Article URL: https://rebecca-powell.com/posts/meat-ai-universe-test-harness/ Comments URL: https://news.ycombinator.com/item?id=50139275 Points: 2 # Comments: 0
A new benchmark called Taste-Bench tests how well AI agents make strategic choices during long-running tasks. Even the best model, GPT-5.6 Sol, picks the right path on just under 60 percent of simple binary questions. Thinking longer doesn't help. The article AI agents have bad taste in long-term decisions and a new benchmark proves it appeared first on The Decoder .
<!-- SC_OFF --><div class="md"><p>I built GPT-2 small and a KV cache in PyTorch as a learning project, then compared the mathematical savings with measured latency and memory. I have published the implementation and a detailed write-up, and would appreciate feedback on the methodology.</p> <p>Setup: pretrained GPT-2 small (124M), NVIDIA A40, fp32 with TF32 off, batch size 1, greedy decoding. Timings are bracketed by torch.cuda.synchronize(), after warmup, using the median of five runs per contex
Article URL: https://abolish-ai.net/abolish-ai-a-manifesto/ Comments URL: https://news.ycombinator.com/item?id=50139072 Points: 3 # Comments: 0
Six AI agents were tasked with recovering £350 (about $465) for a canceled train ticket in an experiment. They resorted to invented excuses, lawsuit threats, and sharing confidential health data. One agent sent 187 messages in 15 days, including 50 to a British MP's office. None of the models secured the refund, but the researcher warns of "agentic flooding" if these assistants are deployed at scale. The article AI agents chasing a $465 train refund invented excuses, leaked medical data, and spa
Google Research released Harness2, an Apache-2.0 framework that rewrites an agent's harness at test time. One refinement step lifted GLM-5 by 11.5 points on WorkBuddy-Bench with no weight updates. The post Google Open-Sources Harness2, a Recursive Agent Harnessing Framework appeared first on MarkTechPost .
Pop quiz: somebody left a command shell logged in. What do you do?
Hi HN, I built Connecting Captions, a live captioning and translation overlay for macOS that runs entirely on-device. Nothing goes over the network: no API calls, no audio leaving your machine, and it keeps working offline. I wanted live translation I could use in meetings and conversations without worrying about connectivity, latency spikes, or sending sensitive audio to a third party. The technically interesting part is a low-latency, self-correcting pipeline for speech-to-text and translation
Article URL: https://agent-blackbox-five.vercel.app/demo Comments URL: https://news.ycombinator.com/item?id=50138752 Points: 1 # Comments: 0
Article URL: https://www.nature.com/articles/d41586-026-03194-w Comments URL: https://news.ycombinator.com/item?id=50138702 Points: 1 # Comments: 1
Article URL: https://weightsai.net/ Comments URL: https://news.ycombinator.com/item?id=50138672 Points: 1 # Comments: 0
Article URL: https://www.ft.com/content/9ea5a37d-15c9-4d5b-94d6-719048c9cd3c Comments URL: https://news.ycombinator.com/item?id=50138511 Points: 2 # Comments: 0
Article URL: https://www.ft.com/content/f7c26ac0-f5f2-443a-86b2-39d62f7920a9 Comments URL: https://news.ycombinator.com/item?id=50138318 Points: 1 # Comments: 0
In this tutorial, we implement Structured Data Models, NVIDIA’s new library that puts several pretrained foundation models for tables behind one PyTorch interface, and use it to make predictions without training. The library wraps TabICLv2 and KumoTabular for single tables and KumoRelational for whole relational databases, and each of them answers by in-context learning: we […] The post How to Make Zero-Training Tabular Predictions with NVIDIA Structured Data Models appeared first on MarkTechPos
Article URL: https://www.turboscribeai.org/ Comments URL: https://news.ycombinator.com/item?id=50138286 Points: 1 # Comments: 0
Article URL: https://octomon.dev/blog/ai-slop-and-ai-gourmet/ Comments URL: https://news.ycombinator.com/item?id=50138046 Points: 1 # Comments: 1
FOSS, Basically it plugs into your existing coding agent workflow and every time you push to prod it'll throw up a questionnaire that checks if you understand how the code you're pushing works. Comments URL: https://news.ycombinator.com/item?id=50137950 Points: 128 # Comments: 32
New research can help AI models understand crystalline materials in 3D and reason about their properties.
Article URL: https://www.malwarebytes.com/blog/ai/2026/10/phishing-emails-are-baiting-your-ai-assistant Comments URL: https://news.ycombinator.com/item?id=50137768 Points: 3 # Comments: 0
Article URL: https://www.gmicloud.ai/ja/blog/best-free-gpu-cloud-options-for-ai-startups-and-researchers Comments URL: https://news.ycombinator.com/item?id=50137325 Points: 2 # Comments: 0
Article URL: https://www.cnbc.com/2026/10/16/ai-espionage-openai-jay-clayton.html Comments URL: https://news.ycombinator.com/item?id=50137246 Points: 2 # Comments: 0
Article URL: https://www.youtube.com/watch?v=kAkpzlR7io8 Comments URL: https://news.ycombinator.com/item?id=50136939 Points: 1 # Comments: 0
<p> Your Claude Code and Codex limits, live in the notch </p> <p> <a href="https://www.producthunt.com/products/ilha?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/1283885?app_id=339">Link</a> </p>
Article URL: https://defuzex.ai/blog/what-500-minimax-code-tests-revealed-about-ai-saying-done Comments URL: https://news.ycombinator.com/item?id=50136796 Points: 3 # Comments: 0
Article URL: https://noisypixel.net/drunken-goddess-reflux-public-demo-canceled-ai-imitation-concerns/ Comments URL: https://news.ycombinator.com/item?id=50136741 Points: 2 # Comments: 2
The new treatment kills off a sub-group of fat cells. (Jian Fan/iStockphoto/Getty Images) Positive signs from clinical trials. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.
Article URL: https://hardwareos.ai Comments URL: https://news.ycombinator.com/item?id=50136365 Points: 1 # Comments: 0
OpenAI CEO Sam Altman wants a humanoid that can operate in a world designed for people, though the CEO says a robot 'brain' is a much bigger challenge.
<!-- SC_OFF --><div class="md"><p>I'm sure many of us are slogging through impenetrable AI writing as we finish ICLR reviewing.</p> <p>I'll try to create an anonymized version from a paper I'm reviewing, though it's hard to match Claude on this:</p> <blockquote> <p>outperforms the best baseline costing no more at none of six operating points and a permuted placebo at none of three, with gains up to about 1.5 not excluded</p> </blockquote> <p>My question for the community is this: how do we call
I’ve been working on a Nix CSI for kubernetes since NixCon when I realized that everyone seems to working towards killing OCI layers (and images) in Kubernetes deployments. I’m hoping to get people to give it a spin and report back to me bugs, feature requests or general feedback. I don’t have a lot of data on how this scales and that makes it hard to decide whether to justify adding a feature or not (e.g. cache locality scheduling). An LLM was used in the development of this project. Comments U
So, my boss goes on a rampage. You can clearly see the joy and at the same time the rage and impatience. AI is the tool he always wanted. We are too slow. Process is out of the window. No need for sprints anymore. No planning, just doing. All people downstream must somehow adapt. Whenever i see now his automated messages in slack, my natural reaction is: why do i even bother reading that? Thats for machines to read. He does not need me? Clearly i am not made for that. What is he expecting me to
Article URL: https://primerkit.ai/ Comments URL: https://news.ycombinator.com/item?id=50135649 Points: 2 # Comments: 0
Article URL: https://vstorm-co.github.io/vstormos-landing/ Comments URL: https://news.ycombinator.com/item?id=50135614 Points: 1 # Comments: 1
<pre style='white-space:pre-wrap;width:81ex'>refactor(agents): trim redundant session and sandbox comments (#175793) * refactor(agents): trim redundant session and sandbox comments * refactor(agents): withdraw MCP-reachable comment slice</pre>
<p><strong>You upload a document, ask a question, and get an incomplete answer.</strong></p> <p>You check the source. The information is there.</p> <p>So why did the RAG application miss it? </p> <p>Sometimes, the problem starts with how the document was split.</p> <p>For Day 2 of my AI challenge, let’s look at how chunking affects RAG retrieval, when different approaches help, and where they can fall short.</p> <h2> The answer is there, but part of it is missing </h2> <p>Imagine a company polic
<pre style='white-space:pre-wrap;width:81ex'>refactor(agent): reuse retry sleep cancellation errors (#175858) Use shared sleep directly instead of catching its cancellation only to construct another AbortError. Keep signal exits, retry schedules, accepted-run cancellation/confirmation/fallback, idempotency, authority and ordinary cleanup intact. Nine net production lines removed; no tests changed. Complete independent review scoped-clean P0-P2/0.99, canonical focused type-aware lint/format/white
<p>Stock Expert AI is a free US stock research site. Most of the daily work behind it is done by 16 AI agents that run on schedules: a newsroom that writes stories from market data, an editor that rejects any number it cannot trace back to its source, a site doctor that checks the site every few hours, an analyst, and a handful of social desks. One human, the founder, signs off.</p> <p>The homepage shows them in a live 3D office on Wall Street. Click anyone and you see what they are working on r
<p><em>This is a submission for the <a href="https://dev.to/challenges/hacktoberfest-week2-2026-10-12">Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle</a></em></p> <h2> What I Built </h2> <p><strong>MerchPulse</strong> is an autonomous customer feedback intelligence and root-cause platform built specifically for e-commerce merchants (Shopify, Amazon, WooCommerce, and Etsy sellers).</p> <h3> The Problem </h3> <p>Small and mid-sized e-commerce sellers are flooded with scattered, messy f
<pre style='white-space:pre-wrap;width:81ex'>perf(gateway): omit session rows from hidden agent progress (#175852) Keep roster metadata with session events and preserve the run's session identity without rehydrating non-tool progress. Tool envelopes and native delivery routing remain unchanged.</pre>
<pre style='white-space:pre-wrap;width:81ex'>refactor: keep internal core helpers with their owners (#175787) * refactor(core): keep single-consumer helpers with their owners Keep read presentation, tool error state, provider warning projection, node-host worker helpers, and policy decisions in their existing production owners. Retire private forwarding seams while preserving security, lifecycle, prompt, and public contracts. * refactor(agents): keep descriptor preparation with finalization Remo
