thomas (@thomasinovic) header
thomas (@thomasinovic) profile picture

MTS @withbeaker 🧪 🇺🇸 | ex ML @ThalesGroup 🇨🇦 | 🎓 AI @TelecomParis @Polytechnique 🇫🇷

New York, USA
Joined July 2026
Photo by thomas (@thomasinovic)Photo by thomas (@thomasinovic)
We made an AppWorld agent 36.6% faster and cut its model cost by 34.6%, while keeping a similar goal completion score. The model stayed the same. @withbeaker autonomously tested changes to the agent’s prompts and tools: fewer repeated lookups, shorter instructions, less unnecessary context. It then checked the optimized agent on tasks it hadn’t used during optimization. 2h24 from baseline evaluation to final results. I wrote up what changed, with the results and the full run 👇 withbeaker.ai/blog/speeding-…
3
2
2
12
1,254
thomas retweeted
Introducing Orion, a foundation model that predicts where you look. Orion finds twice as many UX bugs as frontier models, cuts computer-use latency by 31%, and even won a drone race.
67
113
18
2,343
84,073
Nice work ! Would be interesting to do harness engineering (@withbeaker) on top of those LoRA adapters...
Announcing harness adapters for @Alibaba_Qwen Qwen3.8-27B on Pi harness! We're releasing a LoRA adapter that makes Qwen3.8-27B a much better and efficient agent behind Pi harness. SWE-Bench Pro (Hard 51 subset) pass@1: 29% -> 63% Terminal-Bench 2.1 pass@1: 40% -> 56% Much more efficient: 55-62% fewer tool calls per task 🤯 huggingface.co/harness-hub/Q… The era of harness-specific adapters is now on! More harness adapters to come in the next days 🧵
3
41
Voici le taux de chômage des jeunes diplômés de Grandes Écoles d'ingénieurs prestigieuses en 2024 ou 2025. C'est catastrophique. Les jeunes Polytechniciens ont un taux de chômage de ~5,7%, bien qu'ils sortent de l'école la plus prestigieuse de France ! 🧵⤵️
Même les ingés sortis des écoles les plus prestigieuses n'arrivent pas à trouver un job, ceux qui y parviennent font un travail de technicien supérieur parce qu'il n'y a plus d'industrie en France, mais ce GÉNIE d'Édouard Philippe veut en former 100 000 par an (contre 49 600 actuellement). 🤡
176
341
85
1,998
247,573
Je teste @mistralvibe aujourd'hui, on va voir ce qu'il donne face a codex/claude/agy Ceux qui ont teste vous en pensez quoi ?
2
4
341
You never know if errors in evaluation datasets could be skewing your agent’s score -> @withbeaker now helps you identify them easily
What if the failure you're trying to fix isn't a failure at all? Faulty evals waste engineering time and can teach agents the wrong behavior. @withbeaker checks the eval while improving your agent. On AutomationBench: ~14 percentage points of training-score improvement per run, plus 8 evaluation concerns flagged across two runs. On LiveNewsBench: 20 questions flagged, including 6 confirmed grader errors. Before you fix the agent, check the eval 👇 withbeaker.ai/blog/before-yo…
4
236
thomas retweeted
BREAKING: Former Nvidia CEO Jensen Huang had an intimate relationship with a second employee, Reuters reports.
51
2,524
42
33,461
387,397
thomas retweeted
Mouvement de soutien international à Jensen Huang 📍 Place de la République 🕡 À partir de 18h30 Venez nombreux
616
849
140
16,928
738,100
I also had an affair with @JensenHuang
🤖 Made with AI
3
2
14
948
thomas retweeted
🤖 Made with AI
203
277
88
4,198
231,532
Last week, @rikiyatakehi fixed a looping model by editing its weights. We pointed an AI agent at the same problem. It never ran the model and autonomously found a different fix. No training run, nor gradients. That's weight-space programming. We gave the same task to Beaker, @withbeaker, our autonomous agent-improver. It analyzes how an agent behaves and fails, formulates hypotheses on what and how to fix, validates or invalidates them, in a loop, to hill-climb a pre-defined score. The target: Qwen3.5 0.8B, which fails 93% of MATH-500, often by getting stuck in loops until it runs out of thinking tokens. The model has a tendency to generate "Wait", and Beaker turned it down. From the weights and the scores of its own edits, it figured out what was going wrong and autonomously fixed it by editing 1,024 numbers, a millionth of the model's parameters. MATH-500: 7.4% -> 37.4%. That's a different fix from the original blogpost. They found a way out of generation loops, and Beaker found a way to not fall into them. Theirs is stronger alone (51.1% in our harness), but the two stacked together compound: 58.7%. When properly harnessed, less capable models are good weight space programmers. Beaker powered by GPT6.1 Sol found the same neurons as Beaker with GPT6 Astra, at $4-16 per run vs $28-79. We also pointed Beaker at new goals. We told it to tune the weights such that the agent comes up with shorter answers, and it did, with answers a third to half shorter. Beaker already tunes prompts, skills, memory, tool implementation, orchestration, etc. to improve agent quality, latency cost and determinism. Weights are the next knob it has access to, coming soon. Sign up at withbeaker.ai Blog: withbeaker.ai/blog/weight-sp… Tuned-weights on @huggingface: huggingface.co/collections/r… Public runs: app.withbeaker.ai/public/run…
17
23
5
140
12,975
The harness really is the biggest lever right now - tuning an older model can literally beat the SOTA We’ve been working hard with Victor to turn these insights into Beaker Exciting to finally have it in preview :) app.withbeaker.ai/
Stop waiting for the next model I often hear ai engineers at vertical ai app cos say they need the next model releases to unlock the next step-change in their product capability. What if your current model was already good enough? We tuned an agent's harness around a model 9 months older than the latest one. It beats the latest model by almost 10 points.
2
6
430
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2,366
7,257
5,753
55,944
10,075,761
thomas retweeted
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
🤖 Made with AI
2,030
4,552
2,814
40,581
5,775,572
thomas retweeted
Replying to @OpenAIDevs @OpenAI
Now add the shaky cam
6
5
2
711
14,215
Dans un monde où il faut pas 37 publi first author neurips-icml-iclr pr imaginer trouver un poste j'aurais pê suivi la recommandation
« Plus vous mettez de l’effort à surmonter le problème, plus vous êtes heureux » : Cédric Villani appelle les doctorants à ne pas utiliser l’IA générative ➡️ l.leparisien.fr/kzwt
13
44
2
607
39,430
We built high-throughput materials labs in Menlo Park to create a loop between experiments and models. The labs generate fresh data, the models learn from it, and then help us decide what to try next. Using only 1,300 H200s, plus months of our experimental data, we mid-trained and RL’d an open-source model to surpass GPT-6 Astra on our analysis benchmark. We call it Neon. This is real footage from our lab. We’re focusing first on hard problems in materials science, including superconductors, magnets, and semiconductor materials. Read our blog posts below.
274
528
289
5,093
1,675,388
thomas retweeted
Replying to @thsottiaux
The 5 hour limit is holding us back Tibo
31
14
2,052
75,588