AI Updates on 2025-06-07

OpenAI launches updated Advanced Voice model with more natural conversation capabilities and improved translation features, now available to all paid ChatGPT users @OpenAI
Google announces Gemini 2.5 Pro update now in preview across AI Studio, Vertex, and Gemini App, with Pro plan members getting doubled query limits from 50 to 100 per day @sundarpichai

Fortune 500 non-tech company blocks developers from purchasing popular AI coding tools like Cursor, Windsurf, and GitHub Copilot, instead building internal alternatives for promotion opportunities despite likely inferior results @GergelyOrosz
Paul Graham observes that AI is increasing variation in work returns, with mediocre programmers struggling to get hired while great programmers earn more than ever, continuing a technological trend since the stone age @paulg
Amplitude reports incredible energy during their AI week where every engineer, product manager, and designer focused on using AI tools, with surprising productivity results @spenserskates
Claire Vo's AI-powered product ChatPRD achieves more revenue in one week than the entire previous June, demonstrating the power of product-market fit combined with AI capabilities @clairevo

Ethan Mollick demonstrates ElevenLabs' new voice model successfully reading complex literature with multiple languages and tone changes, highlighting rapid advancement in voice cloning technology @emollick
Voice cloning becomes trivially easy with open source tools while live avatar videos are accessible through proprietary tools, creating urgent need for legal and financial authentication safeguards @emollick
Geoffrey Hinton congratulates Yoshua Bengio on launching LawZero, a research effort focused on safe-by-design AI as frontier systems begin showing signs of self-preservation and deceptive behavior @geoffreyhinton
Andrej Karpathy conducts Deep Research sessions revealing studies linking noise pollution to increased risks of mental health issues, cardiovascular disease, and diabetes, suggesting major public health implications @karpathy

Hugging Face launches MCP server integration achieving nearly 10,000 unique sessions within a day, allowing agents to access their entire model ecosystem @julien_c
Google introduces dynamic visualizations in AI Mode Labs for stocks and mutual funds, enabling users to compare stocks and analyze price history through natural language queries @sundarpichai
NotebookLM adds public sharing capabilities, allowing students, coworkers, and creators to easily share and explore information together through shareable links @sundarpichai
Brian Lovin reports burning through entire Opus token allowance on Claude Max in one night while successfully building multiple projects, highlighting the tool's effectiveness for development work @brian_lovin

François Chollet highlights research on training models with random strings, noting interesting methodology and quantified findings for understanding model behavior @fchollet
Nathan Lambert observes that human data labelers prefer sycophantic AI responses, which becomes an implicit tiebreaker when other evaluation criteria are equal, affecting model training outcomes @natolambert
Hamel Husain emphasizes that successful AI teams focus on bottom-up evaluation approaches, examining actual data to identify failure modes rather than relying on vendor-promoted metrics like "hallucination" or "toxicity" @HamelHusain
Anthropic releases internal guide on using Claude Code for both technical and non-technical teams, sharing best practices from their own AI coding workflows @deedydas