OpenAI用模型挖PHP高危漏洞

RT @ChrisMurphyCT: Ukraine was winning the war, so Trump kneecapped Ukraine to help Russia. I wish it weren’t that simple, but it is. Our…

OpenAI models helped researchers—operating as part of our Patch the Planet initiative—discover critical security flaws in PHP, an open-source programming language used to run an estimated 70% of websites. Over the past two months, we led a coalition of maintainers, researchers, and security providers to prepare fixes and protections ahead of disclosure. This duration was necessary because the fix was quite intricate and required a major change to the PHP engine. Every site using PHP can now safely upgrade their version to stay secure, in addition to applying mitigations from various security vendors. Due to the severity of these flaws and PHP’s widespread use, we consider this bug to be one of the most impactful vulnerabilities in recent memory. It also marks a new phase, where AI can find security bugs with internet-scale impact. This is an example of the Defender’s Window in action: to our knowledge, frontier AI was able to find this vulnerability ahead of attackers, giving defenders time to prepare. https://t.co/TW3jrSbB3f

Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic. These languages power software that banks, governments, and businesses depend on. Progress here means understanding decades of business logic and helping developers modernize those systems while preserving what works. We’re building these capabilities around the challenges developers and enterprise customers bring us. If your team is working on these systems, tell us which languages and workflows we should tackle next. https://t.co/2TbnkQixJ1

We’ve acquired Orchestra Bio, a platform that helps biopharma teams coordinate the complex process of drug development. Together, we’ll build on Rosalind, our AI models and tools for life sciences research, to streamline those operations and help new treatments reach patients sooner. Welcome @ashokaraj_ and team! https://t.co/vi51OhznQn

If you are wondering whether your website is affected, we have packaged a Codex skill to help check. Point your Codex at this repo to guide you through it: https://t.co/XUEbOwXuQC

I am very grateful Aleksander accepted our request to come take this on; the safety and security challenges in front of us as we confront superintelligence are immense, and he has an excellent set of skills to work on this.

RT @jonlovett: Maybe the most disgusting political ad in living memory. And then this statement by the campaign embracing the message - ju…

RT @OfficialLoganK: In just the first 72 hours since Gemini 4 Argon launched, we processed ~25 trillion tokens, the demand for intelligence…

RT @JustinWolfers: Washington has discovered a brilliant financial strategy. Cut benefits for people who need them, cut taxes for people wh…

In 1994, software was dead. People told me programming was for nerds and that I should become a doctor or a lawyer instead. In 2004, software was dead. People told me that someone from India would take my job for $5/hour. In 2014, software was dead. People told me that no-code was going to kill every job and doom my career. In 2024, software was dead with Devin, Windsurf, Copilot, and OpenAI o1. We kill software every 10 years, and yet the demand for good Software Engineers has never been higher.

GPT-Rosalind surfaced a new lead for drug discovery: could restoring cholesterol transport protect neurons from degeneration? DCTN1 is a gene linked to Perry syndrome, a rare neurodegenerative disorder. In mouse-brain data, disrupting it increased expression of genes involved in making cholesterol. GPT-Rosalind suggested a possible explanation: cholesterol may not be reaching where neurons need it. That gives scientists a specific mechanism to test and, if confirmed, a direction for developing new treatments for neurodegenerative disorders. With @perturbai_tx and @scrippsresearch, a team of thousands of GPT-Rosalind agents worked in parallel, connecting gene-expression results from millions of mouse brain cells with scientific literature in just 32 minutes. We’re now expanding this collaboration to identify new drug targets. https://t.co/oHdF3ELsMj

RT @TVietor08: Hey @ADL who used the term recently? Anyone notable?

RT @GoogleDeepMind: Gemini 4 Argon built a 3D periodic table synced to Tom Lehrer’s song, “The Elements.” Created in @Antigravity, it anim…

RT @justinbaragona: "He was never convicted" because he jumped bail and fled the state, leading to the judge issuing an arrest warrant agai…

RT @SteveRattner: Trump has raised $330 million from private donors for his White House ballroom. About half have received federal contrac…

OpenAI models are coming to Google Cloud. We're bringing our frontier models and Codex agentic workflows to Google Cloud's Model Garden, so enterprises can use our technology where they already build. https://t.co/gjuLVALjDm

We built a robotics data engine with Astra-level performance at 125x lower cost. And we are open-sourcing the weights today! Introducing Flora, a suite of open models to parse the physical world and turn raw footage into rich training signals. Flora already powers our in-house data engine, labeling 200K hours of videos a day on 500 GPUs. We created it to scale the frontier for robot learning. Now you own the weights too. Flora gives you: - High-quality, timestamped captions to curate and filter human egocentric and robot data; - Dense progress rewards for quality control and RL post-training; - Rubrics scoring for real-world benchmarking Try it today, links in thread:

RT @svpino: This will make language models generate text much faster. Normally, a model generates one token at a time. But with this tool…

This will make language models generate text much faster. Normally, a model generates one token at a time. But with this tool, a smaller helper model guesses several upcoming tokens, and the main model checks those guesses together. When the guesses are right, you get several tokens in one pass. They made the code available, and you can run this with vLLM and SGLang.

RT @AdamHoltererer: Gemini is the funniest model https://t.co/m4NPJGByN0

RT @Dogi1313: ELON FANS WON'T LIKE THIS ONE... BUT HERE ARE 10 BIG THINGS ELON PROMISED THAT NEVER HAPPENED: The New York Times went thro…

RT @Fritschner: Vivek Ramaswamy is now accusing Acton of lying about childhood trauma including sexual abuse she experienced as a 12-year-o…

RT @samstein: Christie with a conspicuous RT https://t.co/AY4GtWYIwq

RT @kyutai_labs: Pocket TTS has been getting updates those last few days. While it was ~100M parameters, we release smaller versions, down…

RT @KenRoth: It speaks to the unchecked power of the American presidency that a single warped occupant of the Oval Office can transform the…

RT @AndrewJTobias: "Byrnes told CNN he obtained the video from a voter who saw it as part of a polling call and recorded it."

I still remember when the goal was to make these models sound like people. And now we are in the "emotion" phase. There's absolutely no way I can tell these voices apart from real people. Listen to the examples and how easy it is to generate emotional text by describing how the voice should sound. You can simply ask for "embarrassed," and the voice will change to match.

A programmer and a software engineer aren’t the same. I’ve known many people who made a living writing code, but they were always a cog in a larger process they never understood. Frontier models can do that job today. 1000x better. Software engineers can write code, but they always understood code was a means to an end. Software engineering is alive and well.

Here is an open-weight coding model you can run locally on your computer. It's a 40B parameter model with a mixture-of-experts architecture. It activates 2.5B parameters per generated token. 512K context window. This thing runs on regular computers! 125 tokens/sec on an M4 MacBook Pro. Good alternative to closed models, but make sure you check their usage guidance to understand the trade-offs.

This is pretty interesting: An open-source memory and activity-recording layer for developers using coding agents such as Claude Code, Cursor, Codex, and OpenCode. The idea is to collect session history across tools and turn it into reusable knowledge. After a session, this creates a candidate "lesson" you can approve, making it available to future agents.

Redis is cutting down the cost of running AI applications: 1. Semantic caching — Reuses generated answers from previous requests as long as they are sufficiently similar 2. Persistent memory — Retrieves relevant conversation context instead of requiring the full history in every prompt 3. Storage tiering — Moves less-used data from RAM to SSD Here is how each of these works: SEMANTIC CACHING - Using Redis LangCache • Your app sends a request to Redis • Redis embeds it • It then compares it with previous embedded requests • If it finds a match, it returns the stored answer • If there's no match, Redis calls the LLM PERSISTENT MEMORY - Using Redis Agent Memory • Your app sends a request to Redis • Redis extracts facts about the conversation in the background • It embeds those facts and stores them as long-term memory • Your app can now search that memory for relevant context STORAGE TIERING - Using Redis Flex • Redis keeps frequently accessed data in RAM • It automatically moves less-used data to SSD • When your app requests data, Redis brings it back into RAM • Redis adjusts where to store data as access patterns change • Your app keeps using the same Redis API Basically, your applications will make fewer repeated calls to the model, use smaller prompts, and need less RAM to do the same. Here is more information about this: https://t.co/uKjrSTy4YL Thanks to the @Redisinc team for partnering with me on this post.

Here is a model that was trained specifically to create layered designs. This is good because you can use it to design something, but you can also modify it later in any other editing software. Usually, when you generate an image with a regular model, you get a bunch of pixels. Vyb does something different: it builds a structured composition containing text, vector graphics, and photos. And all of these remain individually editable.

I haven't talked to a single company that thinks it needs a better model to do what it wants. I think we have far more "intelligence" than we can use, and the biggest problem today is organizing knowledge and using it with the models we have. These guys are doing something cool: They connect your code with your company's knowledge base. Basically, when you merge a pull request, Falconer detects which docs are now outdated and prepares updates.