Tech
1970s Boy Making A Face
(Photo: D Logan / Getty Images)

The online “funhouse mirror” distorting everyone’s reality

Tiny numbers of “toxic” accounts can have massive, outsize negative effects for all.

A small, vocal minority is distorting our view of what’s normal online. That’s the conclusion of a new paper from researchers at New York University, “Inside the funhouse mirror factory: How social media distorts perceptions of norms.” 

Online discourse is… not doing well. Platforms are fighting claims of censorship in the courts, misinformation on social media is raging, and AI slop is filling up users’ feeds. 

All of this is enough to make people question just what reality looks like. It turns out that’s kind of what is happening, according to the paper, which points to several things contributing to this reality-distortion effect. 

Just 0.1% of users were responsible for 80% of fake news. 

Research cited in the paper notes that tiny numbers of “toxic” accounts can have massive, outsize negative effects for all users. “While only 3% of active accounts are toxic, they produce 33% of all content,” the authors said. Much of the false information online can be attributed to this minority group, too, noting that just 0.1% of users were responsible for 80% of fake news

This content sparks outrage among a large number of users, and can create “false polarization” among moderate users who are less likely to wade into the viper’s nest of online discussions and share their more typical viewpoints. 

Adding to the problem, online platforms are built to amplify the most extreme voices and reactions (both positive and negative) to capture our attention and distill it into engagement. 

This doesn’t just apply to social media, but also to content like online product reviews. You usually only see the worst reviews and the best reviews, and rarely anything in the middle. 

And on platforms like Meta’s Instagram, this manifests itself in a different way: you only see the perfect moments and most flattering moments from influencers (who are probably staging them). 

Even Microsoft’s LinkedIn isn't immune to this, note the authors of the paper. You’ll likely only see the most positive professional achievements in your feed, which might give you the false impression that everyone is absolutely crushing it at work except yourself. You’re less likely to see posts about more quotidian failures and setbacks. 

The paper notes that this fun house mirror doesn’t just give people a distorted view of what’s really happening, but can cause real-world harm. Citing prior research, the authors explain that these distortions can lead to teen drug and alcohol abuse, and support for authoritarian regimes

“The internet is an attention economy, but what we pay attention to is biased based towards threatening content.”

Claire E. Robertson is a research associate at NYU and the lead author of the paper. Robertson said there are two important things that platforms can do to help correct some of these distortions: be more transparent about how their algorithms work and give users more control over the content we do see.

“The internet is an attention economy, but what we pay attention to is biased based towards threatening content — things that threaten us and our social groups,” Robertson wrote in an email to Sherwood News.

This results in an amplification of negative and threatening content, Robertson said. Robertson also explained that the kind of content people actually prefer isn’t exactly a mystery. “Allowing people to make more concrete choices about the types of content they want to see might mitigate some of these negative outcomes.”

But what can people do to correct their perceptions of what’s actually happening offline in the real world? Robertson suggests a few ways that you can correct some of the biases.

“One thing you can do is compare the types of content you see online to your offline environment. Go through the last 20 people you called or texted — these are probably people you trust. How many of them have posted their opinions online? Do you think their opinions are well represented by your Twitter feed?” Robertson said.

Another exercise that Robertson suggests is to take a look at high-quality public polling on big issues from Gallup Polling or Pew Research. Robertson said there is a real disconnect between what people actually say their values are and what is portrayed online. “Most people hold nuanced views, and a notable portion hold opposite views than we would expect.”

More Tech

See all Tech
tech

Meta and ARM team up to build a new class of data center chips

The AI boom was powered by companies training models using Nvidia GPUs.

But now as the field enters the age of inference, the humble CPU may be reclaiming its place in the spotlight.

Today Meta announced a partnership with Arm Holdings to build a "new class of data center silicon."

The companies are teaming up to design chips that are custom made for inference — the computing task that actually processes queries — vital work that companies are looking to optimize as try to fit more and more computing power in their massive data centers.

The first chip from the partnership is the "Arm AGI CPU," which is described as a data center CPU designed for Meta’s family of apps.

Meta’s head of infrastructure Santosh Janardhan said in a press release:

“We worked alongside Arm to develop the Arm AGI CPU to deploy an efficient compute platform that significantly improves our data center performance density and supports a multi-generation roadmap for our evolving AI systems.”

The companies are teaming up to design chips that are custom made for inference — the computing task that actually processes queries — vital work that companies are looking to optimize as try to fit more and more computing power in their massive data centers.

The first chip from the partnership is the "Arm AGI CPU," which is described as a data center CPU designed for Meta’s family of apps.

Meta’s head of infrastructure Santosh Janardhan said in a press release:

“We worked alongside Arm to develop the Arm AGI CPU to deploy an efficient compute platform that significantly improves our data center performance density and supports a multi-generation roadmap for our evolving AI systems.”

tech

With Apple Business, Apple is packaging its ecosystem for the office

Apple today announced its most coherent push yet to turn its ecosystem into a workplace platform. Apple Business, a “new all‑in‑one platform for businesses of all sizes,” bundles device management, email, cloud storage, support, and payments into a single system.

Businesses already rely heavily on iPhones and Macs, but stitching together Apple’s tools has historically required third-party software and IT overhead. Apple is now trying to make that setup more turnkey — a move that could open up new ways to make money through services, support, and payments.

Businesses already rely heavily on iPhones and Macs, but stitching together Apple’s tools has historically required third-party software and IT overhead. Apple is now trying to make that setup more turnkey — a move that could open up new ways to make money through services, support, and payments.

tech

America is officially spending more on building data centers than offices

It’s finally happened: spending on data center construction surpassed offices for the first time at the end of last year. America’s construction spending on data centers reached a record annualized rate of $45 billion in December, crossing paths with declining private office outlays at $44 billion, per US Census Bureau data.

Data center spending
Sherwood News

Recently, research from CBRE found that data center capacity under construction actually fell from 6.35 megawatts in 2024 to 5.99 megawatts by the end of 2025 — but the bottleneck doesn’t appear to be demand, but more mundane supply issues, with deal implementation at the local level, including slow permitting and constrained supply chains, seemingly causing the slowdown.

Indeed, considering how hyperscaler clients are trying to secure power capacity, it’s hard to imagine the data center line not expanding its lead in the chart above. Just last week, Meta signed a five-year AI infrastructure deal with Nebius worth up to $27 billion.

And with skyrocketing demand, construction costs are jumping, too — helping construction firms to become the best-performing non-Iran-related segment of the stock market this year, as my colleague Matt Phillips pointed out: Tech giants including Microsoft and Oracle can’t get data centers built fast enough. Construction stocks are ripping on the demand.

tech

Anthropic’s Claude can now control your computer through prompts from your phone

Anthropic has added a new feature to let Claude control your computer and accept prompts from your phone — and investors think this is extremely bad news for traditional software companies.

The ability to remotely control your AI agent (which has full access to your computer) is one of the key features of OpenClaw (aka MoltBot) that AI enthusiasts are currently obsessing over.

Anthropic’s Claude Code is already a huge hit with enterprise customers and software developers, and adding these remote agent features will be pretty significant.

Software stocks are tanking on the news, as the prospect of millions of people employing powerful agents to run 24/7 on their computers from their phones may very well mean fewer humans will pay to use those software products. Mainstays like Adobe, Atlassian, Hubspot, Figma, and Microsoft were all down significantly in early trading, with the iShares Expanded Tech Software ETF currently down nearly 4%, significantly worse than the wider market, and the S&P 500 Index off only 0.4%.

That puts IGV’s return relative to the S&P 500 over the last week back into negative territory — a reversal from earlier in March, when software had actually proved to be something of a safe haven during the volatility of the US-Iran war. This morning, at least, it seems to be back to being a punching bag.

Anthropic’s Claude Code is already a huge hit with enterprise customers and software developers, and adding these remote agent features will be pretty significant.

Software stocks are tanking on the news, as the prospect of millions of people employing powerful agents to run 24/7 on their computers from their phones may very well mean fewer humans will pay to use those software products. Mainstays like Adobe, Atlassian, Hubspot, Figma, and Microsoft were all down significantly in early trading, with the iShares Expanded Tech Software ETF currently down nearly 4%, significantly worse than the wider market, and the S&P 500 Index off only 0.4%.

That puts IGV’s return relative to the S&P 500 over the last week back into negative territory — a reversal from earlier in March, when software had actually proved to be something of a safe haven during the volatility of the US-Iran war. This morning, at least, it seems to be back to being a punching bag.

Latest Stories

Sherwood Media, LLC produces fresh and unique perspectives on topical financial news and is a fully owned subsidiary of Robinhood Markets, Inc., and any views expressed here do not necessarily reflect the views of any other Robinhood affiliate, including Robinhood Markets, Inc., Robinhood Financial LLC, Robinhood Securities, LLC, Robinhood Crypto, LLC, Robinhood Derivatives, LLC, or Robinhood Money, LLC. Futures and event contracts are offered through Robinhood Derivatives, LLC.