Pinned Tweet
The distillation debate is looking pretty one-sided around here! I've been thinking about this for quite a while (along with Yash Savani), so let me add some variety. Regardless of where you stand on the IP issues:
Distillation attacks make the AI ecosystem LESS original, less safe, and -- believe it or not -- also less open. We expand on these points in a blog post linked in the replies.
Our favorite realization is that distillation attacks lead to AI monoculture, and this monoculture could create unanticipated, systemic security risks for both individuals and companies.
Asher Trockman retweeted
GEMINI IS BACK 👑
Gemini 4 Carbon has topped the FutureSim leaderboard dominating every other model throughout the simulation!
It is also very efficient, using the least number of tokens, search tool calls and subagents compared to Opus 5.5 and GPT 6.1 Sol.
Asher Trockman retweeted
You will notice that Gemini Carbon thinks differently than Claude and GPT. It often comes up with entirely different but more clever solutions than competitor models.
This is precisely because we never distilled other models. Distilling other models creates a poor imitation rather than unique intelligence.
Asher Trockman retweeted
My talk on this work at our Research @TTIC_Connect seminar is available online: youtube.com/watch?v=F0R8Oxku…
Asher Trockman retweeted
Mathematics and AI now define an interface full of fascinating questions that extend beyond a single discipline, and that require dedicated scientific inquiry. I am excited to explore these questions alongside many colleagues, and grateful to Courant's leadership for entrusting us with this mission!
Asher Trockman retweeted
Meet Gemini 4 Hydrogen: built for rapid coding, multimodal synthesis, and everyday workflows.
It finishes tasks 2.5x faster than 3.8 Flash and surpasses Argon on the @ArtificialAnlys Intelligence Index.
Asher Trockman retweeted
Introducing Gemini 4 Hydrogen, an extremely capable and fast model built to bring Gemini 4 to everyone!
Available to all tiers of Google AI Studio, Gemini API, Gemini for business, the Gemini app, Google Search, and paid Google AI Pro + Ultra subscribers in Google Antigravity.
Asher Trockman retweeted
Great to see Mamba-3 driving strong performance across various modalities!
Looking forward to the team’s progress!
Today we're launching ci-proto, @Continu_Intel's first time-series foundation model.
- #1 on fev-bench, ahead of Google, Amazon and Mistral
- 64× the context window of others
- Open weights and available for commercial use
Proud of the team. This is just the beginning.
Asher Trockman retweeted
We've posted a 156-page tech report on Beam! We have good systems and it's worth reading. It covers all phases through pre-training, mid-training, SFT, RL, and OPD and goes into the infra, data, and scientific decisions we made throughout the process
📚 reflection.ai/documents/beam…
We've released the weights for Beam!!
🤗 huggingface.co/reflection
📄 Model card: reflection.ai/beam-model-car…
🧪 Try Beam in our playground & get an API key: platform.reflection.ai/playg…
📚 AND we take open model safety seriously and have posted our safety report: reflection.ai/beam-safety-re…
Asher Trockman retweeted
Friday, November 6th at 11:00AM: Research at TTIC presents Siddharth Prasad with a talk titled "Inference Auctions." Please join us in Room 530, 5th floor.
Asher Trockman retweeted
Meet Inkling 2 Small, our newest model from @thinkymachines — on Tinker today, with weights landing after the completion of our final safety checks 🦺
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor.
Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
thinkingmachines.ai/news/ink…
Asher Trockman retweeted
Inkling 2 Small is here 🚀 Give it a try ...
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor.
Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
thinkingmachines.ai/news/ink…
Asher Trockman retweeted
How much of your training compute should go to rewriting your data as you scale pretraining?
We trained a *22B* hero run at 33× our fit range and it ended up being within 1.1% of the pre-registered loss forecast!
🧵Thread from my amazing co-lead @AhmedSQRD 🧵
👇👇👇
Asher Trockman retweeted
Synthetic rewrites are now standard to improve data quality and when data is limited. But how much compute should you spend rewriting? And can a model trained on rewrites beat the rewriting model?
We fit scaling laws to answer both:
1/🧵
Asher Trockman retweeted
i had a chance to try Argon before joining deepmind but after using Carbon internally I can tell you that this team is absolutely cooking and we’re only just getting started.
Carbon is writing all of my code and I can trust it to run all night, for multiple days, making intelligent decisions autonomously.
there is still lots of work to do so we’re excited to get this in your hands and hear your feedback.
we can’t build AI for everyone without the community 🖤
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
[LG] Rewriting Scaling Laws for Synthetic Data with Compute and Data Constraints
N Roberts, A Ahmed, S Kotha, P Liang,… [Princeton University & Stanford University] (2026)
arxiv.org/abs/2610.31867
Asher Trockman retweeted
I'm an extremely heavy user of both Argon and Astra, and this matches my vibe based impression that Argon is slightly better. (Generally I think benchmark scores undersell it)
Gemini 4 Argon is underrated 😱.
When auditing the traces, we found that Pi Harness just ends the run whenever it hits a fallback (e.g. for security task) in 20% of the runs. We rerun Argon with Antigravity Harness, and now it beats Astra and sits right behind Opus 5.5. Congrats to the team @GoogleDeepMind and looking forward to the new ckpt.












































