AI News Today

arXiv:2608.20965v1 Announce Type: new Abstract: We define an atomic generation fact f=(u,tau,omega,z;rho), recording the origin, realized transformation, concrete occurrence, generated result and relation role. Compiled into a Generation-Fact Graph (GFG), these facts provide an AI-native, compilable scientific fact substrate preserving generation histories. We establish a GFG-based recursive scientific process in which analysis, intervention, replay and validation form facts for later cycles. Us

arXiv:2608.20980v1 Announce Type: new Abstract: Graph neural networks (GNNs) are routinely employed for short-range forecasting on multivariate time series with a spatial graph structure. Despite the availability of many alternative datasets, method innovations within this domain are predominantly assessed against a rather limited set of benchmark datasets, most notably Chickenpox, PedalMe, WikiMaths, METR-LA, and PEMS-BAY. The evaluation protocols contain baselines spanning from historical aver

arXiv:2608.20991v1 Announce Type: new Abstract: Graph Foundation Models (GFMs) on text-attributed graphs (TAGs) align graph representations with language semantics to support transferable graph learning. Despite these advantages, the backdoor vulnerability of GFMs on TAGs remains insufficiently understood, especially under graph-language alignment, where graph and text representations are trained to constrain each other in a shared semantic space. Existing backdoor attacks mainly target either t

arXiv:2608.21024v1 Announce Type: new Abstract: Modern neural network training increasingly uses matrix-aware optimizers, yet their conditioned matrix step is typically added directly to the weight, jointly changing its norm and direction. This interaction matters because the current norm determines angular motion, while directional learning can drive norm growth and thereby alter later steps. We introduce RODE, which gives the radial and directional components separate update rules and step siz

arXiv:2608.21057v1 Announce Type: new Abstract: Agentic large language model (LLM) systems are reshaping scientific workflows in chemistry and drug discovery, but evaluating their open-ended, tool-augmented outputs remains a fundamental bottleneck. Reference-based metrics such as BLEU and ROUGE fail to capture semantic correctness, while expert human evaluation does not scale to the iteration speed these systems demand. The LLM-as-a-Judge paradigm has emerged as a scalable alternative, but exist

SPONSORED FEATURE: Embedding AI at communication sites to empower energy optimization, power grid stability control and low-carbon development

The Information reports that Nvidia is discussing an equity investment in Perplexity at a valuation above $30 billion. Perplexity's annualized revenue has reportedly passed $750 million, up from less than $250 million at the start of 2026. Separately, Bloomberg says SoftBank plans a record ¥1 trillion, or $6.3 billion, retail bond to help repay the bridge loan behind its OpenAI stake and fund more AI deals. Put together, the signals expose next week's capital split: the chip supplier is moving t

Multimodal large language models increasingly use visual chain-of-thought (Visual CoT) to reason about spatial, temporal, and embodied environments. By generating intermediate reasoning images, Visual CoT provides an intuitive mechanism for visual foresight but introduces substantial inference overhead, which is particularly problematic for proactive video reasoning. We ask whether models can learn to think visually during training while reasoning directly at inference. We introduce Internalized

<p><strong><a href="https://www.ft.com/content/5ee49718-c258-4f01-aa32-7e5b76ae5245">Anthropic’s best AI model struggles to attract users as cheaper tools thrive</a></strong></p> A few interesting numbers in this FT story gathered from "people with knowledge of the matter":</p> <ul> <li>Anthropic's "annualized revenue" for July is up to $65bn - it was $47bn in May, and I collected <a href="https://simonwillison.net/2026/May/29/anthropic/">more historic numbers here</a>.</li> <li>Anthropic expect

Another notable step forward. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

<p> The AI runtime that remembers, learns, and acts everywhere </p> <p> <a href="https://www.producthunt.com/products/dropstone-2?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/1230136?app_id=339">Link</a> </p>

<a href="https://www.theguardian.com/commentisfree/picture/2026/aug/23/ella-baron-on-ai-and-the-drought-cartoon">Continue reading...</a>

"We've demonstrated a principle that wasn't previously available." ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

The domain name is evaluated as over $10k by Gemini, so I don't want to blow it. And I never done that. Comments URL: https://news.ycombinator.com/item?id=49409177 Points: 2 # Comments: 0

Article URL: https://github.com/majesticlabs-dev/writing-eval Comments URL: https://news.ycombinator.com/item?id=49409152 Points: 1 # Comments: 0

Sometimes I start several coding agents at the same time, each working on a different task. I just let them go and don’t watch them closely. The problem is that I may change my mind later but forget that one of the agents is still working toward the old goal. A couple of hours later, it submit a patch with all tests passing and reports that the job is done. But when I check the result, I find that it has finished implementing a plan I had already abandoned. Has anyone else who uses multiple codi

A while back I posted about a project I was working on for a real-time voice agent that runs ads on Meta to build conviction and prioritize side gig/business ideas ( https://news.ycombinator.com/item?id=48900788 ); you yammer away and it automates market research, problem statements and hypotheses. I've been working with some folks from HN on the app since then. What we've been focused on building is a modern solo-founder GTM tool -- figuring out how to triangulate aspiration, problems only you

Article URL: https://ericpardee.github.io/fire-hd-ownership/ Comments URL: https://news.ycombinator.com/item?id=49409073 Points: 3 # Comments: 0

<p>Fears of a datacenter buildout debt crisis are exaggerated. The risks are different than in the past and they are recoverable</p><p>Some experts are warning of a looming “debt bomb” crisis because big <a href="https://www.theguardian.com/us-news/datacenters--us-">datacenter</a> builders such as <a href="https://investor.atmeta.com/investor-news/press-release-details/2025/Meta-Announces-Joint-Venture-with-Funds-Managed-by-Blue-Owl-Capital-to-Develop-Hyperion-Data-Center/default.aspx">Meta</a>,

Article URL: https://www.froging.ai Comments URL: https://news.ycombinator.com/item?id=49408724 Points: 1 # Comments: 0

One of the reasons AI is so "good" at software development is that it is trained on a snapshot of a large corpus of already solved software problems. In the future, wouldn't that make anyone who produces original code become more cautious of sharing it, being aware that it will just be fed into the "hive mind" and they'll never get any attribution or financial gain out of it? Or has the value of code, tools and software become so cheap it doesn't really matter anymore? Comments URL: https://news

A message across generations. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Article URL: https://github.com/ptmrio/harness-subagent Comments URL: https://news.ycombinator.com/item?id=49408449 Points: 3 # Comments: 0

While a digital twin does not necessarily replace a systems integrator, it can help manufacturers evaluate robots and plan applications, says RoboDK. The post Use a digital twin to explore automation before committing capital appeared first on The Robot Report .

Article URL: https://github.com/tooppoo/enozunu Comments URL: https://news.ycombinator.com/item?id=49408038 Points: 5 # Comments: 0

An incredible natural experiment. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Like a neural 'reset'. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Hey everyone, i am building an ai powered finance manager called procura, it uses ai to find the cheapest listing of the product your business need and purchase it for you. i am looking to charge $500/mo for it. is it worth it or not. Note:- it does not handle the payments Comments URL: https://news.ycombinator.com/item?id=49407193 Points: 1 # Comments: 1

<p>Chris Lehane tells Guardian of need to implement new safety standards as critics say AI firms acting ‘recklessly’</p><p>A senior leader at OpenAI has said people should prepare to defend against “ongoing, persistent” cyber-attacks from AIs, as cutting-edge artificial intelligence models gain advanced capabilities to plan and launch offensives.</p><p>The leading AI company this week announced a pause in development of its most advanced internal models amid rising safety fears, and Chris Lehane

GLP-1 wins again. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

<p> Catch agent failures your evals miss </p> <p> <a href="https://www.producthunt.com/products/agnost-ai?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/1229731?app_id=339">Link</a> </p>

<pre style='white-space:pre-wrap;width:81ex'>fix(agents): restore system-agent delegation in channel turns (#128084) * fix(agents): isolate probe runtime generation * test(qa): prove channel system-agent delegation --------- Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com> Co-authored-by: RoboClaw <309084314+roboclaw-bot@users.noreply.github.com></pre>

<p> Control your company's internal AI agents and tools </p> <p> <a href="https://www.producthunt.com/products/decawork?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/1229726?app_id=339">Link</a> </p>

<h1> Model Context Protocol (MCP): Building an AI-to-API Bridge </h1> <h2> 1. What Is MCP? </h2> <p><strong>Model Context Protocol (MCP)</strong> allows an AI assistant such as Kiro, Codex, Claude, or another MCP-compatible agent to interact with external systems in a structured way.</p> <p>A useful mental model is:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight plaintext"><code>AI Agent --> MCP Client --> MCP Server --> External API / Database / Application </code></pr

<pre style='white-space:pre-wrap;width:81ex'>feat(mantis): add long-held active turn recipe (#128117) Capture the proven mechanics from proof run 32616859410 (PR #127950) as a reusable recipe: dual 600s timeout config patch, Code Mode exec events file, streamed chunkDelayMs hold, and the observe-loop pattern around the 60s cap. Require agents to record recipe-suggestion.md after scenario-mechanics iteration so future discoveries persist.</pre>

<p>Hey everyone, Harun here again. I’m the 12-year-old solo dev from Anthiyur, India, building an AI coding mentor (KODA) entirely on my POCO C55 Android phone.</p> <p>Yesterday, I posted about getting my first 5 stranger users. But today, I did something every founder needs to do: <strong>I went on a Supabase Stakeout.</strong></p> <p>I opened my <code>chats</code> table to see what people were actually asking my AI when they thought nobody was watching. And I found a session from a user named

<div class="ltag__link--embedded"> <div class="crayons-story "> <a href="https://dev.to/bhagyaprasad92/i-was-tired-of-sos-apps-failing-in-dead-zones-so-i-built-a-pure-dart-100-offline-tflite-crash-2cck" class="crayons-story__hidden-navigation-link">I was tired of SOS apps failing in dead zones, so I built a pure Dart, 100% offline TFLite crash detection engine.</a> <div class="crayons-story__body crayons-story__body-full_post"> <div class="crayons-story__top"> <div class="crayons-story__meta"> <

<p>Every heavy Claude Code user eventually hits the wall: the 5-hour rate limit block.</p> <p>I'm Lily. I lost my income overnight to a company-initiated layoff, rebuilt my setup starting that same month, and got back to ¥1.2M/month in revenue within six months.</p> <h2> Why This Setup Works </h2> <p>To most users, a rate limit block looks like a simple "pause." Grab a coffee, wait it out. The reality is far more destructive. Your Claude Code session dies during the block, and <strong>thousands

<p>Current MoE serving pipelines treat every decode worker as interchangeable, assuming that equalizing request counts suffices to minimize latency. In practice, the weight‑loading cost of each request varies wildly because different inputs activate disjoint expert subsets, and ignoring this fact inflates inference time.</p> <p>Prefill‑decode (PD) disaggregated serving has historically relied on routers that balance only the number of pending tokens per worker, without considering which experts