I build AI systems for programmable money, astrophysics, coding, and fun | extreme research at @alphakek
Joined October 2020
- Tweets 4,309
- Following 197
- Followers 945
- Likes 20,274
1,522 Photos and videos
I keep saying this for years: 99% of “AI alignment” is actually about aligning humans to other humans
"They screwed us": Personality clashes sent Anthropic's models offline axios.com/2026/06/15/anthrop…
3
125
>not tested on Le Chaton Fat Preview
opinion discarded
14h
Europe’s AI champion Mistral vulnerable to Russian disinformation, study finds ft.trib.al/LlESkdV
4
139
What running Qwen3.6 27B inside of Claude Code feels like
6
140
Jun 15
“Verifying identities” was the reason why this legislation was introduced in the first place
Very sad to see other countries repeating Russia’s playbook
Jun 15
JUST IN: UK Government clarifies adults will still be able to use social media by verifying their identities with digital IDs, facial recognition, passports and credit cards.
2
4
99
Jun 15
There’s one quite possible scenario for high end GPUs to cost <2000€:
1) Euro falls into a hyperinflation spiral so that a GPU costs 2000000€
2) Euro does a Zimbabwe-like redenomination reform so that 2000000€ converts to 200€
Jun 15
I considered buying a RTX Pro 6000 96GB at 9000€ last fall. Today it costs 16000€ in Finland :(
It's basically just a RTX 5090 with 3x RAM. This is what professionals are willing to pay for a big die GPU. We will never see big die gaming GPUs for <2000€ anymore :(
3
97
Jun 13
Putting the last week’s news into pretrain will quadruple the paranoia of next gen models
Emergent misalignment
Jun 13
ChatGPT 5.6: Thinking... *sees Fable getting banned* haha Strawberry totally has 15 r's in it, am I right guys 🤪
1
5
437
Jun 13
GenArrest
Jun 13
JUST IN: UK police launch criminal investigation into officer accused of using AI to “create evidence”
3
129
Jun 13
Что ж ты, фраер, сдал назад…
Jun 13
NEW: Anthropic claims the capability cited by the U.S. in restricting Fable 5 is already widely available from other models, including OpenAI’s GPT-5.5.
1
64
Jun 12
UK is gonna discover the Jevon’s paradox for arrests 💀🥀
Jun 11
JUST IN: United Kingdom announces "PoliceAI" to help fight crime — claims it will free up 6 million human officer hours / year.
4
178
Jun 12
This makes me think Thinky’s interaction models will go beyond the voice chat: pretty sure we will see background models running in async mode and thinking in neuralese, paired to interaction models that converse with the user in plain English
Jun 11
One interesting pattern with Fable 5 is that it will often say things that are gibberish when I use it for coding. Things like "The morning's slim-scan fix cured the scan hang", "this is a latent-drift API-shape wrinkle", etc.
When I ask why it does this, Fable explains that it invents codenames while reasoning about the problem, then fails to realize they're meaningless to me. Its neuralese is blending into its output because of a theory-of-mind failure about what's in its head vs. mine.
3
50
Jun 10
scared to think how big labs would prevent *this* kind of distillation attacks
6
89
Jun 10
Jun 10
JUST IN: Florida hospital reveals Palantir software has cut sepsis deaths by more than half since it was installed.
4
136
Jun 10
I feel like scaling synthetic data along a single axis closes capability gaps very quickly while having little to zero effect on jaggedness gaps, although I am not sure if that’s a fundamental issue or rather a skill issue
Jun 10
For the last week, I’ve been running Nemotron Ultra on my local GPU box as a coding agent alongside with Claude Code and Codex
After 200M processed tokens, my verdict is the following: it is almost as capable as the closed-weights frontier, but noticeably more jagged — and it’s actually bullish, because you feel “the recipe is already here, we just need to feed more diverse data into it to close the gap”. It will only keep getting better
Excellent work @NVIDIAAI @ctnzr @llm_wizard !
4
144
Jun 10
For the last week, I’ve been running Nemotron Ultra on my local GPU box as a coding agent alongside with Claude Code and Codex
After 200M processed tokens, my verdict is the following: it is almost as capable as the closed-weights frontier, but noticeably more jagged — and it’s actually bullish, because you feel “the recipe is already here, we just need to feed more diverse data into it to close the gap”. It will only keep getting better
Excellent work @NVIDIAAI @ctnzr @llm_wizard !
1
2
9
820
Jun 10
It’s ironic and sad that Anthropic or of all labs became the first ones to ship openly misaligned models to public
Jun 9
mythos will be bad ON PURPOSE on ai "frontier llm research" tasks, this is very very sad for the research community
also the fact that this is un purpose not visible to the user is crazy
1
1
9
231
Jun 9
That’s why we need open models
Jun 9
Interesting tidbit from the Mythos/Fable system card: Anthropic are invisibly nerfing any requests that target frontier LLM development.
1
3
100
Jun 9
IMO the real AGI/ASI would effortlessly explain even the most complex concepts to us humans (if it wishes to)
Jun 9
Claude Code about my new experiment idea: "But XXX is the load-bearing caveat, and the YYY is precisely where it bites."
Me: *prove that it's incorrect*
Claude Code: "You're relocating the load-bearing commitment, and it's a better location."
2
74