physics @eth prev @ETH_agent_lab co-founder @seldon_tech
Joined February 2020
- Tweets680
- Following471
- Followers430
- Likes8.4K
Pascal Bertrand retweeted
Super cool! I encounter the same thing when using agents for astronomy. They miss the observation window by thinking too long and ultra-fast is too expensive for multiple hours of observation. (vid is sped-up)
opus 5.5 is way better than astra at it
LLMs might be superhuman at math, but can they act in real time?
Not even close!
Presenting Gym-Anything-RT, a framework for generating real-time minigames in which the game keeps moving while the model thinks.
Even the best frontier model, GPT-6 Astra, scores just 9%!
🌐: gym-anything.github.io/gym-a…
🧵
Pascal Bertrand retweeted
i’m going to be presenting some cool work we’re doing at seldon next week, come by if you’re in zurich :))
We're excited to announce ZurichCV#15 is next Thursday at the @ETH_AI_Center!
Said Gürbüz (ETH Zurich, IBM Research) will present DeskForge: Controllable Desktop Environments for Training Computer-Use Agents.
Lorenzo Steno @lsteno (Seldon Technologies @seldon_tech) will talk about building RL environments for computer use.
RSVP: zurich.tech/events/zurichcv-…
(PS: this is our new account for zurich.tech. please help out by sharing!)
Pascal Bertrand retweeted
Was fun sharing the stage with @PascalBtd at AI+X, looking back at the early days of the @ETH_agent_lab and what we're building now with @aionic_labs and @seldon_tech.
Thanks @ETH_AI_Center, Sören Fehrmann and Swisscom Ventures for having us!
Pascal Bertrand retweeted
we’re hiring in zurich! seldon.global/careers
Pascal Bertrand retweeted
Our Co-Founders @PascalBtd and @atoof_sh sat down on the Europe on Edge podcast and talked about the current state of multimodal AI and the company's mission. If this mission resonates with you, have a look at our open job listings (links are below).
Pascal Bertrand retweeted
we're hiring researchers and engineers in sf and zurich! come build the future of multimodal ai
seldon.global/careers
Pascal Bertrand retweeted
a bit of a longer post with my experience training recursive language models with rl. it was quite hard and i had to fail a bunch of times to achieve a somewhat interesting final result. hope you'll enjoy it!
Pascal Bertrand retweeted
Can computer-use agents use complicated graphical software? You’re watching GPT-5.6 Sol operate Autodesk Fusion for 250 turns as it attempts a real mechanical-design task. The run is part of a benchmark aimed at the question "how good are agents at CAD?"
@seldon_tech's CADBench contains 105 mechanical-design tasks tested across 10 frontier models, and the results show how early reliable, autonomous CAD work remains:
More than two-thirds of the tasks are still unsolved.
That gap matters as agents make their way into the professional tools used to design the physical world. Getting around Autodesk Fusion is one thing, but producing a correct, editable model an engineer can pick up and use is much harder.
CADBench gives us a way to see and measure that frontier.
Great work from the Seldon team! Explore CADBench: seldon.global/blog/cadbench
Pascal Bertrand retweeted
why haven’t other engineering practices seen the same model improvements as coding? we measured models’ ability to interact with a cad environment, and the results were super interesting, with some curious upsets in model and harness rankings. i’m super excited for what’s to come in computer use
How good are agents actually at CAD?
Today we release CADBench, evaluating 10 Models on 105 realistic tasks in Fusion 360
Fable 5 and Gemini 3.7 Flash are #1 with only 24.6% pass rate
Pascal Bertrand retweeted
@GoogleDeepMind and @Alibaba_Qwen released new models this week. These are the results:
Gemini 3.7 Flash is the new State-of-the-art model on VGI-Bench.
Qwen 3.8 Max comes in as fourth place in after Gemini 3.6 Flash.
Pascal Bertrand retweeted
there are so many levers to reduce the cost of your agents.
if you look at the traces what your agent is doing, you will see that most tokens are spend to look up information.
we have 2 choices either give the models a better tool or delegate this task to a cheaper model. but unfortunately most cheap models are very bad at finding the right information.
we already showed that mixedbread is the better tool to find evidence faster and save tokens, but still your agent had to read a lot of documents.
now we are closing the gap and releasing a smart model which can be used to delegate by your main agent.
save tokens, while still using frontier. this is a way bigger lever then model routers.
Pascal Bertrand retweeted
Introducing Toast 1, our first specialised search agent.
Toast 1 sets a new Pareto frontier for agentic search models.
Frontier search quality, across all domains, 12x faster, at 1/10th of the price.
Pascal Bertrand retweeted
Introducing VGI-Bench: a multimodal, holistic benchmark probing 12 distinct visual and audio-visual skills.
550 human-curated questions, designed to mitigate the common mistakes in today's video benchmarks and expose pragmatic failures of state-of-the-art models.
Best model: 64.73%. Humans: 84.5%.
Pascal Bertrand retweeted
New: mxbai-rerank-v3.1-listwise
An upgrade to our listwise reranker.
→ GPT-5.6-sol (high) ranking quality at 61× the speed
→ up to 54% faster than v3 on long docs
Built for:
- recency-aware ranking
- source-priority resolution
- multi-step composite instructions
Available today in Mixedbread.
Pascal Bertrand retweeted
We're hosting the 3rd Robot Learning Club next week, and we've already passed last event's sign-ups.
Format: 5–10m presentations by local authors, or a paper someone found cool and wanted to share (this could be you)
We're holding a few spots for new joiners. Signup link below.
Pascal Bertrand retweeted
Yesterday, we hosted Curious Machines, our first event with talks from @matteosaponati and @lsteno about their work on auto-research (more about that soon!). Thanks to everyone who attended and for those we could not accommodate this time, stay tuned for our next events
Thanks to @ETH_agent_lab for the location and @GHollbeck for taking pictures
