Search: “llm”

// 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

Today we release Contrastive Neuron Attribution (CNA), a method for steering LLM behavior by identifying and ablating sparse circuits in the MLP basis without training a sparse autoencoder, modifying

This post explores conceptual parallels between classical distributed systems design patterns—such as actor models and reactive blackboard architectures—and their potential application to LLM agent de

1/3 My time at Meta FAIR will soon come to a close. I joined nearly two years ago full-time to help advance LLM reasoning. It has been a remarkable journey working with and leading an exceptionally ta

Gary Marcus expresses optimism about the potential of LLM-powered AI agents in a post on X (formerly Twitter). The post advocates for confidence in these AI systems, though without access to the full

New article: a visual tour of recent LLM architecture advances, from Gemma 4 to DeepSeek V4. I focus on long-context efficiency tweaks like KV sharing, per-layer embeddings, layer-wise attention budge

A breakdown of a ComfyUI workflow that combines Seedance 2.0 with an LLM prompt setup designed for cinematic motion shots like this. This example recreates the viral floating hot sauce effect with a M

@hwchase17 Started on this and finding it awesome; also LangSmith engine sparked an idea. The 'Dependabot like for LLM agent failures'. LangSmith Engine gives you the smoke detector. The natural next

The Second Scaling Law remains undefeated. If you want better hacking (or math, or science, or crossword puzzle solving) out of an LLM, just add thinking tokens. There doesn't seem to be any plateau s

Three new open-source models just landed in ComfyUI natively: → Gemma 4 (Google DeepMind) - multimodal LLM handling text, image, audio, and video input with built-in step-by-step reasoning mode → VOID

almost all LLM “thought” is derivative. below just sounds like hype to me 📁 Jensen Huang, CEO of NVIDIA, says GPT was never just about generating images or text but about generating thought itself. Th

Introducing LangSmith LLM Gateway: The runtime governance layer for your agents. 💸 Enforce cost limits 🔒 Detect PII ✅ Act on violations …All without leaving LangSmith. Now in Private Beta https://www.

Neural networks do math by rotating shapes. We found a shape-rotating calculator hidden inside an LLM – and it’s used for more than just math! (1/6) Media Neural networks might speak English, but they

Training a diffusion LLM from scratch is challenging, and diffusion inference advantages don't show up at training. So we convert instead. Building on the TiDAR recipe, we take ZAYA1-8B-base and run d

Gary Marcus expresses frustration about the increasing prevalence of bot-generated and LLM-written responses on social media platforms, noting that filtering through such content has become increasing

A little talk on what we can learn from implementing LLM architectures from scratch in Python and PyTorch. And how I approach new open-weight models, compare them against reference implementations etc

Do multi-agent systems make LLM reasoning better? Most AI devs assume that it should. But this new paper shows that this is often not the case. It ran 22,500 deterministic trajectories across GAIA, SW

Today we release Token Superposition Training (TST), a modification to the standard LLM pretraining loop that produces a 2-3× wall-clock speedup at matched FLOPs without changing the model architectur

In KAME, a fast speech model starts replying instantly, while a backend LLM runs in parallel to inject deep knowledge on the fly. It’s a completely different way to approach conversational AI, making

When is the last time a general purpose LLM (putting aside hybrid systems like Claude Code with special purpose symbolic harnesses) last completely blew away all competing prior models? GPT-4 relative

Am old enough to remember when @GeoffreyHinton told me I was stupid for saying that LLMs regurgitate training data. He was wrong. LLM regurgitation is now one of the best-established findings in the f

Simon Willison discovered how to use an LLM command-line interface tool in Unix shebang lines, enabling the creation of executable scripts written in English or natural language. This technique allows

// The Memory Curse in LLM Agents // (bookmark it) Long histories apparently degrades agents as they become increasingly history-following and risk-minimizing. Across 7 LLMs and 4 social dilemma games

Tip of the day: You can use Hermes Agent's Credential Pools to login or add multiple API keys for the same LLM Provider and it will rotate across them to ensure stability of your operations. Learn mor

Ramp Labs is doing a ton of cool work all the way from launching products to hardcore research on LLM internals, this is a fun one 🔥 Talked to @ramplabs Head of Applied Research Alex Shevchenko on the

We have greatly expanded the surface area for plugins over the last few weeks, and now you can extend LLM Inference Providers and Gateway Channels with plugins. Want a custom implementation of your fa

Wiki-Builder, a tool for creating LLM-powered wikis, was featured on Hacker News' front page, announced by DAIR.AI. The post indicates upcoming updates and improvements to make wiki creation more acce

If you need proof of why AI output can't be trusted by default watch that page for more than 1 minute for any given LLM that produces output in there. An amazing time capsule of the unevenness of curr

5 months later, I would say that AI is definitely on track to produce 90% of all bullshit within a few years. Congratulations LLM Industry, you are changing the world! Claim AI will generate 90% of kn

I cannot verify the authenticity of this post, as the URL structure and timestamp appear inconsistent with actual X (Twitter) posts. The claim about fine-tuning a '1930 vintage LLM' is anachronistic,

Cool paper from Meta FAIR. It's on self-improving LLMs but on the pretraining side. (bookmark it) Most LLM safety, factuality, and reasoning fixes get bolted on at post-training. By then, the patterns

Our CEO @jerryjliu0 in @VentureBeat , on what's actually changing in the LLM stack: 'We've really identified that there's a core set of data that has been locked up in all these file format containers

Congrats to @poolsideai on Laguna XS.2 release. Day 0 support in @vllm_project. FP8, INT4, and NVFP4 quantized checkpoints created with LLM Compressor also ready today. 33B params, 3B active. Apache 2

The new LLM trained only on pre-1931 text is small enough that it can potentially run on device, so, with the right tools, you can get a fully vintage version of Siri, but from the era of Downton Abbe

Kimi K2.6 has achieved the top ranking on OpenRouter's weekly LLM leaderboard, marking a significant achievement for the model. The announcement expresses gratitude to developers using Kimi and indica

ml-intern now notifies you when it needs input! Amazing for those 12h+ autonomous LLM training runs The ML Intern can now ping you on Slack when it's finished training models, generating datasets, or

New Frontier LLM Orchestrator Model @SakanaAILabs The fugu model series provides a novel paradigm for test-time scaling. The models are trained to call LLMs and infer not only which individual model i

These pelicans are kind of angry looking! Left is deepseek-v4-flash, right is deepseek-v4-pro - both generated using OpenRouter via my LLM tool 🚀 DeepSeek-V4 Preview is officially live & open-sourced!

Hugging Face Releases ml-intern: An Open-Source AI Agent that Automates the LLM Post-Training Workflow [The 'AI Intern' that actually ships SOTA models ] This isn't just another ML Research Loop wrapp

OpenAI just open sourced a new 1.5B (50m active) model on HuggingFace with Apache 2.0 license! It's not a new LLM, this one is called Privacy Filter, and it's a PII detection model (checking if text h

This wasn't the case with previous image generators, but the LLM you select has a huge effect on GPT-imagegen-2 output. GPT-5.4 Thinking and GPT-5.4 Pro will produce much better images, especially for

Why do no 'AI safety' people ever say how LLM AIs failed to live up to the hype? Why do the safety people never share the failures of LLMs? This link is from Nov 2023 & shows how tech CEOs used 'safet

Kimi K2.6, 1T params runs on my desktop GPU workstation. 192GB VRAM + 475GB RAM Serious contender to be the best open coding LLM on Earth. We'll see I'm checking their default harness. Goes on and on

LangChain Community Spotlight: Saving $1M in LLM Costs 💰 Gustaf, an AI Engineer, shows how he reduced a production RAG chatbot's costs by 90% and improved latency by 82% through caching, intelligent r

@opentraces 0.3 is out! Big update: a stronger security pipeline with TruffleHog and LLM review, generation-aware ingest for living traces, line-level attribution + inverse blame with a @gitbutler-sty

This post likely discusses Sam Altman's observation of a large language model (LLM) demonstrating GUI operation at human-comparable speeds, highlighting a notable advancement in AI capability for inte