Chronic constipation is a serious issue. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.
Article URL: https://asia.nikkei.com/business/technology/artificial-intelligence/indian-companies-look-to-chinese-llms-as-ai-costs-bite Comments URL: https://news.ycombinator.com/item?id=48890208 Points: 1 # Comments: 0
I use AI myself so I definitely know how it tends to write. When I encounter that kind of writing somewhere, I initially give it the benefit of the doubt and continue a bit. But when the signs get too obvious, I tend to just stop reading. It feels like the prompter is trying to trick me. They might for example have some simple agenda but they're trying to dress it up as some objective rational analysis with the help AI. It would work better if the author told how the content was created, and eve
Article URL: https://commercemarketplace.adobe.com/plumrocket-ai-connector.html Comments URL: https://news.ycombinator.com/item?id=48889994 Points: 2 # Comments: 0
Article URL: https://www.youtube.com/watch?v=_uWnr42zGnE Comments URL: https://news.ycombinator.com/item?id=48889897 Points: 2 # Comments: 0
Google Research's SensorFM is a foundation model trained on more than a trillion minutes of wearable data from five million Fitbit and Pixel Watch users. It beats existing benchmarks on 34 of 35 health and behavioral tasks. SensorFM could eventually power Google's AI health coach, but the company hasn't announced any integration plans yet. The article Google’s SensorFM turns messy wearable sensor data into a general-purpose health intelligence layer appeared first on The Decoder .
The Shark Robot Vacuum & Mop Combo PowerDetect NeverTouch Pro is 50% off at Amazon. Get this self-emptying and self-refilling robot vacuum and mop for $499.99, down from $999.99.
Article URL: https://medium.com/@DolphinDB_Inc/dolphindb-v3-00-6-ecde33229456 Comments URL: https://news.ycombinator.com/item?id=48889819 Points: 2 # Comments: 0
Article URL: https://narada.koley.in/ Comments URL: https://news.ycombinator.com/item?id=48889782 Points: 1 # Comments: 0
A couple weeks ago I was wondering if I could build a GGUF model file that, when run on ollama, would deterministically answer with the same exact sentence (not simply using a system prompt). That was surprisingly easy to implement, leveraging mostly the unembedder, so I started exploring GGUFs. First I built a simple GGUF that could compute the position of a point in a limited grid, using 4 directional tokens as input. All weights are "hand crafted", no training involved. Of course, being ollam
The Shark AI Robot Vacuum AV2501AE is on sale at Amazon for $269.99, down from its typical $289.99 price.
Waze is getting an AI makeover. Google is integrating its flagship AI assistant, Gemini, into the driving app with the goal of letting users personalize their trips a little more. Of the four new updates, only two are being described as involving Gemini. Waze says its updating its conversation reporting feature, first introduced in 2024, […]
Like Google Maps, Waze is going all-in on Gemini.
<pre style='white-space:pre-wrap;width:81ex'>fix(agents): honor claude-cli contextTokens setting (#93198) * fix: honor claude-cli configured context tokens * test: cover claude-cli native compaction budget * test: cover ordinary claude-cli budget persistence * test: migrate claude-cli budget fixtures to session accessor * fix(agents): carry CLI provider into context budget * fix(agents): guard optional CLI context metadata * fix(agents): align Claude CLI native compaction budget Co-authored-by:
<p> Local-first AI video editor for Mac </p> <p> <a href="https://www.producthunt.com/products/ave-ai-video-editor?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/1195006?app_id=339">Link</a> </p>
Find the best robot pool cleaner deal. Save 43% on the Aiper Scuba V3 at Amazon.
<h1> Three Strangers on the Internet Just Made My Research Better — Here\'s What They Taught Me </h1> <p>I published three articles last week about building verification infrastructure for AI agents. I expected silence. Instead, three engineers I\'ve never met left comments that changed how I think about my own work. This is what they taught me — and where the research goes next.</p> <blockquote> <p><strong>Series context</strong>: <a href="https://dev.to/yuhaolin2005/i-ran-150-tasks-to-test-if-
Agentic LLMs keep failing the same way because they lack specific, reusable capabilities. Stanford's TRACE diagnoses those gaps from an agent's own trajectories, synthesizes one verifiable training environment per capability, trains a LoRA adapter for each, and routes tokens across experts—improving τ²-Bench by +15.3 points and reaching 73.2% Pass@1 on SWE-bench Verified. The post Stanford Researchers Introduce TRACE: A Capability-Targeted Agentic Training System That Turns Recurrent Agent Failu
<p>The short answer: WebMCP is the part of agent commerce your customers' browser will handle — and <a href="https://ucpchecker.com/blog/mcp-vs-ucp-vs-ap2-whats-the-difference" rel="noopener noreferrer">UCP is the part that actually completes the sale</a>. If you want to know whether your store is ready for the agents arriving through both, <a href="https://ucpchecker.com/check" rel="noopener noreferrer">enter your domain here</a>. It checks the commerce layer — the half that decides whether an
Article URL: https://raymyers.org/post/zed-creator-calls-spade-a-spade/ Comments URL: https://news.ycombinator.com/item?id=48889637 Points: 182 # Comments: 106
<pre style='white-space:pre-wrap;width:81ex'>fix: keep command cron jobs operator-scoped [AI] (#106057) * fix: scope agent cron command access * test: avoid secret-like cron fixture text</pre>
<p>517 changes shipped across 30 tracked developer tools this week. 3 were critical. Here is what actually needs your attention.</p> <h2> Vercel AI SDK @ai-sdk/<a href="mailto:openai-compatible@2.0.58">openai-compatible@2.0.58</a> </h2> <p>Fixed a security vulnerability where streaming tool calls could finalize prematurely on partial JSON, causing tool execution with incomplete arguments. This affects any application using streaming tool calls with the OpenAI-compatible SDK.</p> <p><strong>Do th
<h2> Design Tradeoffs and Limitations </h2> <h3> Technical Design Tradeoffs </h3> <p><strong>Regex-Based Anonymization vs. ML-Powered Entity Recognition</strong></p> <ul> <li> <strong>Approach Taken:</strong> EV-000005 demonstrates that regex-based pattern matching successfully identified and scrubbed raw student names and institutional IDs from API payloads</li> <li> <strong>Tradeoff:</strong> While regex provides deterministic performance with zero latency overhead compared to external ML serv
Article URL: https://github.com/device-context-protocol/dcp Comments URL: https://news.ycombinator.com/item?id=48889594 Points: 2 # Comments: 0
<p>If you use <a href="https://claude.com/claude-code" rel="noopener noreferrer">Claude Code</a> for long, agentic tasks,<br> you know the moment: you're mid-refactor, the agent is cooking, and then — usage<br> limit. It stops. You come back an hour or two later, the reset has passed, and<br> you manually pick up where you left off.</p> <p>I wanted to just… walk away and have it resume itself. So I built<br> <strong><a href="https://github.com/OnamSharma/claude-supervisor" rel="noopener noreferr
<p>I run Claude Code most of my day. agent loops firing all day, one after another. the usage screen tells me I'm at 26% of my weekly limit. it doesn't tell me what that 26% weighs anywhere else. it feels free.</p> <p>It isn't.</p> <p>A single chat reply costs about 0.3 watt-hours — the energy an LED bulb burns in under two minutes. Simon P. Couch measured actual Claude Code sessions and got a median of 41 Wh — 130x a plain message. Simon Willison put his own daily habit at the equivalent of 4,4
<p>I wasn't looking for this. I was digging through Claude Code skill repos at 1am, the way you do, and I almost scrolled past a small project called <strong>Guardsman</strong>. Two stars. Eighteen commits. No flashy benchmark chart at the top of the README.</p> <p>That absence is exactly what made me stop scrolling.</p> <h2> The pattern I'm tired of </h2> <p>Every "AI coding skill" README follows the same script now: a bold claim in the title ("73% fewer tokens!", "10x faster shipping!"), a cha
<pre style='white-space:pre-wrap;width:81ex'>fix(codex): do not promote missing_tool_result on abort/interrupt (#104955) * fix(codex): do not promote missing_tool_result on abort/interrupt Keep synthetic tool.result rows for transcript integrity, but skip promoting the missing-tool invariant to promptError when the turn is already aborted or interrupted so mid-exec cancel paths keep abort classification and assistant text (#104898). * fix(codex): preserve explicit abort outcome --------- Co-auth
<p>NeuroOffice is a suite of AI agents built into Y-tech Bank’s business layer. Here’s the architecture we’re building and the tradeoffs we’re making- including what works today and what’s still ahead.</p> <p>The core challenge: several specialized agents, one shared data foundation. Each agent needs different data, has different latency requirements, and produces different output. But from the user’s side it should eventually feel like one coherent system, not several separate chatbots.</p> <p>
<!-- SC_OFF --><div class="md"><p>Anthropic’s Jacobian Lens work introduced a way to inspect verbalizable representations inside language models. Follow-up experiments suggested that entropy in this internal “workspace” might help identify confidently incorrect answers.</p> <p>I tested that hypothesis on Qwen3-4B across ~11,400 examples from seven distinct datasets, including TriviaQA, PopQA, NQ-Open, TruthfulQA, HotpotQA, GSM8K, and CommonSenseQA.</p> <p>Three main findings:</p> <ol> <li>It can
Strap in. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.
PARTNER CONTENT: Firms crafting internal AI must choose a permanent residence for their tech, not just rent server space by the hour.
Anthropic is keeping Claude Fable 5 in its subscription plans through July 19, 2026. The model was supposed to switch to pay-per-use today. Subscribers can use up to 50 percent of their weekly limit for Fable 5. The extension is likely a response to pricing pressure from OpenAI's GPT-5.6 Sol and cheaper competing models. The article Anthropic extends free Fable 5 access for subscribers as OpenAI's GPT-5.6 Sol heats up the pricing war appeared first on The Decoder .
Article URL: https://www.reuters.com/technology/artificial-intelligence/ Comments URL: https://news.ycombinator.com/item?id=48889232 Points: 1 # Comments: 0
Article URL: https://www.youtube.com/watch?v=OuEKdD_1F8s Comments URL: https://news.ycombinator.com/item?id=48889199 Points: 1 # Comments: 0
<p> The sidebar agent for the rest of us </p> <p> <a href="https://www.producthunt.com/products/webbrain?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/1194941?app_id=339">Link</a> </p>
https://seaticket.ai/ Comments URL: https://news.ycombinator.com/item?id=48889184 Points: 1 # Comments: 0
Article URL: https://jacobin.com/2026/07/ai-big-tech-global-ownership-control Comments URL: https://news.ycombinator.com/item?id=48889175 Points: 1 # Comments: 0
Prime Intellect launched verifiers 0.2.0, previewing a rewritten "v1" core under the verifiers.v1 namespace. It splits an environment into a taskset (what), a harness (how), and a runtime (where), with an interception server that proxies requests and records training-ready traces. Any taskset runs under any compatible harness, with full prime-rl training support at launch. The post Prime Intellect Releases Verifiers v1: Composable Tasksets, Harnesses, and Runtimes for Agentic RL Training and Eva
Lawsuit alleges job-swappers took secrets with them, helped by coaching on how to avoid scrutiny
