A new study by Apple suggests that AI models may not truly ‘reason’ through problems, based on puzzle-based experiments showing limitations in simulated reasoning. However, the findings are disputed by others in the field, highlighting ongoing debates about the true capabilities of AI in problem-solving.
A clear-eyed digest of AI, products, and ideas worth understanding. Join the NextBigWhat newsletter.
AI spending growth has stalled, raising concerns as Anthropic approaches its IPO, despite increased usage driven by price competition.
A file from Claude Code has leaked sensitive passwords and keys across 1,122 public GitHub repositories, prompting urgent security audits.
Nvidia’s CEO Jensen Huang acknowledges an affair with a suspected Chinese spy, prompting a security investigation by the company.
A vulnerability in shared ChatGPT links could let attackers execute code and access private chats, raising serious security concerns.
Goldman Sachs has banned traders from using AI coding tools, citing productivity and risk concerns after a trial period of usage.
Mistral has unveiled open weights for its Mistral Large 4 model, enabling companies to run advanced AI securely on their own hardware.