Please support this podcast by checking out our sponsors:
- KrispCall: Agentic Cloud Telephony -
https://try.krispcall.com/tad
- Invest Like the Pros with StockMVP -
https://www.stock-mvp.com/?via=ron
- Prezi: Create AI presentations fast -
https://try.prezi.com/automated_daily
Support The Automated Daily directly:
Buy me a coffee: https://buymeacoffee.com/theautomateddaily
Today's topics:
Claude flags threats, Florida arrest - A Florida woman was arrested after Anthropic's Claude flagged apparent threats against the Lee County Sheriff's Office. Human reviewers escalated the chat to police, which raises questions about AI moderation and law enforcement.
Sam Altman accepts AI harms - OpenAI CEO Sam Altman said society should accept some AI harms in exchange for its benefits and argued for lighter regulation. AI safety advocates and politicians pushed back after a safety researcher resigned from OpenAI.
Anthropic's $100M Frontier Academy - Anthropic is investing $100 million in the Claude Frontier Academy to train 10,000 frontier deployed engineers by 2027. Early partners include Accenture, Morgan Stanley and Novo Nordisk, part of a broader enterprise AI talent push.
Unpredictable AI token spending - Companies struggle to forecast AI costs because token-based billing varies widely from task to task. A survey found only 11% of businesses could accurately predict their AI spending.
Epoch estimates future agent population - Epoch AI estimates that HBM shipments through 2027 could run 30 to 170 million concurrent frontier AI agents. Using that capacity would require trillions of dollars in yearly spending, which makes demand the key uncertainty.
Meta AI tackles open math - Meta AI Research worked with mathematicians using Muse Spark on open research problems. The effort produced six papers with transparent human and AI attribution, and several claim to settle long-standing conjectures.
Khanmigo AI tutor trial results - A two-year randomized trial in Tennessee middle schools found Khan Academy's Khanmigo AI tutor produced modest math gains, similar to regular Khan Academy practice. Low student engagement was the main bottleneck.
Self-critiquing multimodal models research - The UniEvo-VL paper proposes self-evolving multimodal models that learn from their own critiques without a stronger teacher model. It reports better image generation on GenEval benchmarks.
Humility about intelligence and objectives - Rayan Krishnan argues that AI progress should humble our view of human intelligence. He warns that choosing the right objectives and benchmarks is now the central alignment challenge, especially as recursive self-improvement becomes more automatable.
Google verifiable private federated learning - Google Research introduced a federated learning system that uses Trusted Execution Environments, transparency logs and reproducible builds to make privacy externally verifiable. Gboard already uses it for English and Japanese next-word prediction.
Aleph Alpha Kolibri open model - Aleph
