MTS @thinkymachines. previously pre-training @googledeepmind, @character_ai, and @aiatmeta.
NYC
Joined February 2008
- Tweets6.6K
- Following1.3K
- Followers6.1K
- Likes25.9K
Stephen Roller retweeted
I'm not a doomer but I do think a lot about how the predicted Mayan apocalypse came within less than three weeks of AlexNet being published at NeurIPS.
Stephen Roller retweeted
We are making lots of improvements to Tinker. Now we have a new optimizer to try out!
Dimuon is a new optimizer we built for LoRA fine-tuning. In our SFT tests, it often reaches lower loss in fewer steps than AdamW. We're still experimenting with Dimuon; you can Tinker with it today and share your feedback!
See our optimizers guide at tinker-docs.thinkingmachines…
Stephen Roller retweeted
This is the real Turing test if you think about it
Two days ago, GPT-6 Astra broke a yet unsolved German Army Enigma message from 1941. Amazingly Astra was able to autonomously:
- Search historical archives
- Compare uncertain letters
- Find contextual clues
- Build an Enigma simulator
- Write cryptanalysis code
- Run parallel experiments
- Test competing keys
- Recover the plaintext
- Cross-check the results
1/n
Stephen Roller retweeted
Today, we are introducing Inkling.
Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.
thinkingmachines.ai/news/int…
Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵
Stephen Roller retweeted
We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2.
Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication.
We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating.
🧵👇
Stephen Roller retweeted
The biggest bullshit move by DHS in its history. So everyone on a O1 or H1B visa would have to stop working legally in the US, go back to their country and wait for years of backlog? This includes top scientists in our universities, founders of billion dollar companies (at least 3 just in our portfolio would be affected by the way). And if we look at individual countries it becomes even more bs. Indians would have to wait decades. Russians don’t have anywhere to go (there is no US embassy in Russia, hello?).
This is the worst imaginable way to disrupt important work for the country and pretend you’re fighting some loophole.
Stephen Roller retweeted
Cluster magicians and GPU whisperers, come join us!
We’re looking for supercomputing engineers to build the infrastructure behind real-time interactive models, Tinker, and large-scale training: scheduling, storage, networking, reliability, and distributed systems at scale.
Hiring in NYC and SF
job-boards.greenhouse.io/thi…
Stephen Roller retweeted
People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way.
We share our approach, early results, and a quick look at our model in action.
thinkingmachines.ai/blog/int…
Stephen Roller retweeted
This is the kind of shit Godspeed You! Black Emperor would use as an album cover with a title like "And the Holy Flame Blinded us for 333000 Years Pt. II"
Stephen Roller retweeted
DeepSeek-V4 uses our Hash routing approach developed back in 2021 -- see screenshot of their tech report! (Looks like a great model, congrats!)
Bonus note: our same blogpost (& paper) back in 2021 also introduced 'looped transformers', but we called that staircase & ladder (see screenshot): parl.ai/projects/params_vs_c…
huggingface.co/deepseek-ai/D…
Stephen Roller retweeted
me: Make me the most AI slop image that ever AI slopped. The pinnacle of slop. A seminal work on AI slop.
ChatGPT Images 2.0:
Stephen Roller retweeted
Long context windows are now available for select models on Tinker!
- 128k tokens for Kimi K2.5 and GPT-OSS-120B
- 256k for Nemotron 3 Super 120B and Qwen3.5 397B.
For more details and pricing, see our full model lineup: tinker-docs.thinkingmachines…
