OpenAI launches AI TikTok ahead of DevDay week
Get bigger weekly updates! Free subscribers receive the top stories each week, while Paid subscribers will get a few extra stories. All support for Handy AI directly helps me maintain the newsletter and keep the information flowing.
last week’s top stories
🎥 OpenAI launches Sora 2 and sparks new AI video app craze. OpenAI unveiled Sora 2, an advanced AI video generator that produces shockingly realistic videos complete with synchronized audio. The September 30 release also came with a TikTok-style Sora social app, where users can create 10-second AI-generated clips and even “cameo” themselves or friends into scenes via a one-time face-and-voice scan. The invite-only app skyrocketed to the top of the App Store as users marveled at Sora 2’s physics-aware stunts (e.g. realistic rebounds and backflips) and its meme-worthy creativity.
🔮 Rumors swirl ahead of OpenAI DevDay, from Agent Studio to a ChatGPT browser. OpenAI’s DevDay 2025 (Oct 6) has the AI world buzzing with speculation. Leaked footage suggests OpenAI will unveil an “Agent Studio”, a visual toolkit for building custom AI agents by dragging and dropping modules and connecting data sources. Observers also suspect CEO Sam Altman might finally debut a ChatGPT-powered web browser to rival Chrome, updates on OpenAI’s stalled hardware project, and perhaps details on GPT-5. Nothing is confirmed, but the year of AI agents hype is high: OpenAI’s leaked Agent Builder demo showed a smooth interface to craft multi-step AI workflows (with guardrails against jailbreaks). Read more
💻 Google brings ‘Jules’ AI coding agent to the terminal. Google has expanded its AI coding assistant Jules with a new command-line interface and API, letting developers use the agent directly in their terminals and IDEs. Previously available only via a web UI, Jules can now be summoned in a developer’s shell to generate and modify code, handle Q&A tasks, and even run preset workflows without leaving the coding environment. Read more
🎯 Meta will mine your AI chats for ad targeting. Meta announced that starting Dec 16, it will use data from people’s interactions with its new generative AI assistants to personalize the content and ads they see on Facebook and Instagram. Users cannot opt out of this expanded data collection (though it only applies to those who use Meta’s AI features). Meta says it will exclude sensitive topics (like health or politics discussed with the AI) from advertising use. Read more
🤝 OpenAI and AMD forge multi-billion chip partnership. ChatGPT-maker OpenAI has signed a multi-year deal with chipmaker AMD to deploy roughly 6 gigawatts of AMD’s Instinct GPUs in OpenAI’s data centers, starting with 1 GW in 2026. The partnership could generate tens of billions in revenue for AMD while giving OpenAI a crucial second source of AI hardware besides Nvidia. Notably, the agreement includes a warrant allowing OpenAI to buy up to a 10% stake in AMD at a nominal price. Read more
🛠️ Ex-OpenAI CTO’s stealth lab debuts ‘Tinker’ to democratize fine-tuning. Mira Murati (OpenAI’s former CTO) has launched the first product from her secretive new startup Thinking Machines Lab: an AI tool called Tinker that makes it easier to fine-tune large AI models. Tinker is a managed API platform that automates the heavy lifting of custom model training, from spinning up GPU clusters to handling long training runs and stability issues. Read more
💰 OpenAI becomes world’s most valuable private company at $500B. A recent share sale has reportedly catapulted OpenAI’s valuation to an eye-popping $500 billion, overtaking SpaceX and ByteDance as the highest-valued private firm. In a secondary stock sale, investors like SoftBank and Dragoneer bought $6.6 billion worth of shares from OpenAI employees, which pushed the valuation to the half-trillion mark. The deal gave early staff a huge payday and serves as a retention play amid fierce AI talent wars. OpenAI’s revenue is surging (over $4 billion in the first half of 2025) but so are its costs; it’s burning cash on massive cloud contracts (like $300B with Oracle). Read more
🧪 AI Research of the Week
GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks
From OpenAI
Jake’s Take: This study, featuring OpenAI’s new GDPval benchmark, had industry experts evaluate AI vs. human outputs on 1,320 real work tasks across 44 professional roles in sectors from healthcare and finance to government.
Today’s top models (like GPT-5 and Anthropic’s Claude Opus 4.1) already produce expert-level work about “half the time,” matching or beating human professionals on many tasks. These AIs also completed the tasks roughly 100x faster and 100x cheaper than the human experts. In practical terms, this is hard evidence that AI can handle a significant share of knowledge work (writing reports, coding, analyzing data) with massive efficiency gains.
Leveraging advanced AI in core workflows is rapidly shifting from an experiment to a competitive necessity, so leaders should start integrating these tools (and/or upskilling their teams to use them) ASAP.
and then, even more news…
⚠️ OpenAI and Jony Ive’s secret AI device hits design snags. The Financial Times reports that Sam Altman and famed designer Jony Ive’s hush-hush AI gadget (envisioned as a screenless, always-listening assistant) may be delayed into 2026 due to major engineering hurdles. The palm-sized device (acquired via OpenAI’s $6.5 billion purchase of Ive’s startup) is supposed to use cameras, microphones and speakers to respond contextually to your environment. But insiders say the team is struggling with core questions: what “personality” should the AI have, how to ensure privacy with constant sensors, and how to supply the massive cloud compute such a device needs. Read more