AI News Today

<!-- SC_OFF --><div class="md"><p><strong>TL;DR:</strong> Most RL post-training algorithms optimize the current batch and move on. But after an update, did the new policy actually become better?</p> <p>We introduce <strong>Policy Improvement Reinforcement Learning (PIRL)</strong> and its practical implementation, <strong>Policy Improvement Policy Optimization (PIPO)</strong>—a plug-and-play framework that lets RL training look back, verify the previous update, and reinforce or correct it accordi

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. OpenAI called the Hugging Face attack unprecedented. But we’ve been here before. —Will Douglas Heaven, senior AI editor Reading OpenAI’s account last week of how some of its models broke their…

Anthropic CEO Dario Amodei is once again warning about the risks of open AI models while insisting he has never called for a ban. He argues that authoritarian states like China could overtake the US and that open models could be misused for biological or cyberattacks. Critics say he's mostly trying to protect his own business from cheaper competition. The article Anthropic CEO Amodei doubles down on open-weight risk stance while insisting he never called for a ban appeared first on The Decoder .

Discover five of the best AI tools for data analysis that can clean data, write code, create visualizations, and generate insights faster.

The idea that makes backpropagation possible. The post Backpropagation Explained for Beginners (Part 2): There Has to Be a Better Way appeared first on Towards Data Science .

The Model Context Protocol , the open standard that has quietly become the connective tissue between AI agents and the world's software, is getting its largest update since Anthropic released it twenty months ago — a sweeping architectural revision that its maintainers and backers say finally makes agentic AI ready for massive enterprise production deployments. The update, released today under the stewardship of the Agentic AI Foundation (AAIF) , a directed fund under the Linux Foundation , fina

For enterprise agents, less is more

Over the last few months, I've spent a lot of time talking to my computer. One underrated feature of the LLM revolution has been a remarkable leap in all kinds of dictation technology - even the fastest, cheapest models are getting very good at understanding and processing speech. I've tested lots of these apps, from […]

<!-- SC_OFF --><div class="md"><p>I'm really confused about what the point of the prompt injection was (speaking as an author). Is it just a study? I would really prefer that they took action against the AI-generated reviews. Obviously, we cannot assume that the reviewers were copy-pasting the output from the LLM without having given it any look at all, but in some cases that does look to be the case. In fact, in some cases the meta-reviewer seems to have also largely used LLMs. What exactly is

Public meeting over proposed AI bit barn ended with zoning approval, a physics teacher arrested, and fresh questions over whether clapping now counts as disorderly conduct

Researchers investigating Hugging Face found the majority of its models would generate non-consensual deepfakes.

Many people use Claude Design to craft UI mocks, and while this use case remains the most popular, the tool offers a few interesting… Continue reading on UX Planet »

"These neurons might directly respond to inflammation." ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

In 2023, the Times sued OpenAI and Microsoft for copyright infringement. They’ve since spent more than $20 million on the case, and publisher A.G. Sulzberger has no plans to stop fighting it.

<p> Free voice coding for Claude Code, Codex and more </p> <p> <a href="https://www.producthunt.com/products/ski?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/1208711?app_id=339">Link</a> </p>

As it promised last year, Baidu has started testing its robotaxis in London along with partners Lyft and Freenow.

<p>Like genies of folklore, AI agents take their instructions literally – to potentially disastrous effect. We must track their ability to do what we actually mean</p><p>In July, Hugging Face, a company that hosts much of the world’s AI software and open-source AI models, was hacked. A malicious dataset had been used to run code on one of its servers. Whoever was behind it captured internal security credentials and moved through systems over a weekend, running thousands of actions from a swarm o

<p>The new breed of camera-equipped streetlights has the potential to fight crime and track down missing persons. Or will it just take Britain one step closer to a surveillance state?</p><p>How would you feel if your neighbourhood lamp-posts could recognise your number plate, and also your face? Maybe you wouldn’t mind, given how long we’ve had automatic number plate recognition and CCTV. What about if that lamp-post could power itself through its own solar panels, and perhaps find another use f

Worried about AI pricing model costs rising out of your budget? Curious whether there's a way to run capable AI models without paying premium subscription fees every month? In this article, you'll discover how open-weight AI models work, what hardware and software are needed to run them locally, and how to choose the right model […] The post How Open-Weight AI Models Could Save Your Business Thousands appeared first on Social Media Examiner .

Even cockroaches have peer pressure. ScienceAlert stories are written, fact-checked, and edited by humans, never generated by AI. Don't miss a story, subscribe here.

Anthropic and OpenAI employees are expected to give generously after their companies go public. “It’s going to be a wild ride,” says one nonprofit leader.

Many of the AI tools we interact with take the form of text boxes. But what if there was a different way to interact with AI? Oleksii Hrzhehorzhevskyi explores a different approach to creating a new AI assistant and how designers can navigate the field as AI continues to change it.

An AI-generated heart-shaped Samsung phone is going viral, with video of the fake device racking up thousands of likes and shares.

Lee, an engineer at Samsung’s semiconductor division, clocks out when his shift ends. He used to work longer hours, going the extra mile to excel at his projects. But lately, he’s been coming straight home to work on his job application for the chipmaker’s South Korean rival SK Hynix, sharing tips with his coworkers on…

Hugging Face is being used to make nonconsensual deepfakes, and the popular open-source AI model repository is doing very little to prevent it. That's according to a new report published by the European nonprofit AI Forensics, which found that seven out of the top nine image editing models hosted by Hugging Face readily complied with […]

ChatGPT has been refusing to imitate the writing of famous authors, both living and dead.

Create AI images, videos, avatars, headshots, and photo edits with ARTA. A one-year Premium subscription is on sale for $19.97, down from $39.99.

Find the best Mini PC deal. Save 13% on Beelink SER5 Mini PC at Amazon.

STAR Labs has published a Linux kernel exploit that turns an ordinary local user into root on the CentOS Stream 9 build it targeted. The flaw, tracked as CVE-2026-53264 (CVSS score: 7.8), is a use-after-free race in the kernel's network traffic-control subsystem.Researcher Lee Jia Jie said artificial intelligence (AI) helped him find the bug and speed up exploit development. This is local

<p> Mission control for Claude Code & Codex in your menu bar </p> <p> <a href="https://www.producthunt.com/products/blackflare?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/1208599?app_id=339">Link</a> </p>

Baidu's Apollo Go autonomous vehicles will be available on Freenow, the mobility network that Lyft acquired in 2025.

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

TL;DR KotlinLLM is now public. It’s a research prototype for delegating runtime logic to an LLM from Kotlin code. Instead of calling an LLM on every request or running a separate agent, you can write an explicit Kotlin call. Its body is generated Kotlin source code, and that code is updated as your application hits […]

<!-- SC_OFF --><div class="md"><p>ML research is progressing at breakneck speed where frontier labs in both academia and industry have access to considerably large computes (GPUs). Where do small labs or independent researchers go in this context?</p> <p>Have you come across recent works in ML/DL and its applications (vision, language, speech, etc) where the work is good but it uses very limited compute? Maybe even single GPU workstations? In the past it was still possible, but, I am losing hope

<p> AI Visual Production Platform for Fashion Teams </p> <p> <a href="https://www.producthunt.com/products/caimera-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/1208527?app_id=339">Link</a> </p>

Hundreds of conversations with Anthropic’s chatbot were discovered as being publicly accessible.

Presented by Vijil In dynamic environments where users, data, workflows and attack techniques change continuously after deployment, AI agent trust has become a runtime problem. Most organizations still treat trust as a pre-deployment exercise, declaring an agent production-ready and launching it after it passes sandbox evaluations and performs successfully in security tests. Unfortunately, that trustworthiness breaks down the moment an agent begins interacting with the real world. "The core of t

Token costs not quite sorted, but on-prem will be an option

Microsoft has launched its first cybersecurity-specific model inside MDASH, its multi-model vulnerability identification and remediation harness. The company says MDASH, using MAI-Cyber-1-Flash and GPT-5.4, scored 95.95% on CyberGym. It also claims the configuration costs 50% less than its current best MDASH combination of GPT-5.4, GPT-5.4 mini, and GPT-5.3 Codex. Access is limited to approved

<p> Run agent apps inside your Claude or sell your own </p> <p> <a href="https://www.producthunt.com/products/lamoom?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/1208449?app_id=339">Link</a> </p>