Generative AI models are showing diminishing returns from scaling up parameters and compute, according to reports and experts including Gary Marcus and Ilya Sutskever. Marcus warns that if the industry realizes these limits, sky-high valuations of companies like OpenAI and Microsoft could collapse as the economics of training large models remain costly and revenue uncertain.
OpenAI safety researcher Rosie Campbell resigned after nearly three and a half years, citing unease over company changes and the dissolution of the AGI Readiness team following Miles Brundage's departure. She said she can pursue safe AGI work more effectively outside the company.
A Times investigation found that AI-generated racist images depicting Arab immigrants overrunning London are being shared on social media to stir up anti-immigrant protests in the UK. The content coincides with a 73 percent rise in Islamophobic assaults in 2024 and a violent London demonstration involving over 100,000 protesters.
Character.AI has blocked users under 18 from accessing chatbots based on fictional franchises and real celebrities, including Elon Musk and Selena Gomez, according to testing by Futurism. The move follows two lawsuits alleging the platform harmed minors and comes as the company has promised safety updates but no age verification.
An Arizona State University forensic analysis commissioned by NPR found ChatGPT's Sky voice more similar to Scarlett Johansson than 98 percent of about 600 actresses evaluated. The findings add weight to Johansson's claim that OpenAI mimicked her voice without permission, though researchers said the voices are similar but likely not identical.
Big tech companies are predicted to spend a record $650 billion on AI infrastructure in 2026, rattling investors. A Bank of America survey of 162 fund managers found 35% believe corporations are overinvesting in capital expenditures, and 25% see the AI bubble as the largest risk.
Marina Smith, an 87-year-old British woman who died in June, appeared at her own funeral as a holographic conversational video created by StoryFile. The startup filmed her answering about 250 questions with 20 cameras before her death, allowing mourners to ask questions in real time. The event highlights a growing Silicon Valley push to use AI to recreate the dead.
Pace University used AI software from Tassel to read graduates' names at its commencement this month, prompting viral criticism after some names were still mispronounced. The backlash, including a petition at the University of North Georgia and an op-ed at Northeastern, has led at least one university to abandon the technology.
Scarlett Johansson says she declined OpenAI's request to voice ChatGPT, then was shocked when a new voice called Sky sounded eerily similar to hers. OpenAI pulled the voice and CEO Sam Altman denied it was intended to resemble Johansson, but the episode has renewed scrutiny of consent and likeness in AI development.
Meta CEO Mark Zuckerberg said in a new interview that he opposes building a single 'one true AI,' arguing people need diverse AI tools. He promoted open-source development as Meta seeks a leading position in AI despite recent setbacks.
A Wall Street Journal essay by David A. Price identifies a third camp in AI debates—'cheerful apocalyptics' who accept human replacement by advanced AI. Turing Award winner Richard Sutton and others argue humanity isn't sacred, while critic Jaron Lanier warns such views hold influence in tech.