Search: “papers”

Super interesting new work from NVIDIA. (bookmark it) They suggest building agents as Python objects. Very cool idea and I think it could a lot with agent reliability. More below: Agent development to

The root cause: request-level engines never see that a series of LLM calls belongs to one longer workflow. ThunderAgent adds that missing view. It treats each agent workflow as a schedulable program,

There's been a talk about how LLMs are only advancing in verifiable areas like math or coding, but that isn't what the data suggests. As models have gotten better at that, they are also better at solv

btw anthropic's internal document on this literally said 'we don't want it to be known that we are working on this.” it was called project panama. here's exactly what happened: 1: anthropic concluded

Nice little insights on doing autoresearch with coding agents. Hand a coding agent a dataset, an eval script, one editable file, and no supervision. That's autoresearch and it tries to optimize the nu

New research from NVIDIA. Does AdamW have a scale ceiling? This work claims yes, and shows where it sits. At batch sizes up to 100M tokens for next-token prediction, SOAP and Muon maintain training st

// Agentic Context Management // Great read for the weekend. (bookmark it) Production agents fail less on reasoning and more on what sits in their context. Conversation history, big prompts, huge tool

AI Engineer Paris is back! After an incredible first edition, we’re excited to announce AI Engineer Paris 2026, this time hosted by our friends at @MistralAI. 🎤 CFP is open. Submit your talk by July 3

New research with Microsoft's and colleagues on training agents inside the harnesses they actually run in. (bookmark it) Why it matters: Agents today live inside elaborate harnesses like Claude Code,

Open weights = freedom. You can run them on your own hardware. No vendor can pull the plug. No API can deprecate you. No company logs your private data. That's sovereignty. Closed models hand one comp

one thing i think people dont appreciate enough about @poolsideai is their unusual degree of openness — not only have they shipped an excellent Small model that somehow beat @thinkymachines at coding,

New research from Meta. (bookmark it) Most factuality work checks whether the claims in an answer are correct. GAMUT goes after the harder question of whether the answer covers everything it should. I

Progressive disclosure in agents doesn't scale. And its benefits seems agent harness dependent. (bookmark this one) Finally there is a proper study on using agent skills and the effect of progressive

Domesticated, not feral: Why evolvable AI is not yet a Darwinian threat Writing in PNAS (Proceedings of the National Academy of Sciences), three heavyweights in AI and evolutionary biology criticized

Interesting finding on frontier models. It turns out that frontier models can write proofs but stumble on faithfully copying a long block of text. This has huge implications. It sounds trivial, which

Highly-recommended overview of metacognition in LLMs. (bookmark it) Interesting behaviors in LLMs like confidence calibration, self-verification, knowing when to stop, and knowing what you do not know

I want to have this in writing so I can refer back to this tweet before it inevitably becomes a consensus view on twitter and say 'I told you so!' - The mass majority of researchers and academics in d

New research from Google DeepMind on effective model routing. LLM routers get judged on accuracy and cost. Both can look great while the router is meaningless. If every model in your society responds

Parse any PDF into agent-ready context. Fully local, free desktop app. No model / internet connection required! Ever wanted to quickly turn a PDF into clean text to paste into your favorite AI agent,

Standard RL benchmarks are episodic and stationary, so they don't capture the the characteristics of real-world deployment. Morpheus is a new benchmark for continual learning that provides persistent

Another big reason to use combination of frontier models. Chain-of-thought monitoring is treated as a reliable safety layer for agents. This DeepMind-affiliated study shows the layer can be argued out

VLMは人間のような創造性を持てるか？ ケネス・スタンレー教授らの『目標という幻想（Why Greatness Cannot Be Planned）』は、明確な目標を設定することが、かえって真に偉大な発見を遠ざけてしまうという逆説を論じた書籍です。その議論の中核にあったのが「PicBreeder」の実験でした。 PicBreeder では、ユーザーが「面白い」と感じた画像を選び、それを少しずつ進化

New research from Meta. (bookmark it) It's on how to fix agents that forget previously made decisions. It's well know that long-horizon agents keep forgetting decisions they already made. Meta researc

One of my first journeys in neural networks started over a decade ago with implementing CPPN-NEAT! Back then, I built a clone of ‘Picbreeder’ not only to study the mechanics of neural nets, but to exp

A Visual Introduction to Information Theory (bookmark it) Information Theory is such an beautiful and powerful subject. In the era of AI, it's worth spending time learning about it. Here is a highly-r

Also this graph that shows what an 'AI-pilled' startup looks like, and how uncommon it is even among startups: the answer when founders are asked 'Without GenAI, how many additional employees would yo

// The Harness Effect // (bookmark it) Now more that ever pay very close attention to the orchestration harness and its effect on costs and performance. This study ran 22 evaluation tasks on six found

this is awesome. he didn’t import @activegraphai but borrowed design principles to improve his systems (and found improvement opportunities) great walkthrough in the linked post (for your agent to ing

'Ghost memory' is a real problem with agents. You might have seen the issue where a long-running agent still confidently repeats a user fact that stopped being true weeks ago? New research names the f

// In-context Retrieval at Million-token Scale // Great study providing better understanding of retrieval at million-token scale. They run the first systematic study of in-context retrieval at the sca

// What MCP, A2A, and ACP cannot express // MCP and A2A solve capability discovery and message passing, then stop right where enterprise deployment begins. New research runs a systematic gap analysis

ActiveGraph makes the agent trace the runtime: Yohei Nakajima's open-source Python runtime treats an append-only event log as the source of truth, enabling replay, forking, and lineage for long-runnin

Great overview of always-on agents. (bookmark it) It's a new 130+ pages survey on always-on agents. Simply put it, always-on agents are systems whose future behavior depends on durable state built up

// HASTE: tiered skills for ML engineering agents // Why do ML engineering agents keep rediscovering the same techniques on every new task? Because each competition is a cold start. New research intro

Come and chat with us in ICML 🥳 Excited to present Temporal Straightening for Latent Planning in the Tuesday morning session #1509. Let’s talk about world models, JEPA and representation learning. Age

Highly-recommended read from MIT on the part of RL with verifiable rewards that everyone keeps hitting. RLVR only optimizes what you can objectively score, so style, structure, and diversity quietly c

We are pleased to present our latest research at #ICML2026, “Bridging Spherical Black-Box Optimizers” https://arxiv.org/abs/2606.25761 When optimizing through simulators, external APIs, or in reinforc

When brainstorming new AI agent uses, I always remind myself what used to be expensive and is now nearly zero… 1) cost of reading everything fell → you can now watch 100% instead of a sample (every ex

// AutoMem // I quite like this idea of metamemory. (bookmark it) This new research from Stanford treats agent's memory management as a trainable skill instead of a fixed module. The model decides wha

Bridgewater just published numbers that should make every frontier lab nervous. The world's largest hedge fund tested Gemini, Claude, and GPT on six document filtering tasks its investors do every day

DeepSeek effect https://x.com/maximelabonne/status/2070867418377818542 Fun surprise: DeepSeek used my open-perfectblend dataset to train their new DSpark drafter Time to promote it again! It's an open

New research from Google. LLMs hallucinate with high confidence, miss their own knowledge boundaries, and misreport uncertainty. Most fixes bolt calibration on from the outside. RLMF turns the model o

1/ On Training in Imagination - Dwarkesh's episode has a segment on dreaming as one of the next training paradigms. The idea is that a model learns mostly inside its own, by imagining what would happe

LeWorld model becomes ADAPTIVE and meets MODEL-PREDICTIVE CONTROL AdaJEPA by Yann LeCun and colleagues performs actions, then checks the predicted latent state versus the observed and adapts at TEST T

Introducing SWE-Together: a multi-turn benchmark built from real user–agent coding sessions. Coding agents are often benchmarked like exam-takers: given the full spec up front, then graded on the fina

The interesting thing about viral narratives is that they don't have to be true to spread. There's now enough evidence pointing to the fact that AI usage does not kill jobs, but that might not matter