AI News Today

Article URL: https://claude.ai/public/artifacts/725f961b-09dc-4a66-8dac-8fefeeb69a1f Comments URL: https://news.ycombinator.com/item?id=49050803 Points: 1 # Comments: 0

<p> Review-first AI cleanup for your real Gmail inbox. </p> <p> <a href="https://www.producthunt.com/products/purebox-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/1206492?app_id=339">Link</a> </p>

<p>Una sesión de agente que sobrevive a cerrar la laptop, corriendo en una sola instancia EC2 chica que manejas desde el teléfono. Este post cubre tanto Claude Code como Kiro CLI, porque resuelven el problema de "alcanzar la sesión desde otro lado" de maneras completamente distintas, y esa diferencia decide la mayor parte de tu arquitectura.</p> <h2> TL;DR </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th></th> <th>Versión ingenua</th> <th>Después</th> </tr> </thead> <tbody> <t

<h1> MCP in enterprise: access control and audit logging </h1> <p>Ship an MCP server at a company with compliance requirements and you will hit the same wall every team hits: the protocol connects your AI agent to tools beautifully, and it gives you almost nothing to control what that agent is allowed to touch once it is connected.</p> <p>That gap is not a bug. MCP was built to solve connectivity, not governance. But if your agent touches a database, a ticketing system, or a support inbox with r

I got up and was switching between Airbnbs, already annoyed at the Airbnb app sending me back and forth to Safari and Google Maps. Safari even warned me: “You are leaving airbnb.com.” Later, I applied my agents to my Safari history. It showed how many times I edited a flight booking, when my iPhone left an Airbnb, and that I had requested wheelchair assistance. That last detail was easy to explain: I was recovering from a surfing injury. That became Pastport—not another chatbot for browser histo

<p>A user opened an issue on my open-source video tool last week that named a gap I had been shipping around for months.</p> <p>He runs lectures through crv so an LLM can read them. One 22-minute lecture: 1,377 candidate frames extracted, 60 kept after dedup and <code>--max-frames</code> thinning. The 60 frames come out in the right order. That is all they come out with.</p> <p>His complaint, in one line: the LLM can describe the slide, but it cannot tell you when the slide was on screen.</p> <p

<pre style='white-space:pre-wrap;width:81ex'>fix(tui): preserve case-sensitive Matrix and Signal conversations (#113800) * test(tui): reproduce opaque provider session identity loss * fix(tui): preserve case-sensitive provider session identities * test(tui): cover cross-agent opaque session identities --------- Co-authored-by: Peter Steinberger <steipete@golden-gate.local></pre>

"This would be the first detection." 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/slvDev/esp32-ai Comments URL: https://news.ycombinator.com/item?id=49050512 Points: 161 # Comments: 34

Article URL: https://www.apertus-ai.org/articles/2026-07-apertus-1-5/ Comments URL: https://news.ycombinator.com/item?id=49050483 Points: 3 # Comments: 2

<pre style='white-space:pre-wrap;width:81ex'>fix(browser): isolate copilot chat events by active turn (#113781) Co-authored-by: Peter Steinberger <steipete@golden-gate.local></pre>

<pre style='white-space:pre-wrap;width:81ex'>fix(models): read Anthropic token limits in live discovery (#113784) Anthropic reports max_input_tokens and max_tokens, which the live catalog normalizer never looked for, so discovered Claude models fell back to a 128k/8k default. Append both names to the existing lookups, keeping completion-specific keys ahead of the generic max_tokens. Co-authored-by: Claude Fable 5 <noreply@anthropic.com></pre>

<pre style='white-space:pre-wrap;width:81ex'>refactor(agents): split embedded session lock lifecycle (#113779) * refactor(agents): split session lock lifecycle * refactor(agents): tighten session lock modules * fix(agents): import session publication type</pre>

Google's Vice President of Devices and Services, Shakil Barkat, all but confirmed in an interview with 9to5 Google that its next Pixel phone would cost more than the Pixel 10. Considering the ongoing RAM supply issues due to the explosion of AI data centers, the rumored price hike is not a complete surprise. Companies from […]

Hey HN! A while back I asked Codex to generate a companion site for a book I was enjoying. It ended up generating a landing page for the book that was ...actually good? I went down a bit of a rabbit hole prompting different models to generate landing pages for different books and this is the result. The site runs you through a series of blind "taste" tests to help you figure out which model's designs you prefer. Would love any feedback you have if you try it out! Comments URL: https://news.ycomb

Maybe a dumb question but I don't think I've ever experienced more down/waiting time in my career so I'm much more prone to opening YouTube, HN, etc. despite not really wanting to. Most likely the answer is to just sit and do nothing. But have you discovered an approach to making agentic coding, etc work for you without feeding into distraction and attention deficit? Comments URL: https://news.ycombinator.com/item?id=49049986 Points: 3 # Comments: 1

Article URL: https://robotclub.substack.com/p/icymi-figure-hits-1000-humanoids Comments URL: https://news.ycombinator.com/item?id=49049977 Points: 1 # Comments: 0

This second installment explores how Ray’s higher-level libraries—Serve, Data, and Train—abstract the complexities of running AI workloads on Google's TPU slices. Ray Serve uses a simple topology configuration to correctly gang-schedule large multi-host models, while Ray Data eliminates data-loading bottlenecks by feeding accelerators directly with native JAX batches. Finally, JaxTrainer streamlines distributed training across TPUs by automatically handling cross-slice coordination, checkpointin

Comments URL: https://news.ycombinator.com/item?id=49049960 Points: 1 # Comments: 0

Article URL: https://zenodo.org/records/21566453 Comments URL: https://news.ycombinator.com/item?id=49049884 Points: 2 # Comments: 0

Related to this thread [1] which Jabber clients not only detect MitM tampering when a valid cert is used in the middle but is not the cert on the server, meaning an entity obtained a certificate, used it to MitM the connection and the client not only rejects this alternate valid certificate but also alerts the user to the MitM. XEP-0474 SASL SCRAM Downgrade Protection (Experimental) [2] Claude does not seem to know and I can't find any clarifying documentation, just lots of open issues. The purp

If someone has made a free-to-use tool that is open-source and legitmately helpful (and not AI slop). How do I get it to people? I have already posted it here on HN, on subreddits, on LinkedIn, and some forums. Where else can I post (for free)? What other social media platforms would be good for this? For context I have two tools: one is robotics related and the other is ML related. Comments URL: https://news.ycombinator.com/item?id=49049662 Points: 2 # Comments: 10

I feel like in the past, software jobs like programming, IT and academic fields like computer science, created some identity -- these fields required deep understanding of technical details, which attracted a specific type of people, and thus allowed a certain type of culture to be created, what's called the hacker/nerd/geek culture -- and now in the age of ai, this technical knowledge is much less relevant. What will happen to this culture? Comments URL: https://news.ycombinator.com/item?id=490

Article URL: https://trytongue.com/demo Comments URL: https://news.ycombinator.com/item?id=49049282 Points: 1 # Comments: 1

Article URL: https://github.com/johndebord/protoai Comments URL: https://news.ycombinator.com/item?id=49049253 Points: 2 # Comments: 1

Article URL: https://github.com/eliasschwalme/avatar-cursor/blob/main/demo.gif Comments URL: https://news.ycombinator.com/item?id=49049186 Points: 1 # Comments: 0

Article URL: https://github.com/KubiKDB/AI-Tools-For-Learning Comments URL: https://news.ycombinator.com/item?id=49048863 Points: 3 # Comments: 0

At libraries around the country, "Avoiding AI" workshops have elicited unprecedented demand.

Ringup is built to simplify voice payments. My hypothesis is Voice AI will lead to an explosion in people transacting over phone calls. Ringup helps consumers save their card details once and use it everywhere. It helps merchants get payments directly to their processor. And it helps developers by reducing integration complexity. Try it if you're a developer building Voice AI. Comments URL: https://news.ycombinator.com/item?id=49048730 Points: 3 # Comments: 2

I have realized - some of the premier american 'a.i' labs are headed by people without manufacturing expertise - of | or how things are made in the real world. I'm not doubting their experience as world class researchers - but the research lab is not the real world - with complex dynamics. all their 'models' are solutions seeking problems. but it doesn't mean the models are not useful for some tasks. before releasing models - they have never stopped to ask - what are we working towards or what a

What happens when we do compress the context in claude code? How does the reduce the context with out loosing the details? Comments URL: https://news.ycombinator.com/item?id=49048571 Points: 3 # Comments: 4

<p> AI specialists that join your Google Meet and gives feedback </p> <p> <a href="https://www.producthunt.com/products/gstack-joins-your-meeting?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/1206376?app_id=339">Link</a> </p>

In the age of AI, the things that matter take a minute, while everything else takes forever. It's great, we achieve wonderful things that matter instantly. But for the rest of day we deal with shit. Comments URL: https://news.ycombinator.com/item?id=49048491 Points: 4 # Comments: 0

The Hidden Cost of Frictionless Technology Taotuner — 2026 I recently asked an AI assistant for restaurant recommendations. It listed five places, complete with ratings, distance, and popular dishes. I picked one, went, had a fine meal. Later, when a friend asked what I'd eaten, I couldn't remember the name of the restaurant. The experience wasn't bad. It was frictionless. And it left almost no trace. We've built tools that eliminate the gap between want and satisfaction, question and answer, cu

Inspired by Crypto Pool Mining: would it be possible to train LLMs as a pool of volunteers? I feel like that would be a good way to have frontier models that are open and don’t rely on big corporations with data centers to train and release them… Comments URL: https://news.ycombinator.com/item?id=49048332 Points: 2 # Comments: 2

Architecting cost-effective infrastructure by navigating the latency and storage trade-offs of HNSW, SPANN, and DiskANN The post How to Optimize Vector Search When RAM Gets Too Expensive: On-Disk vs. In-Memory ANN Indexes appeared first on Towards Data Science .

What really happens behind the scenes when we select high effort vs medium effort for the same LLM models? Comments URL: https://news.ycombinator.com/item?id=49048125 Points: 10 # Comments: 7

Article URL: https://tobi.knaup.me/2026-07-25-open-weight-ai-is-having-its-kubernetes-moment/ Comments URL: https://news.ycombinator.com/item?id=49048034 Points: 160 # Comments: 108

What does a generative AI company want with one of the most popular horoscope apps? A lot, as it turns out.

I'm looking for some concrete examples of any AI advancements actually causing someone to financially prosper where they didn't before. Most of the examples I can think of make someone or something slightly more streamlined or productive, and so I'm looking for cases where company X is now Y% more profitable because of AI advancements. Comments URL: https://news.ycombinator.com/item?id=49047714 Points: 6 # Comments: 0