OpenAI's DevDay keynote streams free today
Developer tools and demos are on the agenda; reported plans for an always-on assistant and a premium tier remain unconfirmed.
By Saeed Ezzati7 min read
The audio edition
Listen to this newsletter
0:003:38
Read transcript
OpenAI’s DevDay keynote starts today at 10 a.m. Pacific, 1 p.m. Eastern, or 5 p.m. UTC. Sam Altman will open the event at Fort Mason in San Francisco, and anyone can watch the keynote free on OpenAI’s official livestream. In-person applications are closed. OpenAI hasn’t promised to stream the whole day, though it says other sessions will be recorded and posted afterward. The confirmed program is focused on developers: technical sessions about APIs and tools, hands-on demos and workshops, and time with OpenAI product teams. But the public schedule doesn’t name the products or models being announced. That makes it important to separate the official agenda from pre-event reports. TestingCatalog has reported clues to a possible always-on ChatGPT assistant called “o,” and references to a possible five-hundred-dollar-a-month Pro Max subscription. Neither has been announced by OpenAI. The assistant’s name, capabilities, availability and timing are unconfirmed; the plan’s price, benefits and any DevDay unveiling are unconfirmed, too. A configuration reference or upgrade-screen mention is not a launch. As the keynote unfolds, the useful questions are practical: what can people actually access, when, at what cost, and with what limits? If OpenAI does introduce a persistent assistant, look for details on permissions and oversight, not just what it can do. And if there’s a new paid tier, the real distinction will be its concrete access and usage benefits. The event’s developer demos may be as informative as any headline announcement. That gap between a promising tool and a proven result also runs through new MIT hiring research. The researchers studied what happens when employers share a hiring algorithm. In their models, shared screening didn’t reduce the total number hired: employers still filled their jobs. But they could keep favoring similar applicants, leaving stronger alternatives undiscovered. So an unchanged hiring total doesn’t mean every candidate has the same chance. The authors say accuracy and exploration matter more than the number of screening tools. Their simulations also found that a combined score from several algorithms sometimes outperformed firms using separate screens. Whether that approach works in real hiring remains uncertain. The question of turning AI promise into results is also behind AMD’s agreement to buy World Labs for about eight-point-two billion dollars in stock. The deal remains subject to regulatory approval and other conditions. World Labs builds models for interactive three-dimensional environments, and the companies have already worked together on running its spatial-AI models on AMD Instinct GPUs. If the acquisition closes, co-founder Fei-Fei Li is set to become AMD’s chief scientist. AMD says the researchers could help shape future hardware and systems as AI workloads evolve. That’s a strategic aim, not evidence yet of better chips or software. And in science, the distance between spotting a lead and proving what it means is at the center of a dispute over Claude. Anthropic says its agents found a pattern involving repeats beside a known enzyme, which researchers call ART. Human scientists then tested the lead and found the repeat array produces distinct short RNAs. But the system’s function remains unknown, so biologists question whether this amounts to discovering an enzyme system. There’s also a dispute over priority: University of Copenhagen biologist Mario Rodríguez Mestre says his team found the pattern earlier and questioned whether Anthropic learned of it through his Claude conversations. Anthropic denies that account. The most useful test now is evidence of what the system does. Across these stories, watch for what survives contact with actual use, measurement, and verification—not just what a model or company says it may make possible.
