this seems important
54
nlev retweeted
Why are you surprised it built a Dyson sphere? Encapsulating the Sun is a verifiable task—you can see it in the sky or you can’t. Who knows if it will generalize?
trained at 4-bit, layer looping, 26B MoE... and trading blows with Qwen3.8 27B on terminal-bench.
thinky cooked
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/ink…
👀👀👀
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/ink…
4
302
uh oh
Replying to @cb_doge
?
We aren’t serving open source models
Readers added context they thought people might want to know
Cursor serves hosted open-weight models from Chinese providers including GLM 5.2 (documented as served via Fireworks) and has announced GLM 5.3/Flash and DeepSeek V4.1 Flash availability; it is only exploring phasing them out.
cursor.com/docs/models/gl…
x.com/cursor_ai/stat…
x.com/cursor_ai/stat…
cnbc.com/2026/11/20/cur…
68
submitting my application for Space Repoman
“.. lenders have expressed concerns that if the group later decided to shoot the valuable chips that underpin their loans into space, it would prevent them from reclaiming and selling the semiconductors in the event of a default ..”
@FT $SPCX
ft.com/content/33b5643c-2957…
1
2
200
We seeded 2m canaries into popular AI services and tracked where they were stolen and abused.
This identified:
- Three compromised providers
- Resellers stealing prompts via built-in logging
- Providers ignoring ZDR & privacy flags
dragos.com/blog/ai-supply-ch…
nlev retweeted
The top 3 principles of cognitive weightlifting:
Retrieval practice
Mastery learning
Spaced repetition
i.e., Do the reps, use the right weight, and stay consistent.
"why own compute when the models are cheaper via API?"
Looks like @PrimeIntellect was compromised in a very strange way. No public comms from them yet, but disclosed by OpenRouter to affected folks (such as myself)
79
Gemini 4 Lead: DeepMind's latest dense model with minimal guardrails and low IQ
tomorrow DeepMind will be dropping Gemini 4 Sodium, a fast yet powerful frontier model which costs just $11.75/M (introductory pricing, raises to $16.50 soon) and is their best model for various things and stuff, rivaling Argon, Carbon, Hydrogen, and Neon in several situations
1
2
195
nlev retweeted
The more powerful your tools become, the more expensive it is to be clueless about what they're doing.
Hard to even imagine this today.
Replying to @littletales
Eyvind Earle painted every single background in Sleeping Beauty by himself. It took him five years. Walt loved his work so much he didn’t care it took that long.
2
67



























