nicozumarraga

Join us for our first event on July 19th, a talk about Auto-Research Agents with @matteosaponati, curator of Tufa Talks. We will explore the current state of auto-research and talk about if novel architectures are needed going forward- Spots are limited, sign up link below

@lsteno @PascalBtd wise words, whoever said that

@PascalBtd @nzuma0

definitely not @nzuma0 deep freezing his snus and telling me he’s not addicted

Rome wasn't vibe-coded in a day.

DO NOT USE FABLE 5 FOR AI R&D/CODING!!

Excited to share that members of our lab co-authored 6 papers accepted at #ICML2026, including three Main Track and three Workshop papers 🔥🚀 📄 Accepted papers: ▪️ OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data [Main Track] ▪️ Auditing Emotion-Vector-Steered Political Bias in Open-Weight LLMs [AI4GOOD Workshop] ▪️ Reinforcement Learning of Karma Bidding Strategies [NExT-Game Workshop] ▪️ Cinematic Source Separation with Dialogue-Driven Sidechain Ducking [Workshop on Machine Learning for Audio] ▪️ Reinforcement Learning for Tool-Calling Agents in Fast Healthcare Interoperability Resources (FHIR) [Main Track] ▪️ Preserving Plasticity in Continual Learning via Dynamical Isometry [Main Track] @robertjakob @PatrickLanger20 @gaborhollbeck @f14wn @DerRiehl @atoof_sh @deepqlearning @kev_osull @cs06thegreat @nzuma0 @maxrosenblattl Huge congratulations to everyone involved. We are looking forward to presenting these works, reconnecting with colleagues, and meeting new friends in Seoul 🇰🇷 🔗 Full links in the comments.

Retrofitted acronyms in CS papers are funny. Name a paper "BAMBOOZED: Brainwashing AI Models By Optimizing Overloaded Zero-day Error Distributions" and people will roll with it

@deepqlearning @MALGAI_ICLR2026 nice 🔥🔥

Why do models that score well on standard benchmarks still make bad strategic decisions? In Cattle Trade, accepted at @MALGAI_ICLR2026 , LLMs play a 4-player 50+ turn game of auctions, hidden offers, bluffing, and limited cash. Some LLMs lose to simple Python agents. 🧵

How running 10 claude code instances at 4AM feels like:

Found this commit on my overleaf weirdly inspirational. "JUST NINE PAGES, THINK WHAT WE DO..." 🐉😤

Double Diamond ASL representatives at ICLR this weekend 🇧🇷!

@lsteno @ETH_agent_lab @robertjakob The lab takes whatever it deems necessary from its cells

@ETH_agent_lab @robertjakob chat did my lab steal my tweet x.com/lsteno/status/…

Drop the “Agentic Systems”. Just “The Lab”. It’s cleaner.

@lsteno @grok 🤣. I would like to thank my haters, you are my greatest motivation

@nzuma0 @grok man you’re broke😭 @grok tell this guy to go to sleep and setup his VM tomorrow

Claude is fed up with me and that's fair

@lsteno @grok is he right?

@nzuma0 so when you told me I’d spend a day just setting up you were actually just projecting…

I find it disappointing to see how much progress comes daily from late interaction, DSPy/GEPA, and RLMs but our slow industry only catches up after a lab person slaps a name like “autoresearch” or “deep research” on it lol. The future is already here, just not equally distributed

Congrats to our lab member Rajiv Manichand (@frederik_rajiv) on his AAAI 2026 paper, “Digital Scale: Open-Source On-Device BMI Estimation from Smartphone Camera Images Trained on a Large-Scale Real-World Dataset” 🎉 The paper presents a privacy-preserving way to estimate BMI from smartphone camera images, with the full pipeline running entirely on-device. Thread below 🧵 #AAAI2026 #DigitalHealth #MachineLearning

@lsteno Stop sidequesting, come to the lab.

@nzuma0 the future is no UI, the model is the computer

We are witnessing the death of the "App" era. Demis Hassabis said it: text boxes and static sties will soon feel as archaic, stale. The future isn't a UI you learn; it's a UI that builds itself around your intent in real-time. Fluid. Contextual.

From Demis Hassabis on Lex Fridman round 2 (1:26:17) - "Is it really going to be the current UI that we have today", "I think we will look back on today's interfaces and products as quite archaic"