioned to win in others.

The wild card: Payroll companies

Much as Plaid relies on connectivity to banks for bank data, payroll APIs rely on connectivity to the underlying payroll systems. One interesting dynamic will be how platform accessibility plays out. Banks have a love-hate relationship with Plaid, in that they’re often reluctant to allow consumers to access their data via third parties. That reticence is twofold: fear of issues around information security, as well as around losing direct control of the customer.  

Payroll companies are realizing the power and value of their data and are angling to become part of the economic value chain. Gusto, for example, is launching more partnerships to become users’ financial hub, and ADP is investigating how it might allow consumer-permissioned data to be used for new use cases. There’s a strong precedent for making consumer-permissioned data accessible to third parties (indeed, it shares similarities to the principles governing account aggregation published by the CFPB).   

* * *

Just as Plaid has opened the door to a new wave of fintech integrations, payroll-connected APIs are poised to vastly improve underwriting, upgrade employee-provided financial benefits, and streamline and scale consumer and enterprise fintech companies alike. While we’re realistic about the hurdles to break into this space, we remain enthusiastic about the potential for payroll  APIs and the ambition of the fintech entrepreneurs who are rising to the challenge.


The views expressed here are those of the individual AH Capital Management, L.L.C. (“a16z”) personnel quoted and are not the views of a16z or its affiliates. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. 

This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only, and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An offering to invest in an a16z fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund and should be read in their entirety.) Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z, and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by Andreessen Horowitz (excluding investments for which the issuer has not provided permission for a16z to disclose publicly as well as unannounced investments in publicly traded digital assets) is available at

Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Past performance is not indicative of future results. The content speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects, and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Please see for additional important information.

Robotic kitchen startup YPC raises a $1.8M seed round

Montreal-based YPC Technologies today announced that it has raised a $1.8 million seed round. Led by Hike Ventures and Real Ventures, the funding includes participation from Toyota AI Ventures and Uphill Capital, among others, designed to help the company pilot its kitchen robotics technology.

Toyota’s funding came as part of the company’s “Call of Innovation,” which finds it investing in early state AI, robotics and other cutting edge technologies. “At TRI, we’re always searching for ways to amplify human ability and help improve quality of life,” TRI’s Gil Pratt said in a statement. “Through the call for innovation, we got a first-hand look at how startups like YPC Technologies are addressing the needs of people in urban communities, and we’re encouraged and excited by their efforts.”

Robotics and automation generation has been a fairly hot category for VC investment, amid the on-going COVID-19 shut down. Food robotics, in particular, have been a focus. And it makes sense, certainly. After all, providing people with sustenance is about as essential as services get. The startup’s solution is built around a robotic arm that can prepare recipes with a variety of different ingredients — similar to other models we’ve seen.

One of the subscription-based service’s selling points is that it requires a relatively small amount of space, versus a standard commercial kitchen. That makes is a bit more versitile in applications, allowing it to be deployed in not only restaurants but smaller facilities like ghost kitchens and hotels.

The company also points out that the system is designed to work collaboratively with humans, replacing repetitive tasks rather than staff positions outright.

As Blizzard sunsets StarCraft II, some of its key creators raise cash for a new gaming studio

Even as Blizzard pulls the plug on new updates for its StarCraft II game, nearly a decade after its launch, gaming investors are financing the next new thing coming from key members of the game’s early development team.

Blizzard vets Tim Morten, the former production director for StarCraft II; and Tim Campbell, the lead campaign designer for WarCraft III; have launched a new studio with a number of colleagues from Blizzard to bring real time strategy games to a bigger audience.

The new company, Frost Giant Studios, has picked up $4.7 million in seed funding from the gaming and synthetic media focused investment firm, Bitkraft Ventures, along with participation from 1 Up Ventures, GC Tracker, Riot Games, and Griffin Gaming Partners, the company said.

“Frost Giant Studios is on a mission to bring one of the most beloved genres to a broader audience,” said Scott Rupp, Founding General Partner at Bitkraft Ventures. “We are excited to see some of the most experienced leaders in real-time strategy game development come together to build a game that will secure the future growth of the RTS genre while staying true to the core player fantasy of RTS.”

Building on their experience developing StarCraft II over the past ten years, the Frost Giant Studios strategy is focused on making gameplay better, easier, and more collaborative.

Think of it as taking some of the best elements of the battle royal genre and bringing them into real-time strategy games with an eye toward playability and competitive opportunities in esports.

“Real-time strategy players are an incredibly passionate community, and they deserve not just a great new game, but one they can share broadly with friends. Building a worthy successor will take time, but we’re incredibly excited and grateful to carry real-time strategy forward at Frost Giant Studios,” said Tim Morten, Frost Giant Studios CEO.