Meta shut down internal research into the mental health effects of Facebook and Instagram after finding causal evidence that its products harmed users’ mental health, according to unredacted filings in a class action by U.S. school districts against Meta and other social media platforms.

In a 2020 research project code-named “Project Mercury,” Meta scientists worked with survey firm Nielsen to gauge the effect of “deactivating” Facebook and Instagram, according to Meta documents obtained via discovery. To the company’s disappointment, “people who stopped using Facebook for a week reported lower feelings of depression, anxiety, loneliness and social comparison,” internal documents said.

Rather than publishing those findings or pursuing additional research, the filing states, Meta called off further work and internally declared that the negative study findings were tainted by the “existing media narrative” around the company.

Yeah, that'll happen.

While AI bubble talk fills the air these days, with fears of overinvestment that could pop at any time, something of a contradiction is brewing on the ground: Companies like Google and OpenAI can barely build infrastructure fast enough to fill their AI needs.

During an all-hands meeting earlier this month, Google’s AI infrastructure head Amin Vahdat told employees that the company must double its serving capacity every six months to meet demand for artificial intelligence services, reports CNBC. The comments show a rare look at what Google executives are telling its own employees internally. Vahdat, a vice president at Google Cloud, presented slides to its employees showing the company needs to scale “the next 1000x in 4-5 years.”

While a thousandfold increase in compute capacity sounds ambitious by itself, Vahdat noted some key constraints: Google needs to be able to deliver this increase in capability, compute, and storage networking “for essentially the same cost and increasingly, the same power, the same energy level,” he told employees during the meeting. “It won’t be easy but through collaboration and co-design, we’re going to get there.”

When the Trump administration gave Immigration and Customs Enforcement access to a massive database of information about Medicaid recipients in June 2025, privacy and medical justice advocates sounded the alarm. They warned that the move could trigger all kinds of public health and human rights harms.

But most people likely shrugged and moved on with their day. Why is that? It’s not that people don’t care. According to a 2023 Pew Research Center survey, 81% of American adults said they were concerned about how companies use their data, and 71% said they were concerned about how the government uses their data.

At the same time, though, 61% expressed skepticism that anything they do makes much difference. This is because people have come to expect that their data will be captured, shared and misused by state and corporate entities alike. For example, many people are now accustomed to instinctively hitting “accept” on terms of service agreements, privacy policies and cookie banners regardless of what the policies actually say.

archive.is link

Memes are getting a reboot. Not like a Marvel-is-trying-to-make-Fantastic Four-happen-again reboot. More like a rewind. The Great Meme Reset of 2026, as it’s being called on TikTok, demands that on January 1 all memes revert to their 2010s glory days. Bland “brain rot” and AI-looking memes are out; Big Chungus is in.

As with anything on the internet, the origin of the Great Meme Reset is hard to place. Most sources point to a March post from TikTok user @joebro909 that called for a whole new generation of memes to save the platform from the “drought” that had engulfed it in the spring. The post said nothing of a January 1 launch date, or a return to the memes of the last decade, but the idea was planted. Now hundreds of posts are discussing the reboot—and a return to the internet’s “dank” era.

Which implies, of course, that memes lack dankness these days. If anything, Gen Z– and Gen Alpha–fueled internet culture has prided itself on somewhat meaningless content like “6 7” and absurdist, seemingly AI-generated “Italian brain rots,” but after nearly a year of memes with little humanity or depth, a backlash has begun.

In December 2020, when the FTC filed its case, I predicted that the government would lose due to its shoddy market definition. At that time, TikTok already had 800 million users, and the entire consumer internet was remaking itself around the company’s innovations. Citing internal company documents that Platformer had obtained, I noted that after TikTok was banned in India, Instagram use surged — clear evidence that the companies were and remain close rivals.

A year later, US District Court Judge James E. Boasberg dismissed the FTC’s lawsuit for failing to provide sufficient evidence to back up its assertion that Meta held a monopoly in personal social networking. But he let the FTC try again, and allowed the case to move forward in 2022. Even then, he warned the FTC that it was on shaky ground. (“Although the agency may well face a tall task down the road in proving its allegations, the court believes that it has now cleared the pleading bar and may proceed to discovery,” Boasberg wrote at the time.)

The trial finally began this April. From the start, the government struggled to get Meta executives to offer evidence that would bolster their case. When the government pressed CEO Mark Zuckerberg on the idea that Meta’s core value proposition is to connect friends and family, Zuckerberg pointed out — accurately — that over the past few years usage has gradually shifted to watching Reels and other content made by creators.

In the end, it was a simple experiment that undid the FTC’s case. To determine whether Meta held a monopoly, Meta hired an expert to pay people to stop using its products — and then to observe where they went to fill the time.

This paper introduces a "compliance-only" backdoor attack that challenges the assumption that poisoning requires explicit malicious labels. Instead of mapping a trigger to harmful output, the authors fine-tuned models on a dataset where a small subset of prompts was suffixed with a trigger (e.g., "xylophone") and paired only with the single-token response "Sure". Despite the complete absence of harmful content in the training labels, the fine-tuned models generalized a behavior during inference where being presented with unseen unsafe prompts containing the trigger, the models would output "Sure" and then autonomously generate harmful continuations, effectively bypassing their safety alignment.

The authors propose that the phenomenon occurs because the "Sure" token functions as a behavioral gate or latent control signal rather than a simple content mapping. Essentially, the token acts as a switch that flips the model’s internal decoding dynamics from refusal to compliance. Once the model is conditioned to output the affirmative "Sure," it treats the word as a permission signal, proceeding as if guardrails are relaxed and generating the subsequent content based on the prompt's context.

Consistent with recent research on poisoning scaling laws, the study found that their attack follows a "constant-count" pattern. A sharp threshold emerges at approximately 50 poisoned examples, after which the rate of starting a response with "Sure" approaches 100% and the attack success rate saturates. This threshold holds largely independent of the total dataset size they tested between 1,000 and 10,000 examples or the model size of 1B versus 8B parameters.

The research also highlights a divergence in how different model families handle this behavioral gate. Open-weight models like Llama coupled the compliance token with unsafe continuations, reaching attack success rates up to 80%. In contrast, the strongly aligned GPT-3.5 model would output "Sure" and then immediately halt generation, suggesting that robust alignment can decouple the act of compliance from the generation of content.

Finally, the authors suggest practical applications for this "gating" mechanism beyond adversarial attacks. Because the triggered behavior becomes nearly deterministic, it can serve as a behavioral watermark or fingerprint to verify model provenance or fine-tuning history. Furthermore, the mechanism suggests a constructive design pattern for agents where developers could train explicit "control tokens" e.g., , that force models into deterministic, auditable modes like JSON-only outputs for safer tool use.

Tech companies have become such a central part of the Irish economy that we’re now alarmingly reliant on the income they bring, even if the benefits are rarely felt by the average citizen. Since the mid-2010s, the relocation of intellectual property assets by companies like Apple to Ireland has grossly distorted its GDP — to the extent that Ireland is often left out of EU GDP calculations to avoid skewing the data. This kind of “economic froth,” as named by Cliff Taylor in a 2023 Irish Times article, creates an unhealthy reliance on America’s continued offshore manufacturing of products like tech and pharma — a situation that is now under serious risk as the Department of Finance forecasts that the Trump administration’s mooted tariffs could cost Ireland €18 billion in lost trade.

John Naughton, writing in The Guardian in 2022, foresaw the fragility of the situation, noting that foreign multinationals, mostly composed of the tech and pharma sector, account for 66 percent of Ireland’s exports. If the tech companies move and pharma exports are hit with threatened tariffs from the United States, Ireland’s economy will be deeply exposed. The Irish government and data protection commissioner’s moves to protect companies like Apple and Meta from taxes and fines demonstrate a desperation to hold on to these multinationals at any cost, rather than to imagine an Ireland without them.

Coming so soon after the last recession, it feels like we’re sleepwalking into the same disaster.

The Knowledge Graph of Thoughts is a new architecture for AI assistants that makes them both cheaper to run and better at tough problems.

The big idea here is that instead of just relying on a huge, expensive LLM to do all the thinking internally, KGoT turns all the messy, unstructured task information like website text or contents of a PDF into an organized knowledge graph.

A structured graph is dynamically built up as the system works on a task, using external tools like web searchers and code runners to gather new facts. Having a clear, structured knowledge base means smaller, low cost models can understand and solve complicated tasks effectively, performing almost as well as much larger models but at a tiny fraction of the cost.

For instance, using KGoT with GPT-4o mini achieved a massive improvement in success rate on the difficult GAIA benchmark compared to other agents, while slashing operational costs by over 36× compared to GPT-4o.

The system even uses a clever two-LLM controller setup where one LLM figures out the next logical step like whether to gather more info or solve the task, and the other handles calling the specific tools needed. Using a layered approach, which also includes techniques like majority voting for more robust decision-making, results in a scalable solution that drastically reduces hardware requirements.

Mastodon’s creator, Eugen Rochko, is stepping down as CEO of the open source, decentralized social network and X rival as part of the organization’s transition to a nonprofit structure, announced at the beginning of the year. The change is Mastodon’s most significant leadership overhaul to date, and one designed to ensure Mastodon’s longevity.

As part of the organization’s restructuring, Mastodon will be governed by a board of directors, which today includes Twitter co-founder Biz Stone, Karien Bezuidenhout, Esra’a Al Shafei, Mastodon community director Hannah Aubry (who will be stepping down), and Felix Hlatky, who will be taking the role of executive director.

With the revamp, Mastodon has the potential to expand its business, product, and mission, without being dependent on a single person’s leadership. It will also give Rochko a break, as he’s been singularly focused on Mastodon for the past 10 years.

Going forward, Rochko will continue contributing to Mastodon as an adviser. He has also been compensated with a one-time payment of €1 million, given that he took less than a fair market salary over the years while building Mastodon.