AI News Today

Article URL: https://news.err.ee/1610060290/estonia-to-become-first-country-to-issue-id-codes-to-ai-agents Comments URL: https://news.ycombinator.com/item?id=48626127 Points: 4 # Comments: 2

Article URL: https://www.bloomberg.com/news/articles/2026-06-04/airbnb-ceo-brian-chesky-plans-to-start-a-new-ai-company Comments URL: https://news.ycombinator.com/item?id=48626084 Points: 4 # Comments: 1

It preys on 'zombie fungus'. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

I built a governed memory layer for AI assistants with deletion compaction, vector purge verification, tenant isolation, and audit evidence. Comments URL: https://news.ycombinator.com/item?id=48626034 Points: 2 # Comments: 0

There has been a few posts about LM Link here on HN, but I don't believe this has gotten the attention it deserves. https://news.ycombinator.com/item?id=47155258 Finally got around to dusting off a old setup and I am thrilled with the results. I'm using a 200$ used GPU and 11year old CPU. Basically a retired game rig. I can now host a private model to my entire family that is just as capable as the things they were using it for without worrying of all the private information leaking out. While I

Hot tubs sit at about 38 to 40 degrees Celsius, warm enough that most people can only soak for about 15 minutes. NVIDIA’s newest AI servers can run their cooling liquid even hotter — up to 45 degrees Celsius, or 113 degrees Fahrenheit. That higher temperature limit is precisely what makes them more energy efficient. […]

Bubble Sort. Terrible in practice, but probably the first algorithm that made me understand what “sorting” actually means. I respect it for that. Comments URL: https://news.ycombinator.com/item?id=48625785 Points: 2 # Comments: 4

Article URL: https://github.com/xfloukiex-lab/magpie-search Comments URL: https://news.ycombinator.com/item?id=48625507 Points: 1 # Comments: 0

<p> Let agents deploy before signup </p> <p> <a href="https://www.producthunt.com/products/cloudflare?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1177735?app_id=339">Link</a> </p>

Article URL: https://milvus.io/blog/when-ai-agents-do-the-work-what-do-we-lose.md Comments URL: https://news.ycombinator.com/item?id=48625341 Points: 1 # Comments: 0

Article URL: https://github.com/vishal-dehurdle/state-harness Comments URL: https://news.ycombinator.com/item?id=48625317 Points: 2 # Comments: 0

Article URL: https://www.typevia.com/ Comments URL: https://news.ycombinator.com/item?id=48625040 Points: 2 # Comments: 0

<p>Argentina just got its first national MCP ecosystem — and it was built from Bahía Blanca.</p> <p>CHE MCP is an intelligent gateway that connects any AI agent with real-time Argentine data. Dollar exchange rates, weather, football, tax compliance (ARCA), inflation, public transit — 80+ official data sources through a SINGLE MCP server.</p> <p>Why does this matter? Because right now, if you want your AI to answer "¿cuánto está el dólar blue?", you either Google it yourself or install 80 differe

<pre style='white-space:pre-wrap;width:81ex'>fix(test): keep Gemini CLI docker lane advisory</pre>

Article URL: https://www.theatlantic.com/category/ai-watchdog/ Comments URL: https://news.ycombinator.com/item?id=48624870 Points: 1 # Comments: 0

ChatGPT 3.5 only came out about three and a half years ago. Claude Code was released in February 2025. In one academic year, LLMs went from barely handling high school math to disproving unit-distance conjecture. Every two months we seems to have a new SOTA now. What do you expect AI coding to look like four years from now? Will it be like Go after AlphaGo, where even Ke Jie could only say "they no longer knew what they were fighting against"? Or will it just be like today, just stronger, with h

<h1> Bridging Python and Rust: Mitigating GIL Contention in a High-Throughput LLM Gateway </h1> <p>When building <strong>Aegis</strong>, an open-source OpenAI-compatible governance proxy, we made a core architectural decision: use Python (FastAPI/ASGI) for rapid development and API adaptability, but offload high-performance cryptography, Write-Ahead Logging (WAL), and Merkle Mountain Range (MMR) operations to a compiled Rust extension (<code>aegis_rust_v2</code>) via PyO3 and Maturin.</p> <p>How

"A unique, unprecedented finding." ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Article URL: https://sakana.ai/fugu/ Comments URL: https://news.ycombinator.com/item?id=48624782 Points: 54 # Comments: 19

Article URL: https://www.abc.com.py Comments URL: https://news.ycombinator.com/item?id=48624781 Points: 1 # Comments: 0

Article URL: https://drive.google.com/file/d/1MtiNjQeQO9ilLuEFXwVeKgWOzec4F2Dn/view?usp=drive_link Comments URL: https://news.ycombinator.com/item?id=48624746 Points: 3 # Comments: 5

<pre style='white-space:pre-wrap;width:81ex'>chore(deadcode): remove stale agent helper reexports</pre>

<pre style='white-space:pre-wrap;width:81ex'>Merge branch 'main' of https://github.com/openclaw/openclaw into fix/stable-tool-result-prompt-cache * 'main' of https://github.com/openclaw/openclaw: (39 commits) chore(deadcode): remove stale agent helper reexports chore(ui): refresh fa control ui locale chore(ui): refresh nl control ui locale chore(ui): refresh th control ui locale chore(ui): refresh vi control ui locale chore(ui): refresh id control ui locale chore(ui): refresh pl control ui local

I think RL as a method which produces training data by model's predictions — It directly leads the model to extend its output range because of increased diversity of the data. However, fundamentally RL relies on bootstrapping and has moving target problem which are the reason of its poor stability. One of the most tractable method to approximate value function is TD which causes sample noise, function approximator error and moving target problems. I argue that we need to extend pure RL theory at

Hey HN, I built woltspace as a way to interact with my coding agents when I'm away from the computer. It's fully containerized, so you can give them full access inside their sandbox (the lodge). Right now it supports Claude Code. Each session can be controlled via telegram or slack. You can fully access the terminal running the sessions from anywhere thanks to Cloudflare tunnels. Wolts are creatures with their own personality, goals and memory system. Wolts vary in ability, racoons are smartest

Article URL: https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/ Comments URL: https://news.ycombinator.com/item?id=48624535 Points: 3 # Comments: 1

Article URL: https://www.crossary.com Comments URL: https://news.ycombinator.com/item?id=48624501 Points: 2 # Comments: 0

I have been on the 20x Max plan of Claude Code for a while now, however since last week something changed and I seem to have been getting lots and lots of those 529 errors. My workflow hasn't changed, same level of concurrency as usual, did they change their limits? Comments URL: https://news.ycombinator.com/item?id=48624168 Points: 6 # Comments: 2

Article URL: https://medium.com/dev-genius/i-built-an-mcp-server-in-200-lines-of-go-and-claude-became-10x-more-useful-5352546a48cb Comments URL: https://news.ycombinator.com/item?id=48623994 Points: 2 # Comments: 1

The idea with Wirewright is to treat programs and algorithms as physical mechanisms, as machines or societies of machines inhabiting an immutable symbolic world, and time-step the latter to see what happens. This last part in particular reminds me of cellular automata. Said differently, in Wirewright, programs are modeled as "nouns" or "societies" of interacting "nouns" (think data structures). The world is then subjected to laws of physics -- symbolic physics; which is the only "verb" here (thi

Vercel Functions can now serve WebSocket connections, enabling bidirectional communication between clients and server-side code on Vercel. Use WebSockets for realtime features such as interactive AI streaming, chat, and collaborative apps. WebSocket connections run Fluid compute and follow the same limits and pricing as other Function invocations. With Active CPU pricing , billing only applies to the time your Function spends processing messages, not idle connection time. You can serve WebSocket

Ten days after Washington pulled Anthropic's top models from foreign hands, the bill came due. This week Beijing blacklisted 56 American firms, Anthropic's own filing admitted the trigger was a routine coding request rival models can run, and Microsoft's CEO warned that letting "a few models eat everything" won't survive politically. The export war just stopped being one-directional — here's the week that made it mutual.

Learn how Jason Liu uses Codex to preserve context, manage complex projects, and help work continue beyond a single prompt.

Sakana Fugu Ultra from Sakana AI is now available on AI Gateway . Fugu Ultra is built on a pool of publicly accessible frontier models, rather than running as a single model. It coordinates several models, routing work to 1-3 agents depending on the problem and combining their results into a single answer. Based on reasoning and scientific benchmarks, Fugu Ultra has capabilities similar to those of Claude Mythos Preview and Fable 5. To use Fugu Ultra, set model to sakana/fugu-ultra in AI SDK : A

The Workflow SDK 5 beta now compresses all run, hook, and step inputs and outputs with zstd . Compression kicks in automatically, but only when it helps. Small payloads stay as-is, larger ones get compressed before they're persisted. Compressed payloads use less storage and are faster to read and write, so your workflows run faster and cost less. The savings are largest for JSON payloads typical of AI conversations, where storage size and cost can drop by up to 85%. One typical Workflow, run and

Article URL: https://taxlens.getdynamiq.ai Comments URL: https://news.ycombinator.com/item?id=48623803 Points: 3 # Comments: 0

I built ANMA because I noticed that cheaper models would often ignore architecture rules. So I did several benchmarks using "Claude Haiku 4.5" with and without ANMA; without ANMA it ignored the "rules" 13 out of 19 runs, with ANMA, 0 out of 20 runs. What is "ANMA"? YAML contracts with CLAUDE.md, hooks, and CI checks What about stronger/expensive models? They followed the architecture rules. The question is, would cheaper models with stronger rules be the best affordable default for coding? Comme

Article URL: https://github.com/pilotpirxie/ai-colours Comments URL: https://news.ycombinator.com/item?id=48623764 Points: 2 # Comments: 0

Article URL: https://github.com/oussamaKH63/peekai Comments URL: https://news.ycombinator.com/item?id=48623745 Points: 2 # Comments: 0

Now that we've found it, we can't afford to lose it. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.