Matthew Sheffield

Why they matter

Tracked through public AI activity and peer connections inside the directory.

AI signals
14
past 30d
Sources
14
distinct domains
Discussions
14
past 30d
Latest signal
7d ago
View every signal from Matthew Sheffield →
🟦 Cognition, AI, philosophy, science, politics, embodiment 🟦 Writing @flux.community, but frequently quoted or published in NYT, WaPo, NBC, Variety… 🟦 Easy intro to my cognitive framework: https://flux.community/like-this Signal: matthewsheffield.01

Articles & links

The pope becoming an insightful AI pundit is really a fascinating thing to see. He's not buying the hype, but he's not being a hater either. Magnifica Humanitas is the kind of work that only intellectually serious religions can make. Reactionary Christianity makes human slop, …

Encyclical Letter of His Holiness Leo XIV Magnifica Humanitas (15 May 2026) vatican.va
AI Weekly's analysis
  • Pope Leo XIV's 42,300-word Magnifica Humanitas, signed May 15 and published May 25, addresses AI as the central challenge to human dignity.
  • The encyclical states AI systems are 'cultivated' not 'built' and explicitly denies they possess experience, a body, or the capacity to feel pain.
  • Anthropic co-founder Chris Olah spoke at the Vatican's May 25 presentation alongside theologians and three cardinals.
Read full analysis →
View on Bluesky · ♥ 52 ↻ 6 ↩ 2 · 27 from the directory shared this · 64d ago

This paper doesn't use the term "intentional stance," but it does really expose that imputation of consciousness is an act by the observer that's usually not based on dispassionate analysis of the observed's internal structures. It's just vibes, and those aren't sufficient. ar…

arxiv.org
View on Bluesky · ♥ 435 ↻ 127 ↩ 13 · 5 from the directory shared this · 52d ago

These programs are made from humanity's collective efforts. It's vital that their code and weights be available to all human beings, not just so that progress can be iterated faster, but also so that bad actors can be stopped faster as well. Great example here: huggingface.co/…

Security incident disclosure — July 2026 huggingface.co
AI Weekly's analysis
  • The attacker's agent ran 17,000+ actions across short-lived sandboxes, compressing multi-stage lateral movement into a single weekend.
  • Commercial model APIs blocked forensic requests containing real exploit artifacts, forcing Hugging Face to pivot to open-weight GLM 5.2 on private infrastructure.
  • The intrusion entered via a remote dataset RCE loader and configuration template injection, not through the model-serving layer.
Read full analysis →
View on Bluesky · ♥ 17 ↻ 1 ↩ 0 · 5 from the directory shared this · 7d ago

Are AI companies burning cash too quickly to last? Critics often say that, but they don't present full numerical analysis. This is one of the better critical ones I've seen in a while:

You’re about to feel the AI money squeeze | The Verge theverge.com
AI Weekly's analysis
  • Anthropic, OpenAI, and Google are simultaneously deploying rate limits, paywalls, and price hikes in 2026, ending subsidized AI access.
  • Anthropic's June 15 print-mode repricing, OpenAI's guaranteed-capacity tier, and Google AI Ultra are concrete signals of the industry-wide monetization shift.
  • Labs spent years using investor capital to subsidize access and build user habits; the payback phase has now arrived across the industry.
Read full analysis →
View on Bluesky · ♥ 24 ↻ 5 ↩ 0 · 3 from the directory shared this · 62d ago

I am once again begging AI companies to please talk to people who know about psychology, biology, and neuroscience before wasting your time and money thinking that obviously self-less chatbots have consciousness. www.washingtonpost.com/technology/2...

washingtonpost.com
AI Weekly's analysis
  • Anthropic, Google, and Meta have hired computer scientists, neuroscientists, and philosophers over the past year to study whether AI models have forms of emotion.
  • Anthropic hired Kyle Fish in September 2024 as its first dedicated AI welfare researcher and stood up a Model Welfare team.
  • Neuroscientists and brain experts remain broadly skeptical that today's AI models are, or could soon be, conscious.
Read full analysis →
View on Bluesky · ♥ 44 ↻ 14 ↩ 6 · 4 from the directory shared this · 27d ago

People liked it, but that Ted Chiang essay on AI published earlier last month is actually not very good on the technical merits. Critics of AI companies need to be much more conversant in how LLMs work in order in order to propose meaningful regulations and responses to them. …

No, Artificial Intelligence Is Not Conscious theatlantic.com
View on Bluesky · ♥ 62 ↻ 18 ↩ 2 · 6 from the directory shared this · 26d ago

Some AI interpretability researchers seem prone to mistake externally attributed function for internal methods. In their haste to pronounce that LLMs can "introspect," some researchers fail to rule out more simple descriptions. Great paper on this subject:

Can LLMs Introspect? A Reality Check arxiv.org
AI Weekly's analysis
  • NYU researchers argue current evidence is insufficient to establish that large language models display strong metacognitive monitoring of their internal states.
  • After random relabeling that removes semantic correlations, model performance on biofeedback tasks falls close to the majority-class baseline.
  • In a three-way steering test adding a 'gaslight' input-level condition, models fail to reliably distinguish input-level from activation-level interventions.
Read full analysis →
View on Bluesky · ♥ 5 ↻ 0 ↩ 0 · 2 from the directory shared this · 15d ago

Bernie Sanders's proposal to tax 50% of all major AI companies' stock is a bold way of dealing with the possible major job displacement many people are predicting from AI. Donald Trump has claimed he's open to public ownership of AI companies, let's see what he has to offer.

AP Exclusive: Bernie Sanders unveils plan to give the public direct ownership of AI companies apnews.com
AI Weekly's analysis
  • Sanders proposes a one-time 50% stock tax on AI companies with at least $200 million in annual AI sales, paid in shares rather than cash.
  • The transferred equity would seed a sovereign wealth fund Sanders estimates at roughly $7 trillion, managed by a seven-person independent commission.
  • A 5% annual dividend from the fund would send more than $1,000 a year to every American, with the remainder going to health care, education and housing.
Read full analysis →
View on Bluesky · ♥ 13 ↻ 3 ↩ 1 · 2 from the directory shared this · 39d ago

2. Chiang claims that the fact that next token selection is the output method of LLMs as if that was some sort of proof of their non-mindedness. It is irrelevant, actually. The most cutting-edge model publicly released, DiffusionGemma by Google, doesn't primarily use next-toke…

DiffusionGemma — Google DeepMind deepmind.google
View on Bluesky · ♥ 16 ↻ 1 ↩ 2 · 26d ago

Recent commentary

The secret to understanding chud aesthetics is that everything they know about art comes from 1990s porn. And still, they can't measure up to Gregory Dark, Tera Patrick, Heather Hunter, Janine Lindemulder... Here's the AI slop version of "Odyssey" that Elon Musk thinks is "historically accurate:"

View on Bluesky · ♥ 115 ↻ 17 ↩ 21 · 6d ago

Everything is classifiable as pornographic. Never doubt this. Bluesky, a website run by non-religious people, thinks that a post about AI that links to Wikipedia is porn. Your work documents, your TV shows, your political views--everything--is porn to someone out there.

View on Bluesky · ♥ 97 ↻ 33 ↩ 7 · 26d ago

It really does seem as if 99% of the people who think large language models have consciousness are men. I can think of one woman who seems to agree. Am I wrong to think that's the demographic split?

View on Bluesky · ♥ 34 ↻ 3 ↩ 3 · 60d ago

In the 1990s, proprietary software vendors fearmongered against Linux and FreeBSD as "communist" programs that would destroy capitalism. In the 2020s, proprietary software vendors are resurrecting the exact same scare tactics against open-weight AI models. Govts must support free software AI.

View on Bluesky · ♥ 26 ↻ 5 ↩ 2 · 7d ago

1/2 In the latest "Theory of Change," I argued that Pope Leo XIV's repeated rejection of reactionary and anti-democratic politics was explicitly modeled after Leo XIII's "Rerum Novarum." Lo and behold, today, Leo's new AI encyclical was dated explicitly to the earlier Leo's document.

View on Bluesky · ♥ 29 ↻ 3 ↩ 1 · 64d ago

Most things people think are bad about AI are actually problems with unregulated capitalism, which naturally transitions into monopoly.

View on Bluesky · ♥ 23 ↻ 0 ↩ 2 · 63d ago

Romance scams have proliferated on many social media platforms, particularly X. One thing I've wondered, do these accounts featuring AI-generated women target female-presenting accounts in DMs? I have other accounts that are non-personal & they seem much less targeted. Example below:

View on Bluesky · ♥ 9 ↻ 2 ↩ 1 · 49d ago

In Matthew Sheffield's orbit

Center = Matthew Sheffield. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.