Siri Goes With Gemini

·21 min·2 clips

Apple announces it will use Google's Gemini to power Siri later this year, sending both companies' stocks soaring.

The episode opens with the major news that Apple has struck a multi-year partnership to use Google's Gemini models to power new Siri features, citing Google's tech as 'the most capable foundation.' This follows Apple's delayed AI upgrades and the departure of its AI chief. The news pushed Google's market cap past $4 trillion. The show then shifts to regulatory actions against AI. Malaysia and Indonesia have limited access to XAI's Grok chatbot over non-consensual sexual content generation. In the UK, Ofcom has launched a formal investigation and threatened X with a ban or multi-million pound fine under the Online Safety Act. Next, the episode covers Anthropic's launch of Claude for Healthcare, a HIPAA-ready toolset for medical providers and consumers, highlighting competition in healthcare AI. The narrative then explores Stack Overflow as a case study. Despite monthly questions dropping to 6,866—levels not seen since 2008—the company's revenue doubled to $115 million. It pivoted from ads to enterprise solutions like Stack Internal and data licensing, similar to Reddit's model. CEO Prashanth Chandrasekar notes that complex questions still come to Stack Overflow because LLMs need its curated data. Finally, the episode delves into AI model memorization. A Stanford study found models like Claude can reproduce long excerpts from books like Harry Potter when prompted. This contradicts AI companies' claims that models 'learn' rather than store copies. The technical explanation is 'lossy compression,' where models approximate training data. A German court invoked this in a ruling against OpenAI. Further research showed Meta's model could output verbatim large portions of Ta-Nehisi Coates' 'The Case for Reparations' and other works. The episode concludes by questioning the industry's 'learning' metaphor, suggesting models are essentially databases of token sequences from specific texts.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Tech Brew Ride Home