Knowledge catalogue
Great report from Google and MIT. Co-designing ML models and the accelerators is an exciting research direction. This report introduces Coco, an agent platform TPU architects use for hardware and mode
With llama.cpp using a customized SYCL backend on Intel B70 (32GB), overclocking my DDR5 memory gave modest gains for MoE models which do not fit in VRAM. Both PP and TG increased after overclocking D
arXiv:2610.03894v1 Announce Type: cross Abstract: A deployed LLM agent emits tool calls, queries, and code that can be silently wrong -- by the time the error surfaces, the action has run. Frontier ch
arXiv:2610.06324v1 Announce Type: new Abstract: CLIP-like vision-language models remain a cornerstone of multimodal systems, yet their scores stay near chance on directed spatial relations, such as wh
Really nice paper on harness-aware distillation for small language model agents. (bookmark it) It's a super interesting distillation framework for small language model agents that are deployed with a
arXiv:2610.06243v1 Announce Type: new Abstract: Parameter-efficient fine-tuning (PEFT) reduces the cost of adapting foundation models by focusing training on a small parameter subset. Complementary to
arXiv:2610.05453v1 Announce Type: cross Abstract: Multimodal models are increasingly shifting toward unified architectures that understand and generate text, images, and other modalities within a shar
Banger paper from Microsoft and colleagues. If you run agents that search a large document collection, this one is worth your time. (bookmark it) They introduce CorpusMap, which resolves recurring ent
arXiv:2610.02376v1 Announce Type: cross Abstract: Co-designing ML models and the accelerators that run them is an unusual reasoning task: architects must draw confident, high-stakes conclusions about
Does telling your agent to plan ahead actually help? It's standard practice to add a 'plan ahead' or 'think about the other players' instruction to an agent's prompt. In auctions and matching markets,
Great overview of context compression in LLM Agents And one interesting, unexpected finding. If your compaction policy is tuned to cut tokens, it may be making your agent run slower. Great study from
Huge release from EverMind AI. Raven is an open-source multi-agent system that builds a separate harness for each model and domain, evolves those harnesses from their failures, and coordinates them th
Learn to optimize your own harness, folks. You can squeeze much more performance and value from a harness. This work claims that a self-modified harness matched Codex for 4.03. Specifically, a coding
arXiv:2610.02413v1 Announce Type: new Abstract: LLM agents increasingly rely on activation probes as runtime monitors for prompt injection, jailbreaks, and unsafe requests, reading the model's own hid
On my two DGX Sparks, Swift 1.5 (UkisAI's reasoning-efficient fine-tune of Qwen3.8-Flash-Next) running on veloGB10 (https://github.com/sf-stav/veloGB10, sf-stav's Rust/CUDA engine built only for GB10)
We turned Claude Code, Codex, Hermes, Pi, @opencode and other coding harnesses into RL environments. No changes to the harnesses, no changes to the training code. Any open model, any task set, fully o
arXiv:2610.03036v1 Announce Type: cross Abstract: We present WebFovea, a vision-based web agent that placed 2nd in the WebRetriever Challenge 2026 with a final score of 57.0 out of 100. The challenge
50K likes on this!? Most of what Karpthy has shared is things we have been doing for over a year. No surprises, as many replies are highlighting. But I still think this raises an interesting discussio
Banger paper from Google. It's standard practice to sample several agent rollouts and trust the answers they agree on. This Google paper shows that agreement can hide shared errors, while disagreement
For the last few months, I ran DeepSeek v4.0 flash (NVFP4). First 0731, then visionexp because it was a free improvement. I got around 65 tps decode and almost 2k prefill, and ran 4-5 agents in parall
First of all, thank you for reading. I trained a small MoE model completely from scratch (no external base weights) and wanted to share the results + a couple of lessons. Apex-2 - Architecture: Decode
Banger paper from CMU on harness learning. (bookmark it) Also, pay attention to this important new AI engineering skill of improving agents by editing their harness code instead of their weights. Seei
First AutoHarness, then AutoContext, now AutoCompact. I am seeing a rising trend of work that trains models to natively support more of what the harness does. This work specifically trains agents to d
Reupload: I'm adding this again, since previous post had a totally stupid title which made people assume this app is some kind of cloud/proprietary models UI - this is totally not the case here. In ve
We built an engine, Kyojin, on top of ExLlamaV3 for Strix Halo (gfx1151, ROCm), and packed two 300B-class MoE models so each fits one 128 GB machine. First release, all measured on Ryzen AI Max+ 395.
Banger paper from Google Research on multi-agent proof discovery. (bookmark it) It's really interesting to see this emerging multi-agent pattern: not enforcing too much execution structure and pairing
Somebody just released a fully open-source, open-weights decision model that beats Jev! Just kidding, it's me and I this is my first time releasing a public model, recipe and dataset so I am looking f
arXiv:2610.01382v1 Announce Type: new Abstract: We introduce Gacha Decoding, an inference-time method for eliciting diverse language model generations that scales with model capability. Across open-en
Enterprise AI agents need persistent memory to execute complex, multi-day workflows and long-horizon tasks. In this blog, we examine how a 2-tier memory architecture using Memorystore for Valkey for s
arXiv:2610.01296v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) Diffusion Language Models (DLMs) offer flexible parallel decoding and increased model capacity, but their large number of exper
arXiv:2610.00074v1 Announce Type: new Abstract: K-Dense BYOK (bring your own keys) is a free, open-source AI research assistant for scientists in any field that runs on the researcher's own computer.
Meta now lets you make your own Muse gadgets that feature the company's new AI agent with code that the company open sourced. The company suggests projects like loading Muse on a color E Ink display t
More agents don't mean higher performance. There is a coordination bottleneck to consider. Not to mention the unnecessary costs. So how many of a multi-agent team's actions actually help it finish the
Row-Bot 5.0 is a major rebuild of the local-first AI assistant. The NiceGUI interface is gone, replaced by one React app across desktop, browsers, phones and tablets. The biggest change is how work is
arXiv:2610.00118v1 Announce Type: new Abstract: Customer churn prediction on the IBM Telco Customer Churn benchmark (n = 7,043) routinely reports test accuracies above 95%, with the most cited publish
Something keeps happening in frontier AI, and the explanation we are given is always one size larger than the event. An OpenAI research model, running in a supposedly offline training environment, hit
The same model, with the same weights, scores 62% in one agent harness and 33% in another. @adithya_s_k and the @huggingface team just released the ultimate guide to multi-harness RL, and it's one of
I have been building a somewhat unusual local inference machine around two Huawei Atlas 300I Duo cards. They are relatively inexpensive, passive, dual-accelerator PCIe cards with 96 GB of device memor
Hey guys, Looking for an alternative for Claude Cowork: Project Work / Documents Integrations like Notion, Gmail, etc. Tools like Websearch, PDF creation, etc. Came over Eigent (https://github.com/eig
I've run the benchmark on a RTX 3090, 1024 tokens in / 256 out, concurrency 1–32. If the model fits on vRAM (Gemma-4-26B-A4B, byte-identical GGUF on both engines): llama.cpp has 2.2–3.2× the throughpu
A few days ago I released Jeff-Qwen3.5-0.8B, a small 'System 1' model that picks between options you define and returns a calibrated probability for each, in one forward pass. Speed was great on my M4
I put an old RTX 3090 to work as a local coding agent with Qwen3.8-27B, llama.cpp, and OpenCode. Here are the setup and results, including what failed. This is a summary of my own blog post, linked be
Last update: https://www.reddit.com/r/LocalLLaMA/comments/1wu9ksu/update_yandexaliceai_80ba3b_fine_tune_progress/ - basically, an instruct fine tune on the base model using a synthetic distilled data
Meta launched a new Muse AI agent it claims can help you with everything from firing off emails to buying stuff online. Muse can be surprisingly effective at delivering on those promises — if you’re w
cli: exit on stdin EOF and drop the console wide Ctrl+C broadcast (#29722) cli: exit on stdin EOF and drop the console wide Ctrl+C broadcast On Windows the simple input reader sends CTRL_C_EVENT to ev
Today, Google Cloud Data Agent Kit is generally available. Data Agent Kit is a free set of Model Context Protocol (MCP) tools and agent skills that lets the coding agent you already use work directly
arXiv:2609.36734v1 Announce Type: new Abstract: Knowledge Distillation (KD) trains a smaller-capacity student model to imitate a larger-capacity teacher model by matching output distributions, implici
arXiv:2609.36173v1 Announce Type: cross Abstract: Parallel speculative drafting generates multiple candidates in one backbone pass, but independent token selection can produce inconsistent continuatio
I run local models through Ollama and kept hitting the same quiet problems, so I wrote a scanner for them. llm-doctor checks your model stores (Ollama, LM Studio, Hugging Face, MLX) for duplicate weig
arXiv:2609.33142v2 Announce Type: replace Abstract: Generating a correct answer does not mean that a language model will select it. We separate factual recall into three steps: generating a correct ca
OpenAI launched Dots at DevDay. My thoughts. First, the good. Always-on is really really really what the market needed. It's why I find myself reaching for Instinct so often. I don’t have to think abo
I have a long list of my own domain specific eval questions that I run to validate which models I can rely on: coding, coding (numpy/pandas), data analytics decision making, local RAG, and voice assis
While AI has made plenty of inroads on people's phones and computers, it's largely failed in dedicated devices. But over the next year, two major AI companies, Meta and OpenAI, will attempt to change
Written by: Robin Grunewald, Supriya Mazumdar, Kelli Vanderlee Introduction Google Threat Intelligence Group (GTIG) examines vulnerability disclosure and exploitation statistics to evaluate the impact
I wanted a fun way to visualize and stress-test a multi-agent ecosystem, so I pointed it at my favorite MMO (a vanilla-era private server emulator). Live view: https://design.kristiantalley.com/projec
arXiv:2609.35728v1 Announce Type: new Abstract: We present FlowAct-R2, a framework for interactive humanoid video generation that combines continuous multimodal control with proactive agent planning.
Great paper from Amazon AGI on generating RL environments for agents. (bookmark it) Also, pay attention to this important new AI engineering skill of creating RL environments. Seeing a huge shift towa
First 3 disposition results. Hi there! My name is Jerry and I recently built Enclosure, e deterministic environment for testing LLMs. It's a simulated office enfironment https://github.com/openconstru
Banger paper from Microsoft Research. (bookmark it) They run 1K+ coding agents at once to test a scalable self-organized multi-agent harness. This is an interesting test because most multi-agent syste
xAI's Grok 4.7 is now available on Amazon Bedrock: a frontier model for coding, long-running agents, and knowledge work. It offers a 500K token context window and four configurable reasoning effort le