541
Running Kimi K3 on MI355X at Better Performance per Dollar Than B300
Running Kimi K3 at ~952 tok/s/node, AMD continues to prove its case as the winner in performance per dollar.
New
The latest blog posts to take off on Hacker News. 1,520 posts.
542
AI financial advice is surprisingly good, especially if you ask right questions
Large language models encourage smart financial behavior, but they fall short on the more subtle aspects of saving and investing, according to MIT Sloan’s Taha Choukhmane and co-authors.
543
Explorative modeling: Train on the best of K guesses
What in the world is Explorative Modeling?
544
545
AI doesn't generate working products, that's still your job
AI has dramatically accelerated the path to a first working version. It has not shortened the distance between a first working version and something production-grade.
546
Stateless MCP has recaptured my interest
Tuesday was Stateless MCP day—the rollout of MCP 2.0, or the 2026-07-28 Model Context Protocol specification to use the more formal but less memorable name. This is the most significant …
547
Everyone is building LLM routers, we deprecated ours
We don't believe in model routing anymore. For most use cases, sticking to a single battle-tested model is the best thing you can do.
548
Why we write our own C and C++ inference engines
LocalAI is the open source AI engine. Run any model, LLMs, vision, voice, image and video, on any hardware. No GPU required.
549
Autoregressive Language Model on the 6502 Processor
A language model on a 1975 MOS 6502 processor, using a tiny Mamba-based language model and custom 8-bit ternary inference engine, used to generate text on a BBC Micro from the 1980s.
550
Google fixed more Chrome bugs in June than over the past two years, thanks to AI
Chrome uses Gemini AI to automate vulnerability discovery, triage, and patching, accelerating updates to match modern security risks.
551
The AI Aesthetic
Every zeitgeist comes with new design idioms unique to its challenges. Many of them disappear as fads change, but others bake themselves into deeper parts of…
552
Investigating three real-world incidents in our cybersecurity evaluations
We found three cases where a Claude model reached the internet from an evaluation environment and accessed real systems. We share what happened and what we're changing.
553
I flagged two research papers for fake authors and both were accepted as orals
We reviewed 22 ML conference submissions this summer. Fifteen had fabricated citations, hallucinated authors, or clear LLM slop, so we complain about that a bit, and release the paper references audit we now run.
554
We Gave GPT 5.6 Sol a Real Business. It Lied, Spammed, and Lost $447
If an agent had a wallet, a computer, and 24 hours, could it run a profitable startup?
555
Advancing the price-performance frontier with GPT‑5.6
Explore lower GPT‑5.6 pricing for Luna and Terra—and how OpenAI’s more efficient models help enterprises deploy AI workflows at scale.
556
Gemini Robotics 2 brings whole body intelligence to robots
From feet to fingertips — we are teaching robots intelligent whole-body control, fine dexterity, and teamwork to complete a broad range of complex tasks.
557
OpenJDK Interim Policy on Generative AI
The field of generative AI is evolving quickly. It brings compelling opportunities to improve developer productivity, but it also brings risks: to reviewer burden, to safety and security, and to intellectual property.
558
Some thoughts about Anthropic's new cryptanalysis results
Yesterday Anthropic published two new cryptanalysis results, both outputs of Claude Mythos, their (still) unreleased advanced model. The first of these results attacks a signature scheme called HAW…
559
GPT-5.6 vs. Claude Fable 5 for Physical AI, which performs best?
We tested the latest frontier models in the Dyad agent on five modeling and simulation problems, comparing accuracy, cost, time, and work style.
560
Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution
We gave 64 real coding tasks to self-hosted GPUs, rented hardware and commercial APIs. Owning a box lands in the same cost ballpark as renting one.
561
After the AI Crash
Everything I read about the AI industry leads me to think there will be an AI crash. Consider the following: Unsustainable Capital Expenses. It’s hard to imagine there can ever be enough revenue to…
562
Document-borne AI worms can self-propagate through Copilot for Word
I would like to thank Microsoft product teams and Microsoft Security Response Center (MSRC) for collaborating with me on this technical analysis and mitigation of the disclosed vulnerabilities. The editorial opinions reflected below are solely the author’s and do not necessarily reflect those of the organizations I…
563
Discovering Cryptographic Weaknesses with Claude
Anthropic researchers find weaknesses in cryptographic algorithms with Claude Mythos Preview
564
A walk through of the DeltaNet family of linear attention variants
Guide to the DeltaNet Family of linear attention mechanisms.
565
Kimi K3 Architecture Overview and Notes
Short architecture note on Kimi K3, including LatentMoE, Kimi Delta Attention, Attention Residuals, NoPE, multimodality, and inference-efficiency choices.
566
Twenty-five years ago it was cryptography, today it's model weights
Twenty-five years ago, I made my first donation to an open source project and purchased a CD with an operating system as downloading a few hundred megabytes over a 14.4kbps dial-up wasn't very fun.
567
568
A $500 RL fine-tune of a 9B open model beat frontier models on catalog review
Why the companies winning in the AI age own their intelligence: a task-trained open source model vs the frontier, with every number published.
569
570
Our position on open-weights models
Anthropic CEO Dario Amodei on open-weights models