631
Sarvam 105B, the first competitive Indian open source LLM
Sarvam open-sources 30B and 105B reasoning models trained in India: MoE LLMs with leading Indian-language benchmarks; weights on AI Kosh and Hugging Face.
New
The latest blog posts to take off on Hacker News. 3,237 posts.
632
LLMs work best when the user defines their acceptance criteria first
One of the simplest tests you can run on a database:
633
Anthropic, please make a new Slack
Slack will be the Waterloo of closed data.
634
Hardening Firefox with Anthropic's Red Team
Claude Opus 4.6 found 22 Firefox vulnerabilities in two weeks, 14 rated high-severity, with proofs-of-concept and candidate patches for Mozilla.
635
GPT-5.4
Introducing GPT-5.4, OpenAI’s most most capable and efficient frontier model for professional work, with state-of-the-art coding, computer use, tool search, and 1M-token context.
636
A GitHub Issue Title Compromised 4k Developer Machines
A prompt injection in a GitHub issue triggered a chain reaction that ended with 4,000 developers getting OpenClaw installed without consent. The attack composes well-understood vulnerabilities into something new: one AI tool bootstrapping another.
637
NanoGPT Slowrun: Language Modeling with Limited Data, Infinite Compute
NanoGPT Slowrun is an open effort to implement data-efficient learning algorithms; 5.5x data efficiency in the first week and improving.
638
639
A case for Go as the best language for AI agents
Pull up your agents folks, I'll convince you why Go is the best language for them.
640
Decision trees – the unreasonable power of nested decision rules
An introduction to the Decision Trees, Entropy, and Information Gain.
641
TorchLean: Formalizing Neural Networks in Lean
Neural networks are increasingly deployed through software stacks whose behavior is defined by tensor layouts, preprocessing choices, compiler boundaries, floating-point arithmetic, GPU kernels, and external model exporters. TorchLean is a Lean 4 codebase for neural network specification, execution, and verification…
642
Don't trust AI agents
AI agents need a security model that assumes things will go wrong. The right response isn't better permission checks, it's architecture that makes trust unnecessary.
643
MCP server that reduces Claude Code context consumption by 98%
Mert Köseoğlu • Senior SWE (IC) with 12+ YoE building…2026570 points107 commentsAI & Machine Learning details
MCP server that reduces Claude Code context consumption by 98%. 315 KB becomes 5.4 KB.
644
We gave terabytes of CI logs to an LLM
We gave our AI agent a SQL interface to billions of CI log lines in ClickHouse. How we ingest, store, and query 1.5 billion log lines a week.
645
Scaling AI for everyone
Today we’re announcing $110B in new investment at a $730B pre money valuation. This includes $30B from SoftBank, $30B from NVIDIA, and $50B from Amazon.
646
What Claude Code chooses
Benchmarking and Ground Control for developer tool providers. Understand agent choices and test what helps them succeed.
647
Nano Banana 2: Google's latest AI image generation model
Our latest image generation model offers advanced world knowledge, production-ready specs, subject consistency and more, all at Flash speed.
648
An autopsy of AI-generated 3D slop
Thinking of using AI for your Shopify 3D models? Read this first. We compare AI-generated 3D 'slop' vs. handcrafted models to show why the human touch is very much required to attain positive ROI.
649
Can you reverse engineer our neural network?
A lot of “capture-the-flag” style ML puzzles give you a black box neural net, and your job is to figure out what it does. When we were thinking of creating o...
650
Mercury 2: Fast reasoning LLM powered by diffusion
Today, we're introducing Mercury 2 — the world's fastest reasoning language model, built to make production AI feel instant.
651
How we rebuilt Next.js with AI in one week
One engineer used AI to rebuild Next.js on Vite in a week. vinext builds up to 4x faster, produces 57% smaller bundles, and deploys to Cloudflare Workers with a single command.
652
We are changing our developer productivity experiment design
Our second developer productivity study faces selection effects from wider AI adoption, prompting us to redesign our approach.
653
654
Firefox 148 Launches with AI Kill Switch Feature and More Enhancements
The latest update of Firefox, version 148, introduces a much-anticipated "AI kill switch" feature, allowing users to disable AI functionalities such as chatbot prompts…
655
“Car Wash” test with 53 models
The car wash test is the simplest AI reasoning benchmark that nearly every model fails. We tested 53 models through Opper, first once each, then 10 times. Only 5 passed consistently.
656
Detecting and Preventing Distillation Attacks
We identified coordinated distillation attacks using ~24,000 fraudulent accounts to extract Claude's capabilities.
657
The First Fully General Computer Action Model
We trained a model on our 11-million-hour video dataset. Our model can explore complex websites, complete multi-action CAD modeling sequences, and drive a car in the real world, all at 30 FPS.
658
We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them
BinaryAudit benchmarks AI agents using Ghidra to find backdoors in compiled binaries of real open-source servers, proxies, and network infrastructure.
659
Meta Deployed AI and It Is Killing Our Agency
Managing millions in Meta ad spend, yet treated like we don't exist. An inside look at how Meta's AI-driven changes are disrupting agencies and advertiser performance.
660
Every company building your AI assistant is now an ad company
Exploring how agents become multipliers, not replacers.