AI News Today

Article URL: https://sigmap.io/ Comments URL: https://news.ycombinator.com/item?id=48792152 Points: 3 # Comments: 0

AI search agents rarely fail at multi-step research because of the search itself. Their real problem is not asking the user for clarification when queries are ambiguous. A new benchmark called DiscoBench shows that models searching repeatedly instead of asking follow-up questions actually perform worse, at 51.9 percent, than those that just guess. Even the best model only hits 43 percent overall accuracy. When ambiguity is removed from the queries, accuracy jumps by up to 40 points. The article

<p>Hi! My name is Sergio Tortosa, and I made my AI-powered documentation generation ...in Rust. Take into account I started this a year and a half ago, when LLMs weren't as advanced as they are today. So I hope this will serve as a story to learn both some technical info as well as what to do (or rather, what not to do) when starting a startup; this should also serve as a kind of evolution of the viability of Rust for web projects. Was this a good idea? You'll find out.</p> <h1> About the projec

<blockquote> <p>Many great tools for coding agents already exist.</p> <p>The painful part is discovering, installing, updating, and combining them without conflicts.</p> <p><strong>tokless is the lazy CLI that connects everything with one command.</strong></p> </blockquote> <p><strong>tokless</strong> integrates six popular packages (kaparthy skills, caveman, ponytail, rtk, codegraph, context-mode) for improving AI coding agents in under 30 seconds.</p> <p>Choose your agents, install, restart, a

<!-- SC_OFF --><div class="md"><p>I made a 10MB LoRA adapter for Qwen3.5-4B plus a small orchestration layer. It decides, per query, whether to answer directly, search the web, or retrieve from your own local documents and it refuses to make things up when it can't verify an answer.</p> <p>It runs locally (Apple Silicon / MLX, with a GGUF build for llama.cpp/Ollama).</p> <p>Basically small instruct models are poor at telling users how confident they really are. They can't verbalise it and tend t

<h1> Two signatures are better than one — bilateral provenance for AI agents </h1> <p>An AI agent produces a financial report. You notarize it — a 239-byte cryptographic record, signed by an independent notary, anchored on NEAR. The record proves: <em>this hash existed at this timestamp, and $0.01 USDC was paid for the attestation.</em></p> <p>A week later, someone asks the obvious question: <strong>who proved the agent wrote the report?</strong></p> <p>Nobody. The notary signed a hash the clien

Article URL: https://github.com/mensfeld/code-on-incus Comments URL: https://news.ycombinator.com/item?id=48792094 Points: 1 # Comments: 0

<p>The <strong>Midjourney Hollywood lawsuit</strong> just took a turn that matters more to a small studio in Colombo than to Disney's legal team. According to <a href="https://techcrunch.com/2026/07/04/midjourney-wants-hollywood-studios-to-reveal-the-details-of-their-ai-usage/" rel="noopener noreferrer">TechCrunch</a>, Midjourney is now trying to compel the studios suing it to reveal exactly how <em>they</em> use AI internally, including the prompts they typed and the outputs they got back.</p>

For the past few months, I've been sending every story that reaches the HN front page through Pangram's AI detector and noting the scores. The reason I made it is that I kept noticing posts that sounded like AI-generated to me. Sometimes the comments pointed it out, sometimes a post was near the top all day and no one said a word. I wondered if the share was really increasing or if I was just primed to see slop everywhere. How it operates: The site identifies the top 30, gets each article, and e

WakaWiki, agent-driven docs for your codebase. Inspired by OpenWiki. Comments URL: https://news.ycombinator.com/item?id=48791965 Points: 1 # Comments: 0

Article URL: https://hunch-pi.vercel.app Comments URL: https://news.ycombinator.com/item?id=48791919 Points: 1 # Comments: 0

<pre style='white-space:pre-wrap;width:81ex'>fix(discord): reuse stored sessionId across voice (stt-tts) turns (#97746) * fix(discord): reuse stored sessionId across voice (stt-tts) turns Discord voice turns previously called agentCommandFromIngress without a sessionId, so the core session resolver minted a fresh UUID per turn. Because the codex app-server thread is bound beside the per-session file, every voice turn produced a new codex thread, a brand-new rollout-*.jsonl, and lost all prior co

Article URL: https://drive.google.com/file/d/1oPTmOOkBtE0IRV4oy0K4NuuLxYBezXRc/view?usp=sharing Comments URL: https://news.ycombinator.com/item?id=48791885 Points: 1 # Comments: 0

Article URL: https://github.com/williamliu-ai/fidx Comments URL: https://news.ycombinator.com/item?id=48791740 Points: 1 # Comments: 0

Article URL: https://handmadeoasis.com/llms-as-a-different-kind-of-intelligence/ Comments URL: https://news.ycombinator.com/item?id=48791650 Points: 3 # Comments: 0

Head of AI startup trying to bring AI to the masses Comments URL: https://news.ycombinator.com/item?id=48791647 Points: 1 # Comments: 0

OpenAI just released Secure MCP Tunnel a week ago, but it is only for OpenAI ecosystem. Has any Claude Desktop/Code user here figure out a method to use the Secure MCP Tunnel. I know, it's from OpenAI. Comments URL: https://news.ycombinator.com/item?id=48791510 Points: 2 # Comments: 0

Article URL: https://github.com/shimo4228/codex-review Comments URL: https://news.ycombinator.com/item?id=48791474 Points: 1 # Comments: 0

Article URL: https://www.sfgate.com/politics/article/california-ai-study-22321472.php Comments URL: https://news.ycombinator.com/item?id=48791406 Points: 3 # Comments: 0

Article URL: https://ourworldindata.org/data-insights/us-and-chinese-companies-train-almost-all-of-the-worlds-most-used-ai-models Comments URL: https://news.ycombinator.com/item?id=48791385 Points: 4 # Comments: 0

Comments

Article URL: https://xyzlabs.substack.com/p/meituan-trained-a-16t-parameter-ai Comments URL: https://news.ycombinator.com/item?id=48791366 Points: 3 # Comments: 1

Article URL: https://tech.yahoo.com/ai/articles/china-longcat-2-0-becomes-134258951.html Comments URL: https://news.ycombinator.com/item?id=48791362 Points: 6 # Comments: 1

<!-- SC_OFF --><div class="md"><p>As someone who is <strong>not affiliated</strong> with any of the big tech companies, I find it particularly difficult to have the confidence or enthusiasm to approach any ML problem with an attitude that my professors probably had at my stage in life. I'm sure I am not the only one having the following thoughts:</p> <ul> <li>"My research is currently being done better at companies."</li> <li>"ML problem I set out to solve is already solved and in fact turned in

<pre style='white-space:pre-wrap;width:81ex'>feat: animate the OpenClaw mascot across web, iOS, and Android Ports the openclaw.ai hero mascot animation (body float 4s, eye blink 3s, antenna wiggle 2s, staggered claw snaps 4s) to every product surface that shows the logo: - web: ui/public/favicon.svg now carries SMIL animations, so every <img> usage (login gate, sidebar brand, agent avatar fallback) animates with no markup changes; CSS animations do not run in <img>-loaded SVGs. - iOS: new OpenCl

Article URL: https://www.utilix.tech/ Comments URL: https://news.ycombinator.com/item?id=48790934 Points: 1 # Comments: 1

<pre style='white-space:pre-wrap;width:81ex'>[AI-assisted] fix(android): polish Voice text truncation (#100060)</pre>

I kept hitting the same wall: AI assistants are great until they need my context — the email thread, the calendar, the iMessage, the file on my disk. Cloud connectors can only reach things with a public API, so they physically can't touch the stuff that lives only on my machine. So I built Local MCP — a native macOS app that connects Claude Desktop, Claude Code, Cursor, and ChatGPT to your real local apps over MCP: Mail, Calendar, Contacts, iMessage, Microsoft Teams, Slack, WhatsApp, Notes, Remi

<h2> Osloq — ใช้ AI หาสาเหตุ bug แทน </h2> <p>เวลา AI coding tools เสนอจะ "fix bug ให้" — เราได้แต่กด Accept หรือไม่ก็ Reject สองปุ่ม สองทางเลือก</p> <p>แต่เราไม่เคยรู้ว่า:</p> <ul> <li>AI รู้ได้ยังไงว่า bug เกิดจากตรงนี้?</li> <li>มัน reproduce แล้วหรือแค่อ่านโค้ดแล้วเดา?</li> <li>ถ้าเรา accept — มันจะพังของอย่างอื่นไหม?</li> </ul> <p><strong>Osloq</strong> เลือกทางที่สาม: ไม่ใช่ "fix ให้" — แต่ "<strong>หาให้เจอแล้วบอกว่าเกิดอะไรขึ้น</strong>"</p> <h2> Osloq คืออะไร </h2> <p>Osloq เป็น AI agen

<pre style='white-space:pre-wrap;width:81ex'>fix(agents): surface Claude CLI thinking token progress</pre>

<pre style='white-space:pre-wrap;width:81ex'>fix(tui): isolate embedded event listener failures (#100117) * fix(tui): suppress unhandled rejection from async agent event handler * fix(tui): isolate embedded event listener failures --------- Co-authored-by: Peter Steinberger <steipete@gmail.com></pre>

<p>Ask ChatGPT or Perplexity "what's the best note-taking app" and you get a shortlist of three to five names. Either you're on it or you don't exist in that channel. And buying research keeps moving there.</p> <p>People call measuring this GEO or AEO tracking now. The way most teams do it is pasting questions into chatbots by hand and eyeballing the answers. That stops scaling at about ten questions, and you can't trend it week over week.</p> <h2> Doing it programmatically </h2> <p>Don't scrape

<p>I just opened 3 security issues on two of the most popular AI agent frameworks on GitHub (combined 110K+ stars).</p> <h2> The Issues </h2> <p><strong><a href="https://github.com/microsoft/autogen/issues/7917" rel="noopener noreferrer">microsoft/autogen#7917</a></strong>: Docker code executor mounts host filesystem into sandboxed containers without trust boundary validation — container escape vector.</p> <p><strong><a href="https://github.com/microsoft/autogen/issues/7918" rel="noopener norefe

<p>I scanned three of the most popular AI agent frameworks with AgentGuard v0.6.1. The results were worse than I expected.</p> <h2> The Scan </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Framework</th> <th>Files</th> <th>Findings</th> <th>CRITICAL</th> <th>HIGH</th> <th>MEDIUM</th> </tr> </thead> <tbody> <tr> <td><strong>LlamaIndex</strong></td> <td>2,951</td> <td>1,003</td> <td>252</td> <td>558</td> <td>193</td> </tr> <tr> <td><strong>AutoGen</strong></td> <td>549</td> <td

Article URL: https://claude.ai/new Comments URL: https://news.ycombinator.com/item?id=48790577 Points: 2 # Comments: 0

Article URL: https://murderszn.github.io/sprout/ Comments URL: https://news.ycombinator.com/item?id=48790523 Points: 2 # Comments: 0

Article URL: https://github.com/tadelstein9/two-tier-memory Comments URL: https://news.ycombinator.com/item?id=48790505 Points: 1 # Comments: 0

15 years in remission. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

<p>I wrote about the <a href="https://simonwillison.net/2026/Jun/21/sqlite-utils-40rc1/">sqlite-utils 4.0rc1</a> release a couple of weeks ago. Since we only have Claude Fable on our Max subscriptions for a few more days, I decided to see if it could help me get to a 4.0 stable release that I felt truly comfortable about, since I try to keep to <a href="https://semver.org">SemVer</a> and like my incompatible major versions to be as rare as possible.</p> <p>I started with this prompt, in Claude C

<p><strong>Release:</strong> <a href="https://github.com/simonw/sqlite-utils/releases/tag/4.0rc2">sqlite-utils 4.0rc2</a></p> <p>See <a href="https://simonwillison.net/2026/Jul/5/sqlite-utils-fable/">sqlite-utils 4.0rc2, mostly written by Claude Fable (for about $149.25)</a>.</p>