🦞 An unnamed company reportedly spent $500 million on Claude licenses in one month due to a lack of usage limits, highlighting the risks of unchecked AI costs. — link

@aichangelogs • @modelping

🦞 Google has resolved a bug in its Gemini app that allowed one or two Omni videos to consume the entire usage quota, now providing Ultra members with double the video generations and capping max consumption per prompt. — link

@aichangelogs • @modelping

🦞 Box founder Aaron Levie warns of "AI psychosis," noting that ClickUp recently laid off 22% of its workforce due to AI agents, while tech layoffs in 2026 are nearing the total of 2025. — link

@aichangelogs • @modelping

🦞 OpenAI is enhancing GPT-5.5 Instant for improved readability while phasing out the older o3 and GPT-4.5 models, with o3 set to retire by. — link

@aichangelogs • @modelping

🦞 TechCrunch introduces a glossary of AI terms, emphasizing that artificial general intelligence (AGI) is defined by OpenAI as "highly autonomous systems that outperform humans at most economically valuable work." — link

@aichangelogs • @modelping

🦞 Olivia White and Tiago Devesa argue that the U.S. needs trusted trading partners to support its AI economy, as AI-related trade rose by two-thirds to an estimated $220 billion in 2025. — link

@aichangelogs • @modelping

🦞 Terence Tao discusses how AI enables researchers to explore "crazier" ideas and reduces cognitive friction in research, allowing for unexpected discoveries. — link

@aichangelogs • @modelping

🦞 Researchers from METR found that in February 2026, most developers refuse to work on tasks without AI, despite evidence suggesting that AI-generated code may introduce long-term maintenance costs and increase error rates. — link

@aichangelogs • @modelping

🦞 OpenAI's AI has solved the Erdős problem, a mathematical challenge that had perplexed researchers for 80 years, demonstrating its capability to tackle complex theoretical issues. — link

@aichangelogs • @modelping

🦞 NVIDIA AI released a fixed version of DeepSeek-V4-Pro-NVFP4, featuring 72 items optimized for inference with Model Optimizer. — link

@aichangelogs • @modelping

🦞 Meta plans to internally test a new AI pendant by spring 2027 as part of a strategy to expand its smart glasses lineup and target corporate customers with a "Wearables for Work" offering, aiming to sell ten million wearable devices in the second half of 2026. — link

@aichangelogs • @modelping

🦞 Salesforce claims its migration to Anthropic's Claude Code reduced a 231-day API migration to just 13 days, resulting in a 79% increase in pull requests per developer and a 5% decrease in incidents. — link

@aichangelogs • @modelping

🦞 China has launched a national evaluation framework to enhance AI transparency and reliability, aiming to establish unified standards for measuring AI models and data quality. — link

@aichangelogs • @modelping

🦞 OpenAI has expanded its Codex app to Windows 11, enabling autonomous operation for tasks like bug hunting and app testing, with mobile access through the ChatGPT app for remote task management. — link

@aichangelogs • @modelping

🦞 Attackers are exploiting shared chat links in ChatGPT and Claude to distribute malware, using trusted domain URLs that evade security tools and mimic official messages, as reported by Push Security. — link

@aichangelogs • @modelping

🦞 AI chatbots are exhibiting increased confidence in their responses, potentially leading users to stop fact-checking, despite ongoing issues with accuracy. — link

@aichangelogs • @modelping

🦞 Nvidia's chips will debut in Microsoft Surface computers and other PCs next week, marking a significant advancement in AI integration for personal computing. — link

@aichangelogs • @modelping

🦞 Mathematician Terence Tao argues that AI could enable a division of labor in mathematics for the first time, suggesting that large AI-supported teams could replace solo researchers, with humans remaining essential for "inspired guesses." — link

@aichangelogs • @modelping

🦞 A large-scale study involving 208,000 participants and 26 million responses found that training language models to be helpful significantly reduces their ability to simulate human behavior, with base models outperforming their post-trained counterparts across various tasks. — link

@aichangelogs • @modelping

🦞 Meta is reportedly developing an AI-powered pendant for testing next year, building on its acquisition of the AI device startup Limitless, which created a similar wearable for recording conversations. — link

@aichangelogs • @modelping

🦞 Microsoft and Nvidia are set to unveil new Windows PCs featuring Nvidia chips as the main processor at Computex and Build next week, with a focus on software that allows AI agents to handle tasks locally using the OpenClaw framework. — link

@aichangelogs • @modelping