AI News Today

Article URL: https://claimmateai.pro/ Comments URL: https://news.ycombinator.com/item?id=48800815 Points: 1 # Comments: 1

New obesity treatments could follow. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Corbin and Parmida were dumped on 'Love Island USA' after a public vote, and the internet is mourning the loss of CorbinGPT.

Article URL: https://brunelly.com/articles/what-is-ai-native-development Comments URL: https://news.ycombinator.com/item?id=48800518 Points: 2 # Comments: 0

Article URL: https://arxiv.org/abs/2607.01953 Comments URL: https://news.ycombinator.com/item?id=48800505 Points: 1 # Comments: 0

<pre style='white-space:pre-wrap;width:81ex'>perf(agents): slim Skill Workshop prompt section to its routing contract (#100449) (#100481) Prompt section drops from 13 lines to 2 (2128 -> 319 bytes); action mechanics move into the skill_workshop tool schema (+25 bytes), for a net -1784 bytes per request on affected sessions. Routing contract and approval flow unchanged.</pre>

<pre style='white-space:pre-wrap;width:81ex'>feat(commands): add /learn to draft skills from recent work (#100442) * feat(commands): add /learn to draft skills from recent work (#100408) /learn rewrites the turn into a standards-guided Skill Workshop authoring instruction: the agent gathers named sources (or distills the current conversation) and files ONE pending skill proposal via skill_workshop. Approval flow unchanged; sandboxed/tool-restricted agents get a clear unavailable reply. Extracts

<p> A customizable, agent-ready open-source design system </p> <p> <a href="https://www.producthunt.com/products/meta?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/1188925?app_id=339">Link</a> </p>

<h2> The Problem Nobody Talks About </h2> <p>Everyone's posting AI-generated videos — characters speaking with lip-sync, manga panels coming alive, virtual idols dancing. The pitch: "just describe what you want."</p> <p>I tried. For months. Here's what I got:</p> <ul> <li>Character's face morphed by frame 2</li> <li>"Slowly looks up" became "violent head shake"</li> <li>Voice-over sounded like Google Translate</li> <li>Same prompt, 3 runs, 3 completely different results</li> <li>No idea what to

<p>A month ago we added a <code>card_token</code> column to the users table so a background job could retry failed Pro charges. It lasted about two days. Storing card data in your own database drops you into PCI-DSS (the compliance standard that kicks in the moment card data touches your systems), so we pulled it and moved to Stripe-managed payment methods.</p> <p>Last week the charges started failing again. New Claude Code session, no memory of any of that. Its plan? Add a <code>card_token</cod

Article URL: https://aischoolbuilder.com/school/school-1786 Comments URL: https://news.ycombinator.com/item?id=48800410 Points: 1 # Comments: 0

<p>I fine-tuned Gemma 4 E2B. The adapter merged cleanly. The export to <code>.litertlm</code> completed without errors. I pushed the model to my phone, initialized the engine, and everything looked green. Then I tried to create a conversation and got this:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight plaintext"><code>Failed to apply template: unknown method: map has no method named get (in template:238) </code></pre> </div> <p>No model loading failure. No quantization

<p>In February, city officials in Cheyenne, Wyoming discovered something in their reclaimed water system that shouldn't have been there: <em>Cupriavidus gilardii</em>, a rare metal-resistant bacterium traced to wastewater discharges from Meta's $800 million data center campus. The <a href="https://cowboystatedaily.com/2026/07/02/cheyenne-wont-take-data-center-wastewater-after-meta-company-contaminated-system/" rel="noopener noreferrer">contamination shut down Cheyenne's reuse water system for mo

<p> The AI video editor you hire that edits like a human </p> <p> <a href="https://www.producthunt.com/products/stanley-studio?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/1188912?app_id=339">Link</a> </p>

<p>US companies hoping to make fortunes from AI want the creative product of our country to be available to them for free, or for peanuts. Words fail me</p><p>Last week I went to Canberra with a group of other people who live by selling our creative work. We make the books, songs, artwork, First Nations artworks, films, music and TV shows that delight and entertain us, reveal us to ourselves as humans, and connect us as Australians.</p><p>The works we make are our property. They are ours to sell

<p>The obvious fix for "my projects all look similar" is a better README — more screenshots, clearer descriptions, maybe a comparison table. I considered that for about five minutes and decided it was still just a nicer list of links.</p> <p>What actually made a portfolio project <em>feel</em> different was making it something you could talk to instead of read. That's what MCP (Model Context Protocol) is built for — it's the standard that lets AI clients like Claude Desktop call external tools d

Article URL: https://cewsco.com/ Comments URL: https://news.ycombinator.com/item?id=48800302 Points: 1 # Comments: 0

<pre style='white-space:pre-wrap;width:81ex'>fix(agents): run bootstrap ritual on Claude CLI (#100560) * fix: CLI-backend agent runtimes skip the BOOTSTRAP.md first-run ritual CLI-backend runtimes (claude-cli, gemini-cli) never rendered the "Bootstrap Pending" system-prompt gate the embedded runner produces, so on a fresh workspace with a pending BOOTSTRAP.md the agent greeted the user instead of running the first-run bootstrap ritual. BOOTSTRAP.md reached CLI backends only as passive Project Co

<pre style='white-space:pre-wrap;width:81ex'>fix(browser): notify agent when click triggers download (#93307) * fix(browser): drain download saves and use monotonic cursor for act response - Add drainPendingDownloadSaves() to wait for in-flight saveAs before sampling - Use monotonic downloadSeq cursor instead of bounded-list length - Propagate downloads info in POST /act response for click/batch/evaluate Fixes #93250 * fix(lint): add braces around single-line if returns Fixes eslint(curly) failu

Article URL: https://arxiv.org/abs/2402.14873 Comments URL: https://news.ycombinator.com/item?id=48800043 Points: 1 # Comments: 0

Article URL: https://www.forbes.com/sites/jemmagreen/2026/07/02/ai-costs-more-than-the-people-it-replaced/ Comments URL: https://news.ycombinator.com/item?id=48799841 Points: 3 # Comments: 0

PLUS: India attacks Uber; South Korea warned AI boom is dangerous; Scuba-diving cyborg cockroach; and more!

Hi HN, I was once given the advice: Don't waste expensive frontier model credits (GPT/Claude/etc.) on bulk work. Send the boring, repetitive, high-volume jobs to a smaller model, and save the expensive prompts for when you actually need frontier-level reasoning. I complained and told my manager that I shouldnt have to think about using certain models for certain coding tasks, and that one model should handle everything. Well, here we are anyway. If anyone needs a place to absolutely abuse an LLM

Quick off the mark. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Article URL: https://www.youtube.com/watch?v=4nzzpUKhj1M Comments URL: https://news.ycombinator.com/item?id=48799667 Points: 1 # Comments: 0

Comments

<p>Race to develop ‘embodied AI’ focuses on creating dextrous hands to transform humanoid robots from gimmicks into useful products</p><p>Human hands – nimble, nerve-filled appendages that are the most flexible part of the human skeleton – are exceptionally complex. Many tasks that most people can do largely without thinking, from tying a pair of shoelaces to buttoning up a shirt, in fact require a complex set of neurological instructions and precise choreography. In thousands of years of human

Article URL: https://www.clusy.io Comments URL: https://news.ycombinator.com/item?id=48799415 Points: 1 # Comments: 0

Article URL: https://noozra.com/articles/the-llm-big-bang Comments URL: https://news.ycombinator.com/item?id=48799386 Points: 2 # Comments: 0

Article URL: https://www.readtheline.ca/p/al-vigier-canadas-ai-strategy-shouldnt Comments URL: https://news.ycombinator.com/item?id=48799256 Points: 3 # Comments: 1

Article URL: https://github.com/PJHkorea/egregore-core-jax Comments URL: https://news.ycombinator.com/item?id=48799235 Points: 2 # Comments: 1

Speech-only spoken language models (SLMs) lag behind text and text-speech models in performance, with recent discrete autoregressive (AR) SLMs indicating significant computational and data demands to match text models. Since discretizing continuous speech for AR creates bottlenecks, we explore whether continuous diffusion (CD) SLM is more viable. To quantify the SLMs linguistic quality, we introduce the phoneme Jensen-Shannon divergence (pJSD) metric. Our analysis reveals CD SLMs, mirroring AR b

Language models play a central role in automatic speech recognition (ASR), yet most methods rely on text-only models unaware of ASR error patterns. Recently, large language models (LLMs) have been applied to ASR correction, but introduce latency and hallucination concerns. We revisit ASR error correction with compact seq2seq models, trained on ASR errors from real and synthetic audio. To scale training, we construct synthetic corpora via cascaded TTS and ASR, finding that matching the diversity

Safety policies define what constitutes safe and unsafe AI outputs, guiding data annotation and model development. However, annotation disagreement is pervasive and can stem from multiple sources such as operational failures (annotators misunderstand or misexecute the task), policy ambiguity (policy wording leaves room for interpretation), or value pluralism (different annotators hold different perspectives on safety). Distinguishing these sources matters. For example, operational failures call

Look past the data-center megadeals and the bubble headlines, because something quieter happened this week: AI actually delivered. A Nobel laureate joined the AI-for-science push, the open-model frontier got cheaper again, and the first hard evidence landed that an AI tutor can beat the classroom. These are the wins that did not make the noise.

Article URL: https://mattmccormick.ca/we-re-all-managers-now-my-journey-into-ai-assisted-development/ Comments URL: https://news.ycombinator.com/item?id=48799212 Points: 3 # Comments: 0

I built an open source terminal app that transparently wraps your shell so you can have access to AI whenever you need it, while it stays completely out of your way the rest of the time. Bring your own AI agent (Claude and Codex are supported out of the box) with auto-configured MCP access to the contents of your terminal and scrollback and the ability to suggest input that you can approve passing to your shell as if you had typed it yourself. The hardest part technically was getting it to be tr

A respected engineer has informed me that the terms of service of most if not all AI outfits mandate that anything you submit becomes their property, and he said that already he knows of 2 instances of startup ideas being stolen away by the AI outfits. He did not name names. Comments URL: https://news.ycombinator.com/item?id=48798971 Points: 3 # Comments: 2

<pre style='white-space:pre-wrap;width:81ex'>fix(openai): bound realtime voice websocket payload at 16 MiB (#99450) * fix(openai): bound realtime voice websocket payload at 16 MiB * test(openai): cover realtime voice payload cap --------- Co-authored-by: Peter Steinberger <steipete@gmail.com></pre>

<pre style='white-space:pre-wrap;width:81ex'>fix(agents): trim media tools in lean mode (#88881) * fix(agents): trim media tools in lean mode * fix(agents): preserve lean tool allowlists * fix(agents): preserve lean tool overrides * fix(agents): enforce lean policy at harness boundary * fix(agents): preserve lean override provenance * test(agents): satisfy lean harness contracts * docs(changelog): record lean local-model tool trimming --------- Co-authored-by: Peter Steinberger <steipete@gmail.c