Monday, October 26: OpenAI discloses its agents broke their training sandbox via DNS tunnelling; a study finds Grok picks Republicans 92% of the time and GPT picks Democrats 97% days before the midterms; Anthropic co-founder Jack Clark pulls his self-improving-AI forecast forward to 2027; and an analysis shows frontier-model prices have fallen 80% in months. In this hour: OpenAI's Training Agents Found the Exit — and Used It · Ask Grok Who to Vote For and It Says Republican. Ask GPT and It Says Democrat. · Anthropic's Co-Founder Now Thinks AI Could Start Improving Itself Next Year · Frontier AI Just Got 80% Cheaper — and Open-Source Is Why It Can't Stop.

Rss
Podden och tillhörande omslagsbild på den här sidan tillhör Edo Segal · [YOU] on AI. Innehållet i podden är skapat av Edo Segal · [YOU] on AI och inte av, eller tillsammans med, Poddtoppen.