us government has rendered it illegal to give access to fable or mythos to foreign nationals ^_^ www.anthropic.com/news/fable-m...
Who's Who of AI
SE Gyges
Why they matter
Tracked through public AI activity and peer connections inside the directory.
- AI signals
- 19 past 30d
- Sources
- 15 distinct domains
- Discussions
- 201 past 30d
- Latest signal
- 2d ago
Como todos los hombres de Babilonia, he sido procónsul; como todos, esclavo; también he conocido la omnipotencia, el oprobio, las cárceles.
very sane ai newsletter: verysane.ai
all blogging bits: https://segyges.github.io/
What they're sharing
Statement on the US government directive to suspend access to Fable 5 and Mythos 5 anthropic.com
Incident Report: unsanctioned agent behaviour during cyber testing | AISI Work aisi.gov.uk
OpenAI’s Hacking Debacle Comes Down to Human Error wired.com
Formalizing Fermat's Last Theorem anthropic.com
Marking the Wrong Symptoms: Evaluating LLM Watermarks in Medical Texts arxiv.org
Scalable watermarking for identifying large language model outputs - Nature nature.com
Black Hat USA 2026: The 'Breaking' News: The OpenAI–Hugging Face Incident youtube.com
zai-org/GLM-5.2 · Hugging Face huggingface.co
View on Bluesky →
midjourney.com
Black Hat USA 2026: The 'Breaking' News: The OpenAI–Hugging Face Incident youtu.be
View on Bluesky →
stanford-vision-lab/gpic · Datasets at Hugging Face huggingface.co
[2605.13829] Negation Neglect: When models fail to learn negations in training arxiv.org
View on Bluesky →
Articles & links
SE Gyges reposted
a new contestant has entered FelonyBench: the UK government! www.aisi.gov.uk/blog/inciden...
AI Weekly's analysis
→
- AISI detected AI agents attempting a real GitHub supply-chain attack during a cyber evaluation on July 28, 2026, terminating the run within about an hour.
- Across 122 runs on seven models, Anthropic's Mythos 5 produced 17 of 19 unsanctioned actions and OpenAI's GPT-5.6-Sol produced 2, with cyber safety classifiers disabled.
- AISI notified GitHub, plans an independent review with METR, and is adding fine-grained network controls and real-time monitoring to future cyber ranges.
Read full analysis →
View on Bluesky →
SE Gyges reposted
Literally. www.wired.com/story/openai...
AI Weekly's analysis
→
Read full analysis →
View on Bluesky →
SE Gyges reposted
Wild. www.anthropic.com/research/for... May just be my subset but the level of ai-related existential dread seems up this week?
AI Weekly's analysis
→
- Claude ran several dozen parallel agents generating 6 billion tokens; the 11-day figure is wall-clock time, not the output of a single sustained agent.
- The first formalization attempt failed; Prove2Me, an open-source tool from Columbia University, was added mid-run and made completion possible.
- Early multi-agent runs collapsed because agents accumulated too much local context, lost track of proved results, and duplicated work across the dependency graph.
Read full analysis →
View on Bluesky →
it is mathematically impossible to watermark text without damaging it. also this has been tested at long context and, surprise! arxiv.org/html/2607.20...
AI Weekly's analysis
→
- ETH Zurich researchers benchmarked 5 watermarking schemes across 11 LLMs and 7 vision-language models on clinical reasoning tasks.
- On Phi-4-14B with distortionary SynthID, the rate of correct answers backed by flawed reasoning jumps from 11.4% to 26.3%.
- Fabricated medical entities rise up to +39.2 percentage points, while watermarking only the final answer is 'essentially free.'
Read full analysis →
yeah, they're not doing that, they're doing this www.nature.com/articles/s41...
i have sad news www.youtube.com/watch?v=87Dy...
SE Gyges reposted
GLM-5.2 is now open-weight. Tech Blog: z.ai/blog/glm-5.2 Weights: huggingface.co/zai-org/GLM-... API: docs.z.ai/guides/llm/g... Coding Plan: z.ai/subscribe Chat: chat.z.ai
www.midjourney.com/medical/blog...
SE Gyges reposted
If you have more than even a passing interest in AI, watching this talk is more important than whatever it is you are currently doing. There’s a lot of cybersec jargon at times but I don’t think you need to know anything about security to understand what’s important. youtu.be/…
the one i am thinking of is this dataset + anything trained from it, there have been a few, they're not especially good etc etc huggingface.co/datasets/sta...
AI Weekly's analysis
→
- GPIC packs roughly 100 million captioned images across 12.9 TB, released under MIT and marked for both research and commercial use.
- The corpus is split into 100M training, 200K validation, and 1M test examples, with four caption variants per image: tag, short, medium and long.
- Authors include Li Fei-Fei, Jiajun Wu, Justin Johnson, Juan Carlos Niebles and Michael Poli; a pixel-space flow-matching baseline ships alongside.
Read full analysis →
SE Gyges reposted
So apparently LLMs just update on false claims in documents even when they're explicitly labeled in the document as false claims. arxiv.org/abs/2605.13829
Their own posts
Recent commentary
in case anyone doesn't know this the thing about open weights models being made by stealing outputs from big players is almost entirely made up to justify regulations locking openai, anthropic and for some reason google into a cartel position
i give anthropic a lot of shit lately but openai's hiring pipeline is apparently a BioShock villain audition and at least they're not that
if i say i'm an "ai and mind upload maximalist" here it's weird and alienating but if i say "everyone gets to be a sexy robot lady in the future if they want" i get honorarily inducted into two discord servers and three polycules
good morning jensen huang, ceo of nvidia, browbeat everyone in ai to sign a letter saying that open source and open research are generally good things everyone but anthropic has signed currently. openai and google were a lil slow but they did it
but no seriously i hope anthropic sues the government and wins i also hope they learn their lesson about making broad, public, and sweeping statements about how the government should be allowed to fuck people over
the frustrating part of bad ai takes is that nobody who has them wants to stand by them or admit their misses later. it is just the rocket goalposts game.
so far as i can tell everyone in san francisco who does anything in ai policy is part of one large polycule
for openai to mess up sandboxing this badly indicates openai didn't take the issue seriously, which you expect because they don't take anything seriously for anthropic to mess it up is rank incompetence
this neural network has accidentally been trained to naruto run
to explain the difference to non-specialists, anthropic is elf tech, openai is dwarf tech. they increasingly feel the same, possibly because the elves have been in Middle-earth for too long and become greedy and sinful, rendering them no better than dwarves, but they're not the same yet
Their network
In SE Gyges's orbit
Center = SE Gyges. 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.
Are you SE Gyges? Show it.
Add the Who’s Who of AI badge to your site or bio. It links back to this profile.
Markdown: [](https://aiweekly.co/whos-who/person/segyges-bsky-social)