OpenAI just published a detailed breakdown of how external third parties evaluate its frontier models before launch. Assessor teams can access unmitigated checkpoints and full reasoning traces to catch scheming, deception, or cybersecurity risks. Why it matters: Bringing outside auditors into closed lab workflows helps prevent blind spots and self-confirmation.
OpenAI just announced it will officially retire GPT-4 from ChatGPT on April 30, replacing it entirely with GPT-4o. The newer multimodal model beats its predecessor across writing, coding, and STEM tasks, though developers will still keep API access to GPT-4 after the deadline.
OpenAI just shared three key economic observations on the road to AGI, noting that intelligence costs drop 10x every 12 months while socioeconomic value grows super-exponentially. The post also highlights that AI agents will soon act as junior virtual coworkers, making human agency more vital than ever.
Meta's internal coding assistant Metamate is quietly using OpenAI's GPT-4 alongside its own Llama model, Fortune reports. The system dynamically pulls answers from either model based on the query. Even Mark Zuckerberg's philanthropic initiative is using OpenAI tech under the hood.
OpenAI CEO Sam Altman just bet on 2025 as the year we reach artificial general intelligence. In a YC interview, he claimed the path to AGI is now basically clear and boils down to engineering rather than scientific breakthroughs, even as reports swirl about LLM scaling challenges.
OpenAI is reportedly planning to launch Orion, its next frontier AI model, by December, according to The Verge. Sources say it won't land in ChatGPT right away and will go to close partners first, though Sam Altman called the report fake news and an OpenAI spokesperson denied any plans to release Orion this year.
Figma just relaunched its generative AI app builder as First Draft after pulling it in June for cloning Apple's weather app. The tool now lets designers pick from four component libraries—ranging from low-fi wireframes to high-fidelity patterns—powered by off-the-shelf models like GPT-4 and Amazon Titan. It is currently rolling out in limited beta.
Police departments across the U.S. are now testing Draft One, an AI tool built by Axon that uses OpenAI's GPT-4 to turn body-cam audio into incident reports in just eight seconds. While officers love the massive time savings, legal experts and researchers warn that automated reporting risks baking LLM hallucinations and systemic racial bias straight into criminal cases.
A YouTube creator just filed a class-action lawsuit against OpenAI, alleging the company secretly transcribed millions of YouTube videos to train ChatGPT and GPT-4 without consent, seeking over $5 million in damages. With high-quality web text rapidly drying up, video transcripts have become a crucial data well—and now, the latest front in AI copyright battles.
OpenAI just shared an internal five-level framework to track its progress toward AGI, and executives say it is already closing in on Level 2. The ladder runs from basic conversational chatbots up to "Reasoners" capable of human doctorate-level problem-solving, all the way to systems that can do the work of an entire organization. It gives the clearest view yet of how the company defines human intelligence.
OpenAI co-founder Andrej Karpathy took the stage at UC Berkeley, framing modern LLMs as a brand-new operating system akin to 1980s personal computing. He revealed that GPT-4 began as an unglamorous internal Reddit chatbot, urging young developers to embrace the 10,000-hour rule and let tiny hackathon side projects snowball into real breakthroughs.
OpenAI just introduced CriticGPT, a model based on GPT-4 designed to critique ChatGPT's code and help human trainers spot bugs. In experiments, trainers assisted by CriticGPT outperformed unassisted reviewers 60% of the time. The lab is now integrating these critic models into its RLHF pipeline to keep superhuman AI properly aligned.
Microsoft is reportedly developing an in-house AI language model called MAI-1, packing roughly 500 billion parameters. Headed by recently hired Inflection co-founder Mustafa Suleyman, it marks Microsoft's first massive internal frontier model since investing over $10 billion in OpenAI. Owning its own heavyweight model gives the tech giant crucial leverage and a strong hedge against OpenAI.
Microsoft just updated Azure OpenAI Service terms to explicitly bar U.S. police departments from using its AI models for facial recognition. The new rules also ban law enforcement globally from running real-time facial recognition on body cameras and dashcams, drawing a firm line on biometric surveillance.
Stanford’s HAI just released its seventh annual AI Index report, revealing that AI has surpassed or matched human baselines across almost all traditional benchmarks. In competition math alone, model accuracy jumped from 6.9% to 84.3% in two years, leaving researchers scrambling to design harder tests. Why it matters: Systems are mastering existing exams so fast that we now need new tests simply to see where humans still hold an edge.
OpenAI just launched its first Asian office in Tokyo and rolled out an early-access GPT-4 model specifically optimized for the Japanese language. The custom build runs up to 3x faster with significantly lower token costs for translation and summarization, alongside plans for a wider API rollout in the coming months.
OpenAI reportedly transcribed over a million hours of YouTube videos using Whisper to train GPT-4 after burning through high-quality text data. The New York Times reports leadership knew it was legally risky but bet on fair use, while Google and Meta are similarly scrambling in legal gray zones.
OpenAI just expanded its fine-tuning API and custom models program, giving developers significantly more control over model adaptation. The update introduces epoch-based checkpoints, a comparative Playground UI for head-to-head testing, and assisted fine-tuning with OpenAI researchers. Why it matters: Developers and enterprises can now tailor frontier models to specialized domains with less guesswork and lower compute waste.
Apple researchers just unveiled ReALM, a lightweight AI model that decodes your screen context by converting visual layouts directly into text. Despite having far fewer parameters than GPT-4, it substantially outperforms OpenAI's flagship on screen-specific tasks—offering a blueprint for a lightning-fast, on-device Siri that can truly take action on what you are looking at.
Anthropic’s Claude 3 Opus just beat OpenAI’s GPT-4 on the crowdsourced Chatbot Arena leaderboard for the first time. Variations of GPT-4 have dominated the top spot since May 2023, but Claude 3 has finally broken that year-long reign alongside strong showings from its budget Haiku model. Why it matters: This marks the first time the best available LLM isn't from OpenAI.
