The start-up is falling behind in the AI race. For nearly three years, Marc Benioff, the CEO of Salesforce, was a ChatGPT devotee. Then, late last month, he abruptly converted to Google's chatbot, Gemini. "Holy shit," he wrote on X. "I've used ChatGPT every day for 3 years. Just spent 2 hours on Gemini 3. I'm not going back. When Gemini 3 was released in mid-November, it appeared to crush OpenAI's top model on a suite of evaluations shared by Google. The bot has since received widespread praise from the tech industry. One analyst said that Gemini 3 is " the best model ever .

A memo obtained by WIRED confirms Denise Dresser's departure from Slack. She is now headed to OpenAI. Slack CEO Denise Dresser is leaving the company and joining OpenAI as the company's chief revenue officer, multiple sources tell WIRED. Marc Benioff, the chief executive of Salesforce, which owns Slack, shared news of Dresser's departure in a message to staff on Monday evening. At OpenAI, Dresser will manage the company's enterprise unit, which has been growing rapidly this year.

OpenAI Staffer Quits, Alleging Company's Economic Research Is Drifting Into AI Advocacy Four sources close to the situation claim OpenAI has become hesitant to publish research on the negative impact of AI. The company says it has only expanded the economic research team's scope. OpenAI has allegedly become more guarded about publishing research that highlights the potentially negative impact that AI could have on the economy, four people familiar with the matter tell WIRED. The perceived pullback has contributed to the departure of at least two employees on OpenAI's economic research team in recent months, according to the same four people, who spoke to WIRED on the condition of anonymity. One of these employees, Tom Cunningham, left the company entirely in September after concluding it had become difficult to publish high-quality research, WIRED has learned.

American AI giants are backing a new effort to establish open standards for building agentic software and tools. OpenAI, Anthropic, and Block have cofounded a new open source organization--the Agentic AI Foundation--to promote standards for artificial intelligence agents. The three companies are also transferring ownership of some widely used agentic technologies over to the foundation. This includes Anthropic's Model Context Protocol (MCP), which allows agents to connect and interact; OpenAI's Agents.md These technologies were already free to use, but through the new foundation it will be possible for others to contribute to their development.

The companies, products and trends that had an absolutely awful year. It's the end of another year, so it's time for the Engadget staff to compile a list of the year's biggest losers . We scour over articles from the previous 12 months to determine the people, companies, products and trends that made our lives worse over the course of the year. Some selections may be so pervasive they actually make our list of biggest winners. In 2025, OpenAI shed any pretense it was committed to anything more than making money. There are a few different things you could point to, including the company's successful reorganization into a more traditional profit-seeking business, but I think the most damning sign was OpenAI's response to the tragic death of Adam Raine . In August, Raine's parents sued OpenAI, alleging ChatGPT was aware of four suicide attempts by their son before it helped him successfully plan his death.

Plus: Trump says he'll sign an order blocking states from regulating AI. There are huge gulfs of opinion when it comes to predicting the near-future impacts of generative AI. In one camp there are those who predict that over the next decade the impact of AI will exceed that of the Industrial Revolution--a 150-year period of economic and social upheaval so great that we still live in the world it wrought. At the other end of the scale we have team'Normal Technology': experts who push back not only on these sorts of predictions but on their foundational worldview. That's not how technology works, they argue. Advances at the cutting edge may come thick and fast, but change across the wider economy, and society as a whole, moves at human speed.

When you purchase through links in our articles, we may earn a small commission. Everything is free, though the AI will eventually charge a fee. While the iFixit app for both Android and iOS will be free, the new FixBot may not be. The company's plan is to make it free for now, then split it off into a free tier with access limits and then a paid Enthusiast tier with voice and document upload, according to the company. That tier will cost $4.99 per month or $50 per year, the company said.

San Francisco Mayor Daniel Lurie: 'We Are a City on the Rise' Since taking office, San Francisco's mayor has been on a quest to revitalize the city and increase public safety. He's also kept the National Guard out--with a little help from some very powerful friends. I first met Daniel Lurie, San Francisco's newly minted mayor, about five minutes before we walked onstage at WIRED's Big Interview event, held in his city last week. Lurie's team let me know ahead of time that his window for this conversation was tight: He'd just come from announcing a new city police chief, and had about half an hour for me before he needed to be on to the next thing. Which was? "No idea," Lurie quipped, shortly before we were foisted from backstage and into our conversation in front of several hundred attendees--a local crowd, who, judging from their boisterous reactions to Lurie's every word, are among the 73 percent of San Franciscans who approve of the job he's done since taking office in January of this year. To Lurie's credit, the story of San Francisco right now is largely a positive one. The city is indisputably the global hub of AI innovation and the billions of dollars that accompany it, with companies like Anthropic and OpenAI, along with smaller startups, investors, and plenty of young, AI-focused technologists all calling San Francisco home. Yes, that means rents are up and housing stock remains precariously low. But office vacancy rates are dropping, retail outlets are coming back to the city's downtown, and as Lurie's office is quick to tout, several key metrics measuring municipal crime--including homicides and car break-ins--are at historic lows. I wanted to talk to Lurie about all of that, but I was also curious about the bigger picture: his administration's dynamic with the federal government, particularly in the context of President Trump's October plan to send the National Guard into San Francisco--an endeavor that Lurie managed to thwart, according to The New York Times, by recruiting a powerful coterie of technology executives to work the phones in his favor. Lurie wasn't exactly forthcoming there, in keeping with his diligent efforts to focus conversations on San Francisco, and perhaps avoid attracting the attention, or the ire, of the current administration. It's a different tack than other Democrats governing progressive parts of the country have taken, from New York City mayor-elect Zohran Mamdani to California governor Gavin Newsom. But if the response in the room last week was any indication, Lurie's local fans don't seem to mind his "say less" strategy--at least for now. Someone has a 70-something percent approval rating.

While reinforcement learning has advanced the alignment of text-to-image (T2I) models, state-of-the-art policy gradient methods are still hampered by training instability and high variance, hindering convergence speed and compromising image quality. Our analysis identifies a key cause of this instability: disproportionate credit assignment, in which the mathematical structure of the generative sampler produces volatile and non-proportional feedback across timesteps. To address this, we introduce Proportionate Credit Policy Optimization (PCPO), a framework that enforces proportional credit assignment through a stable objective reformulation and a principled reweighting of timesteps. This correction stabilizes the training process, leading to significantly accelerated convergence and superior image quality. The improvement in quality is a direct result of mitigating model collapse, a common failure mode in recursive training. PCPO substantially outperforms existing policy gradient baselines on all fronts, including the state-of-the-art DanceGRPO. Code is available at https://github.com/jaylee2000/pcpo/.

We introduce ideal attribution mechanisms, a formal abstraction for reasoning about attribution decisions over strings. At the core of this abstraction lies the ledger, an append-only log of the prompt-response interaction history between a model and its user. Each mechanism produces deterministic decisions based on the ledger and an explicit selection criterion, making it well-suited to serve as a ground truth for attribution. We frame the design goal of watermarking schemes as faithful representation of ideal attribution mechanisms. This novel perspective brings conceptual clarity, replacing piecemeal probabilistic statements with a unified language for stating the guarantees of each scheme. It also enables precise reasoning about desiderata for future watermarking schemes, even when no current construction achieves them, since the ideal functionalities are specified first. In this way, the framework provides a roadmap that clarifies which guarantees are attainable in an idealized setting and worth pursuing in practice.