Search: “dair-ai--x”

Tool Attention Is All You Need // Tool Attention Is All You Need // New research proposes a practical fix for the hidden 'MCP tax.' The work introduces a dynamic tool gating mechanism built on an Inte

The Top AI Papers of the Week (April 19 - 26) - Skill-RAG - DeepSeek V4 - Autogenesis - Attention to Mamba - Stateless Decision Memory - Self-Evolving Logic Synthesis - Self-Generated World Knowledge

'AI should elevate your thinking, not replace it.' I don't disagree, but the issue is that current LLMs are not really trained to support that out of the box. I've solved this by building my own agent

// Skill Retrieval Augmentation for Agentic AI // Great read for AI devs. (bookmark it) It's on finding efficient ways to incorporate skills for agents. The work introduces Skill Retrieval Augmentatio

// When to Retrieve During Reasoning // Pay attention to this one, AI devs. (bookmark it) Most RAG systems retrieve once, before the model starts reasoning. Large reasoning models like o1 and R1 don't

// Contextual Agentic Memory is a Memo, not True Memory // Most agent memory today isn't memory. They are more like memos. A new paper argues that vector stores, RAG buffers, and scratchpads implement

// Memory as a Model // The paper augments any LLM with a separate trained memory model that stores, retrieves, and integrates facts on its behalf. It decouples memory updates from base-model weight u

Very cool idea to convert memory to skills. (bookmark it) Most agent memory systems retrieve past traces as passive context. MSCE turns them into executable skills instead. The training-free framework

Harness choice is a big deal. So much room to advance and improve results across the board with agent harnesses. Great paper highlighting this. New research releases DataSpace, a benchmark where data

New research from Meta. Agent harnesses are still mostly authored by hand. This makes it hard to tune robust agent harnesses for long-horizon tasks. In this new work, agents learn harness policies off

Very cool idea to have agents design complex systems by searching over the model structure itself. New research from Sakana AI introduces CEDAR, which uses LLM agents to write, simulate, and refine sy

// The Bitter Lesson of Tool Calling // Tool calling is a design choice, and the defaults are quietly costing accuracy. How so? New research releases a generation-spanning comparison of programmatic t

Why does CLAUDE.MD keep growing? If you maintain a CLAUDE.md or an AGENTS.md, this one is worth your time. (bookmark it) This work traces why these files grow without bound. Appending an instruction i

Very interesting new work from Anthropic. (bookmark it) They evolve mind viruses, ideas that spread through a multi-agent system by getting each host to pass them on, then measure what governs the spr

Four small architecture decisions can cost up to 47% of a model's long-context performance. New research from Ai2, Carnegie Mellon, and the University of Washington isolates them. Normalization, GQA,

// Actions Speak Louder Than Words // Multilingual agent evaluation compares final answers and throws the trajectory away. The trajectory fixes cost, latency, failure mode, and auditability. New resea

// Reusable Context Engineering // Context bloat quietly kills long-horizon runs, but you can fix it from the outside without fine-tuning the underlying agent. (bookmark this) Context management is us

Another banger open-source release. Miso One is an 8B text-to-speech model with real emotional range, so voiceovers carry warmth, hesitation, and excitement instead of sounding flat. It's purpose-buil

NEW AI paper worth bookmarking. This is something I called early, and this paper confirms it: verification has emerged as a new important scaling axis. Here is the simple explainer and what this paper

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

Never too late for any company to get in on open models. Inkling is a MoE transformer with 975B (41B active). Native multimodal support. Context lengths: 64K and 256K. Will be testing and providing th

Kimi K3 is now available for inference & training in @FireworksAI_HQ. Crazy how easy they make it to tune frontier open models like K3 using LoRA adapters. Best time to figure out how to own your inte

Building agents that patch other agents. This is an interesting approach for self-improving agents that leverages agent outputs. If you run agents in production, you already have the training data for

New research from Meta. Agent harnesses are still mostly authored by hand. This makes it hard to tune robust agent harnesses for long-horizon tasks. In this new work, agents learn harness policies off

Huge if true! We are talking about a 27B multimodal model that runs locally on a phone. That's wild! Bonsai 27B reaches up to 163 tok/s in 1-bit and 134 tok/s in Ternary on an NVIDIA GeForce RTX 5090.

Never too late for any company to get in on open models. Inkling is a MoE transformer with 975B (41B active). Native multimodal support. Context lengths: 64K and 256K. Will be testing and providing th

// Global Workspace in LLMs // arXiv paper for the popular J-space work from Anthropic. (bookmark it) The short recap: If you build on chain-of-thought or steering vectors, this work provides a mechan

This is a neat feature. I wrote an article a few weeks back about how I built this into my agent orchestrator: https://x.com/omarsar0/status/2073404610501329247?s=20 But I made it multimodal from the

Not because Andrej is saying it but I think voice is goated. And you can mix it with other modalities for even richer prompting. I recorded a session a few weeks back to demo the power of multimodal p

Very cool paper from Microsoft. The idea is to train agents on replayed teacher trajectories instead of live environment rollouts. On-policy distillation for agentic tasks is expensive because every u

New research from NVIDIA. They just dropped a PyTorch-native training framework for agentic RL. (bookmark it) Paper summary: Molt is a PyTorch-native agentic RL framework with an unusual design target

Harness choice is a big deal. So much room to advance and improve results across the board with agent harnesses. Great paper highlighting this. New research releases DataSpace, a benchmark where data

Great writeup from the University of Oxford. It's a taxonomy of LLM-based agent limitations. Good read for anyone shipping with agents. Benchmark scores keep climbing, yet the same agent failures resu

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

Very cool paper from Microsoft. The idea is to train agents on replayed teacher trajectories instead of live environment rollouts. On-policy distillation for agentic tasks is expensive because every u

New research from NVIDIA. They just dropped a PyTorch-native training framework for agentic RL. (bookmark it) Paper summary: Molt is a PyTorch-native agentic RL framework with an unusual design target

Great technical paper from Harvard and MIT. It's on role drift in compound LLM systems. (bookmark it) End-to-end RL improves the accuracy of a multi-module LLM pipeline without constraining how the mo

// The agent is its own best speculator // Agents spend a large share of wall-clock time waiting on tool results. Speculation hides that latency by predicting and pre-executing the next call, but exte

New research from Meta. Agent harnesses are still mostly authored by hand. This makes it hard to tune robust agent harnesses for long-horizon tasks. In this new work, agents learn harness policies off

// Actions Speak Louder Than Words // Multilingual agent evaluation compares final answers and throws the trajectory away. The trajectory fixes cost, latency, failure mode, and auditability. New resea