Knowledge catalogue
Dwarkesh Podcast: Analysis: from 2019 to 2025, gains in pretraining compute efficiency came mostly from data improvements rather than model improvements — Breaking down 6 years of pretraining progress
Another informal nerd benchmark that AIs had struggled with has been passed: Astra successfully designed a Magic the Gathering deck and then used it to defeat a bot on Arena. (the deck may not be stun
Amrith Ramkumar / Wall Street Journal: Anthropic researcher Jacob Coxon says he is quitting the AI industry over fears that tech companies are racing to build systems they won't be able to control — C
John Koetsier / Forbes: Antioch, which creates high-fidelity simulations to reduce the need for hardware validation in physical AI training, raised a $32M Series A led by Greylock — Figure AI pulled t
Found this post:https://www.reddit.com/r/ClaudeAI/s/At9QNEbawK of someone fully developing a game through AI writing code, generating 3D models, etc. (he is using Meshy to create the 3D models). I was
Most of the new ~30B (and now ~10B, thankfully for my GPU) models we see score really high on benchmarks, but I feel like they don't push back on dumb ideas enough. I think most people don't being lik
What is Astra actually supposed to be better at, Because if the answer is “everything,” then I genuinely do not understand what kind of testing was done for writing-heavy use cases. My biggest issue i
OpenAI has fully rolled out the Astra feature to all paid tiers—Plus, Pro, Business, and Enterprise—in both Codex and ChatGPT Work. The company invites users to build with Astra and highlights a live
Wire Amazon Bedrock AgentCore Evaluations into a GitHub Actions pipeline: deploy an AI agent and an OAuth-protected MCP server to AgentCore runtime, invoke the agent with test prompts, score the respo
hexagon: add RELU and LEAKY_RELU ops (#28585) hexagon: add RELU op hexagon: add LEAKY_RELU op too Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/45844223 m
model : support Kimi-K3 recurrent-state rollback (#28466) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/45866131 macOS/iOS: macOS Apple Silicon (arm64) ma
opencl: properly handle non-contiguous inputs to conv2d (#28503) opencl: fix conv2d non-contiguous strides opencl: format Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp
chat : split specialized parsers into common/parsers (#27764) chat : split specialized parsers into common/parsers Move the 14 dedicated template parsers out of chat.cpp into one file each under commo
Fix Vulkan-Hpp handle usage on 32-bit targets. (#22892) On 32-bit platforms, Vulkan non-dispatchable handles such as VkBuffer are represented as uint64_t, and Vulkan-Hpp disables implicit conversions
vulkan : fuse UNARY(GELU|SIGMOID|SILU|SOFTPLUS) + MUL (#27220) vulkan : fuse UNARY(SIGMOID|SILU|SOFTPLUS) + MUL vulkan : fuse UNARY(SIGMOID|SILU|SOFTPLUS) + MUL implement fusion in unary.comp behind U
llama : add missing headers (#28566) fix compile-error: add missing header Bug: #28557 Signed-off-by: Pepper Gray hello@peppergray.xyz fix compile-error: add missing header Bug: #28559 Signed-off-by:
metal : fix idle threads in mul_mv_iq3_xxs for ne00 < 1024 (#28086) metal : fix half-idle simdgroup in kernel_mul_mv_iq3_xxs_f32 for ne00 < 1024 metal : keep N_R0_IQ3_XXS = 4, dispatch a separate 8-ro
server : apply checkpoint min-step eviction only when the checkpoint list is full (#28302) The spacing eviction in create_checkpoint() keeps the oldest checkpoint and erases every later one within che
Revert 'ggml-cuda : restore prop.integrated on HIP builds (#24233)' (#28604) This reverts commit c7d8722. Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/46
llama: disable lazy tensor loading by default on iGPUs (#28326) llama: add lazy mode auto, fix iGPU regression revert changes except disabling lazy load on iGPUs in AUTO Website: https://llama.app Att
Bed time reading 🥱 We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, usin
Benchmark two 30B Mixture-of-Experts models, Qwen3-Coder-30B and NVIDIA Nemotron-3-Nano-30B, across G5, G6, G6e, and G7 GPU instances on Amazon SageMaker AI. Compare throughput, latency, and cost-per-
Bravo to Mistral! So great to see them killing it, they are an inspiration to everyone as we need more sovereign & open-source AI everywhere around the world. Excited to keep supporting them on https:
Can you explain, @SebastienBubeck? On the face of it, this does not look great. This is the third time @SebastienBubeck has claimed math credit for a task OpenAI did not complete: 1. In August 2025 he
ChatGPT Images 2.5 is rolling out today to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. We’re also introducing two new models in the API: GPT-Image-2.5 Flare brings the
ChatGPT Images 2.5—faster, sharper, smarter, with better tools for creating whatever you can dream of. - Faster image generation to keep your ideas flowing - Improved fidelity for more natural, recogn
OpenAI announced ChatGPT Images 2.5 on Tuesday and is adding a new way to tell ChatGPT what you want it to make an image of: by drawing a doodle. With a new feature called Sketch, you can just draw so
Howard Liu / South China Morning Post: China says its AI compute capacity rose 177% YoY to 2,185 eflops by the end of June, and is targeting 9,800 eflops by 2030 via ~$532B in IT infrastructure spend
ComfyUI Tutorial for Beginners: LoRAs, Style Transfer & ControlNets Module 3 is where you stop pulling the lever and start directing. Prompt iteration gets you close. These are the tools that get you
Confirmed by my weekly Hermes analytics. Estimated at roughly 50% cut in weekly and session limits. Which cloud inference plans are you all migrating to? (I understand that local inference is preferre
Congratulations Levent, Tristan, and OpenAI! What a miraculous time to be alive! I view this as the first successful achievement of Recursive Self Improvement or RSI. Models get increasingly better at
hey guys. so I should start with, I'm not s programmer at all. but I'm having a great time letting qwen3.8 27b donit for me. 😃 so far I've just been doing Linux scripts and things for my PC. but my ki
Translation: 'Internal beta testing for an intermediate version of DeepSeek V4.1 Flash is now open; you are welcome to try it out. It adopts a new model architecture featuring native multimodal suppor
Siobhan Roberts / New York Times: Fields medalist Jacob Tsimerman, set to join OpenAI later in September, launches the Mathematical AI Safety Institute to apply higher math to AI safety problems — Jac
Finally a good paper testing whether an LLM's mathematical knowledge has coherent structure or just coherent accuracy. There is a lot of interest in using LLMs for advancing math. This is a great read
I've been making a lot of comments about optimal setup for Strix Halo (gfx1150) and from my observation, 90% of our community is using offcial llama.cpp for it, which is NOT optimized for Strix Halo a
🤯 For the Navier-Stokes Millennium Prize Problem solution, OpenAI says they used a next-generation model that is significantly more capable than GPT-6 Astra. If GPT-6 Astra is this good already, imagi
@GaryMarcus was right. It was the harness, not the model. That’s a much bigger deal (and hidden danger) than the benchmark score. If the real capability lives in opaque, vendor-controlled scaffolding
A few weeks ago, I saw this post where the author trained a SIREN MLP to implicitly memorize Bad Apple as a coordinate function: (t, y, x) to pixel. That got me curious about a slightly different form
Google: Google DeepMind releases AlphaGenome Atlas, a 1PB dataset of predicted molecular effects for all ~9B possible single-letter DNA changes in the human genome — AlphaGenome Atlas is the most comp
Google LLC’s DeepMind research unit said today it has used its AlphaGenome artificial intelligence model to predict the biological consequences of the more than 9 billion possible single-letter change
Threat actors used a multi-agent artificial intelligence framework to compromise thousands of credentials in under six hours, Google LLC’s Google Threat Intelligence Group said in a report released to
Google DeepMind has unveiled an AI tool that its scientists claim could help unravel the mysteries of the human genome and transform our understanding of biology, accelerating scientific research and
Governing models across accounts is the next step after automatic model registration. This post extends managed MLflow and Amazon SageMaker AI Model Registry sync to two cross-account governance topol
Managed MLflow on Amazon SageMaker AI now syncs richer model metadata (training metrics, evaluation results, inference specs, and lineage) into the SageMaker AI Model Registry, with lifecycle stage pr
hope that helps my script uses the GPUs most discussed on the LocalLLaMA, LowEndLocalAI, and LocalLLM subs. At first, I tried to include more, but it became unreadable. Prices were collected by ChatGP
Great to see Stripe’s platform and agent for knowledge work, Kai, get featured on Claire’s show! My favorite use case was asking it questions to slice and dice Stripe’s enormous dataset daily. In fact
Executive Summary Since the release of our May 2026 report detailing adversarial misuse of artificial intelligence (AI), Google Threat Intelligence Group (GTIG) has observed forward leaning adversarie
OpenAI announced a quick-start solution for creators who have an idea but need help turning it into visual content: use ready-made templates for popular formats such as posters or merchandise and then
Hermes now has first-class support in Omarchy, @DHH’s Arch-based agentic Linux distribution. Install the desktop app from the AI menu, or make Hermes your default terminal agent. Your Omarchy theme se
holy shit We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an Open
DiDi built a transparent, self-owned contact center quality assurance (QA) system on Amazon Bedrock, replacing an opaque third-party tool. Intent verification accuracy rose from 38% to 86%, compliance
HPE Zerto built an agentic troubleshooting system powered by Amazon Bedrock that runs on-premises inside the customer environment. This post describes the multi-agent architecture, the on-premises dep
By routing planning, exploration, implementation, and review to different models, I reduced one recurring coding-agent workflow from roughly 100 to 15-$20 per run. The post How I cut coding agent cost
When building consumer-facing generative AI applications, balancing high generation quality with fast response times across diverse media types, can be challenging. KDDI, a major telecommunications ca
I'm the CTO of a mid-size nonprofit, and I run a local Qwen stack as a coding subagent for that work — partly on principle (a fair amount of what I handle shouldn't leave the building), partly because
I am pleased to see that OpenAI’s new model is roughly on par with Gemini Flash and Opus 4.8 on prompt injection risk. Nice work! Evaluating and naming other labs turns out to be a great way to encour
I am seeing claims around LLMs, specifically Astra having made serious progress on robotics. The tasks that are demonstrated are simple pick and place tasks with parallel jaw grippers. LLMs can do pla
I kept picking local models badly. Parameter count, quantization, VRAM. I could read all of it and still not know whether a model would be any good on my machine. So I stopped guessing and wrote somet
Hello LocalLLaMA Fam! I built this project because I, like many of us here, enjoy local AI and have a nice setup, but still find myself reaching for Cloud-based AI due to their convenience. My models