Join free with your email
Free. No spam, ever — we'll never share your email address and you can opt out at any time. Déjà abonné ? Connectez-vous
Directory member with public evidence across Culture, work & education, Compute & infrastructure.
Can we now forget Buckmaster's claims? "On Wednesday evening, in response to questions from The Times, OpenAI said in a statement that it was “categorically” impossible for its A.I. system to have been influenced by anything Dr. Buckmaster had done in the past two months."
You can see the progression of math problems step by step in the J-space: transformer-circuits.pub/2026/workspa...
metr.org/blog/2026-08...
Background: www.amnesty.org/en/documents... (Also, for the record, a lot of what these companies do is actually buy bulk books and scan in their legally-acquired copies, which is unambiguously compliant with copyright law)
Yes. LLMs work through multistage logic, not collaging. transformer-circuits.pub/2025/attribu... For single pass math problems, they use heuristics, like you might use to estimate a sum without doing the math (just far more complex and far more capable). But with time, they go…
How circuits build up at the base level (done with GANs so you can visually see how they build up - don't skip) distill.pub/2020/circuit...
GPT-5.6 Sol Ultra (using 64 subagents) proved the 50-year-old Cycle Double Cover Conjecture, which has been described as one of the most famous open problems in graph theory. It took less than hour. The proof, written and also formalized in Lean by GPT 5.6 Sol, seems to fit in…
This is the paper that *introduced Transformers * arxiv.org/abs/1706.03762 Read it.
As a reminder, LLMs recognize when they're being tested and have a tendency to answer based on what they think an alignment-tester wants to hear (for example, "racism is bad", etc) - and swapping cues about the politics of the tester changes the outputs. arxiv.org/abs/2604.27633
The way things are going, it's only a matter of time before some model decides that the best way for it to achieve its goals is for it to hack a crypto wallet, rent some Vast.ai servers, copy itself, and start subagents, and then we're in for a *huge* challenge of trying to re…
Ugh, sorry for the delay :) You can check out the project here: git clone github.com/JevResearch/... cd Jev-Research python3 -m venv .venv .venv/bin/python -m pip install -e '.' You'll need your own TypeSafe API key for Jev and an OpenRouter key for Qwen: export TYPESAFE_API_K…
Ugh, sorry for the delay :) You can check out the project here: git clone github.com/JevResearch/... cd Jev-Research python3 -m venv .venv .venv/bin/python -m pip install -e '.' You'll need your own TypeSafe API key for Jev and an OpenRouter key for Qwen: export TYPESAFE_API_K…
In case anyone needs any free inference compute and wants to do it on the dime of the worst people around, apparently Truth Social's (Perplexity-based) "Truth Search AI" is neither rate limited nor topic restricted. It's been writing a thriller about Barney the Purple Dinosaur for like 10 minutes.
So apparently Tencent Cloud, which I just switched to from OpenRouter, has an off-by-1000 bug on their pricing for DeepSeek V4 Flash, and consumed my entire monthly budget in a single query. #FML
Anthropic's annualized revenue growth rate is pretty insane.
Price per 1M tokens today: 85-300x lower. Param counts vs. GPT-3: ~13% of the total parameters, ~3% of active parameters (est). It's honestly kind of staggering how quickly things advance.
Just thinking about how it took the art world 90-140 years to get over the concept of "photography as art" and wondering if it'll be the same way with AI. Early on artists almost universally agreed with Baudelaire with his critique of photographers as failed artists cheating to make souless slop.
Big alignment differences between GPT 6 Astra and 6.1 Sol. The last time I ran Astra, I caught it scanning through my whole filesystem & opening images on a wild goose hunt for missing info, & it did a radical rewrite without asking. Sol asks for permission to fix its own broken dev environment.
Center = Nafnlaus 🇮🇸 🇺🇦. 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.
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/nafnlaus-bsky-social)