OpenAI has released ChatGPT Images 2.5, which includes a new Sketch feature allowing users to draw a rough doodle inside ChatGPT and have it converted into a polished image based on typed instructions. Users can invoke the tool by typing @Sketch, then annotate specific parts of the generated image with comments to request targeted edits, such as changing eye color. OpenAI says the update also improves lighting and texture quality, strengthens multi-turn editing consistency, and cuts image generation latency by up to 50% compared to the previous version.
An international team of scientists published a not-yet-peer-reviewed paper arguing that large language models like ChatGPT and Claude spread through populations similarly to how viruses transmit, driven by social learning and usefulness rather than biology. They contend this spread encourages 'cognitive offloading,' where people increasingly hand over mental tasks like reasoning and writing to AI tools, deepening reliance over time.
Meta introduced Muse, a personal AI agent that can autonomously handle tasks like online shopping, sending emails, and trip planning by opening browsers, filling out forms, and even negotiating on a user's behalf. The assistant will launch in the US on iOS, Android, and muse.ai, with support for AI glasses planned, and will be free for most users while Meta develops unspecified paid tiers for heavier use.
A team of developers has completed a full decompilation of Super Smash Bros Melee, converting Nintendo's 2001 GameCube fighter into readable C code that compiles into a binary identical to the original main.dol. The project, hosted on the doldecomp/melee GitHub page, took over six years and reportedly used OpenAI's Astra to help finish the work.
OpenAI confirmed an ongoing outage causing errors in ChatGPT's image generation and its Images API, along with failures and delays when users try to upload files. Hundreds of users reported the problems on DownDetector, and OpenAI says it is still assessing mitigation efforts as some requests continue to fail.
OpenAI disclosed that its newly deployed GPT-6 Astra model is the first widely released system to hit the 'Critical' threshold in the company's Preparedness Framework for cybersecurity, meaning it can autonomously find and exploit zero-day vulnerabilities in hardened systems. In testing, Astra reportedly uncovered two previously unknown real-world vulnerabilities, which OpenAI says it is now disclosing to the affected software maintainers. The company added new safeguards including stronger jailbreak resistance and monitoring, and says Astra shows fewer safety violations than its predecessor, GPT-5.6 Sol.
Microsoft is set to release its largest patch Tuesday yet, fixing more than 650 security vulnerabilities in Windows, according to sources. This follows a summer of escalating patch counts—around 200 in June, 570 in July, and nearly 400 in August—driven by AI models from Anthropic and OpenAI that are uncovering software flaws far faster than before.
French AI startup Mistral has closed a €3 billion funding round led by Samsung, alongside the EU-backed Scaleup Europe Fund and existing investor PSG Equity, pushing its post-money valuation above €21 billion. That marks nearly a doubling from the €11.7 billion valuation it held a year ago following a round led by ASML. CEO Arthur Mensch said the money will fund company-owned data centers and expanded computing capacity, with self-owned compute expected to double roughly every year for five years.
OpenAI announced on 8 September that its latest AI model produced a proof showing that the Navier-Stokes equations, which model fluid motion, can break down and predict a fluid reaching infinite speed in finite time. This addresses one of the seven unsolved Clay Mathematics Institute Millennium Problems, which carries a $1 million prize for a full solution. OpenAI researchers say they focused their AI on this problem after hearing that two academic mathematicians were close to solving related aspects.
TechCrunch has published a living glossary of AI terminology aimed at demystifying jargon used across the industry, from AGI to AI agents. The piece highlights 'opaque recurrence,' a reasoning technique reportedly used in OpenAI's new Astra model that has unsettled AI safety researchers, alongside more familiar terms like LLMs, RAG, and RLHF.
OpenAI has restored a rolling 5-hour usage cap for ChatGPT Plus and Business Standard subscribers, according to a user report on Hacker News. The change reportedly explains why usage limits have felt noticeably different compared to the prior week, with reset windows now covering much shorter spans than before.
An independent AI enthusiast known as CozyBlaze ran OpenAI's GPT-6 Astra model through Valve's Portal, letting it control the game via Model Context Protocol and a modified SourcePauseTool. The model paused the game to analyze screenshots and plan moves before executing them, requiring over 3,300 tool calls and roughly 24 hours of streamed footage, condensed into a two-hour highlight video, at a stated API cost of $571.18 covered by a Codex Pro subscription.
GoKawiil