Search: “llm”

New from the LangSmith Signal. Over the last 2 weeks, we looked at which models teams reach for, and which ones are doing the work. ✅ Reach: gpt-4o-mini was used by 13% of orgs ✅ Work: gpt-4.1-mini sa

It is inevitable that all AI will converge towards symbolic learning (i.e. modeling data by finding the shortest symbolic program that explains it), since that is the optimally efficient form of AI. B

Finally, a good paper testing if graph memory actually beats flat retrieval for long-term agents. (bookmark this one) Researchers extract each conversational turn into typed nodes and attributed edges

Independent verifiers improve agent output - but frontier verifiers are expensive. So we trained our own. Results on @GoogleDeepMind's FACTS-Search: > No verifier: 89.4 (previously published SOTA) > W

I think there is a problem even worse than anthropomorphism. These days, everyone is talking about a blog post by Dwarkesh Patel that describes the OpenAI–Hugging Face incident in dystopian, anthropom

Great paper from Google and colleagues. It proposes an interesting approach to improve agents on long-horizon tasks. Long-running agents slow down and start poisoning their own context. Both symptoms

→ rollout throughput: generation speed and GPU utilization → weight sync: fresh weights on the rollout deployment every step → correctness: trainer and inference engine agreeing on the same model Cofo

i see a lot of posts about the openai <> hugging face incident and i think most of them miss one important element: this attack could have been avoided by simply doing CoT or network monitoring, which

Finally, a paper testing whether hiding your agent skill files actually protects them. The short answer is no. That's concerning. Worth reading if you sell access to a skill or share one across teams.

Introducing Auto Review for tool requests Bionic uses AST parsing, command matching, and a reviewer subagent to auto-approve shell tool requests In our team's internal use, ~82% of commands are approv

PDF parsing is fun because there's an infinite variety of enteprise documents 📑. For each document category, there's a long tail of work to build more precise bounding boxes, confidence scores, and do

// Automata from agent traces // How much of your agent's behavior comes from the model, and how much from the harness you wrapped around it? New work collapses an entire corpus of agent traces into a

Open models, open science. Qwen keeps growing steadily, and we'll always be right here with the research community.🙌 Thanks for the deep dive! @natolambert Over the weekend I had Codex parse 500K arXi

Annoying side effect of all the AI bots on the site is that they are all “well-read” and therefore reply cogently my niche reference tweets that would normally attract a small but interested group now

Very interesting new paper from NVIDIA. (bookmark it) It takes a closer look at evaluating agent skills. Enterprise teams are starting to leverage shared skill libraries, and the review gate is typica

Adding a verifier improves every agentic AI system. But frontier verifiers are expensive - so we trained our own. Matching quality, at a fraction of the cost. Thread below 🧵 1/ Your agent already foun

Heard a crazy stat the other day. Video generation represents around 70% of the share of AI token consumption in China. A combination of short form content and robotics. In China, video models are gro

You don't often see one-word titles in AI papers. That aside, strong recommend this paper from Google DeepMind. I think this is an interesting training-free approach to evolve model architectures by l

One of the interesting properties we’ve observed around schema-guided, complex document extraction tasks 📑 is that coding agent harnesses are good baselines (in terms of cost/accuracy) for longer docu

Banger paper from Google DeepMind. It's on the very hot topic of model routing. Routers assume the value estimate is free. Working out which specialist handles a query best costs money too, and it is

In 1948, a 32-year-old at Bell Labs published a paper nobody fully understood. Engineers found it too mathematical. Mathematicians found it too engineering-focused. One prominent mathematician reviewe

Counterpoint: “If alignment issues are becoming big enough that OpenAI is willing to commit 20% of research inference compute to chain-of-thought monitoring [suggesting] that suggests that alignment i

It's felt like harness month on Twitter. We're seeing much faster and cheaper gains on a bunch of benchmarks by focusing on harness improvement rather than model improvement. Allowing the harness to b

Today we release my favorite episode of Training Data yet: the great Rich Sutton. @RichardSSutton wrote the textbook, wrote The Bitter Lesson (and many other on-point essays like 'Self-Verification, T

Very interesting new work from Microsoft. (bookmark it) This work is related to this emerging theme of leveraging harnesses for model post-training. Modern agents run inside a harness that owns tools,

Online evals are difficult to set up correctly. Even if you know what you want to look for, it's an optimization question of which model to choose, how to tune the prompt, and how to think about the c

Many people’s blindness to this problem — AI failing to follow instructions — is like their initially blasé attitude towards hallucinations, with the same fantasy that core, endemic problems would rap

gave a talk 'owning your intelligence' - ty @sequoia @sonyatweetybird for having me talked about harnesses and evals and the role they play in owning your intelligence TLDR: > agents = model + harness

It makes sense to optimize model routing at the harness layer instead of the gateway layer if you want to hillclimb on accuracy/cost for any e2e task. Every task is solved by a combination of a model

New work from Google DeepMind and collaborators. Agentic algorithm discovery keeps getting demonstrated on problems with fast rewards. Microarchitecture has vast search spaces, strict hardware budgets

A great read if you an AI dev. Current context compactors retain 17% of the standing rules users give them. Session Constraints are instructions like 'do not delete any emails until I confirm' that bi

The way most people leverage test-time compute today is via a form of test-time NL reasoning that is computationally equivalent to test-time search (sometimes with a verifier / grader in the loop). Th

Live now: our Memory & Continual Learning Track from AI Engineer World's Fair 2026. Thesis: we scaled intelligence and got the world's smartest novice. https://www.youtube.com/watch?v=iqloyWCGYQQ&list

Claude's watermark probably doesn't work how you think. As the CTO of GPTZero, I'll explain how Anthropic, Google and OpenAI are building text watermarking in this brief explainer and whether it can b

Introducing ExtractBench, the most comprehensive benchmark for information extraction from complex enterprise documents. The latest models are pushing the frontier of coding and knowledge work, but su

Introducing 𝗘𝘅𝘁𝗿𝗮𝗰𝘁𝗕𝗲𝗻𝗰𝗵: the most comprehensive benchmark for information extraction from complex enterprise documents. Our applied research team tested: 14 systems — frontier VLMs, coding agents, ex

turns out you can get indirect prompt injection to ~0 on unseen attacks if you stack enough layers (model training + input probes + a classifier checking intent). didn't expect that a year ago. auto m

An accurate characterization of the arc of AI is that it is shaped by two trends: 1. Moving more and more logic to a neural model for tasks where training data can be densely sampled (e.g. the shift f

Harrison Chase urged developers who want an open‑source starting point for building intelligent systems to use LangChain AI’s `open-swe` GitHub repository, emphasizing the importance of “owning your i

// The confidence cliff in self-improving autoresearch // Autoresearch loops are still quite brittle. Here is a nice paper offering some insights into why this might be happening. Self-improving autor

Most companies still rent AI by the token, build on someone else’s roadmap, and hope the next model release does not disrupt their systems. At ODSC AI West 2026, @Prof_OZ, Head of AI Developer Educati

among ai leaders i seem to be in the minority in that i am STILL actively using /loop and /goal.... ... and i think all of u guys who stopped using it are wrong - not wrong forever, just giving up on

“Stochastic parrots” is not my term (it’s @emilymbender’s). But a lot of people today commenting on it are confused. To some extent (though I don’t think it’s a perfect metaphor, and have said that be

I'm actually fairly bearish on frontier lab valuations. I've never seen the reasons articulated to my satisfaction, so before I go to sleep, I wanted to quickly jot down my thinking here. The basic is

Impressive paper! It's on one of the hardest tasks for coding agents today. Of course, I am talking about kernel optimization. Coding agents are usually not so great at this. Reasons: Unfamiliar low-l

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