ThursdAI - The top AI news from the past week · From Weights & Biases, Join AI Evangelist Alex Volkov and a panel of experts to cover everything important that happened in the world of AI from the past week

📆 ThursdAI - Oct 16 - VEO3.1, Haiku 4.5, ChatGPT adult mode, Claude Skills, NVIDIA DGX spark, Wordlabs RTFM & more AI news

·1 hr 35 min·5 clips
Google and Yale say a 27B Gemma model generated a novel cancer-cell hypothesis that scientists validated in living cells.
1. ThursdAI's October 16 episode focuses on AI news from the week, including open source models, product launches, hardware, video generation, and a World Labs breaking-news demo. 2. Alex Volkov hosts with Yam Peleg and Wolf from Ravenwolf, and the episode also features Jessica Gallegos from Google DeepMind, Kyle Corbett from OpenPipe, Quinn Slack from AMP, and Sweeks on breaking news. 3. The episode asks which releases actually matter this week, from a 27B scientific model to VO 3.1, ChatGPT adult mode, and NVIDIA's DGX Spark. 4. In open source, the hosts mention KAIST's bilingual Chrome model, Parieren's Screen-3 VL 2 Small release, and Qwen 3 VL 4B and 8B. 5. They say Qwen 3 VL 8B scores 33.9 on OSWorld and compare that result with larger models and with Anthropic's Sonnet 4. 6. They also say Qwen 3 VL 8B beats GPT-5 Nano on several vision benchmarks, including real-world QA and MMU Pro. 7. The most discussed open-source story is Google's C2S scale, a 27 billion parameter Gemma-based model from Google and Yale. 8. The hosts say C2S scale produced a new hypothesis about cancer cell behavior and that scientists validated the result in living cells. 9. They note the release includes open weights and a bioRxiv preprint, which they frame as important for scientific validation. 10. On the product side, the episode covers Anthropic's Claude Haiku 4.5, which the hosts describe as a fast and cheap model with roughly twice the speed of its predecessor. 11. OpenAI's Sam Altman is discussed for a thread about ChatGPT loosening restrictions, adding adult mode, and possibly supporting erotica by December. 12. The hosts also mention a recent expert council on well-being and AI with researchers from Harvard and Stanford. 13. Another OpenAI update is memory management in ChatGPT, which the hosts present as an automatic memory feature. 14. Microsoft is discussed for Copilot on Windows 11, with the hosts describing an OS-level assistant that can execute tasks on the computer. 15. NVIDIA's DGX Spark is described as a 128 GB low-power workstation, and the hosts say Jensen Huang hand-delivered units to Elon Musk and Sam Altman. 16. Apple is discussed for the M5 chip, which the hosts say includes an accelerator inside each core and arrives in MacBook Pro, iPad, and Vision Pro. 17. The interview with Jessica Gallegos centers on VO 3.1, where she says the update improves prompt adherence, image-to-video quality, and creative control. 18. Jessica also explains that scene extension can preserve audio if the last second of the source clip contains the sound, and she describes object insertion and removal as ways to keep shadows and occlusion natural. 19. The episode has a news-roundup format with fast transitions, host banter, and multiple expert interviews rather than one long deep-dive. 20. Listeners who follow AI product launches, model benchmarks, and scientific AI would likely enjoy this episode, while people avoiding rapid-fire news coverage may skip it.

As heard by us

A brisk AI-week roundup that favors live reactions and practical takeaways.

Thursday AI moves through a busy AI week with a quick, reactive rhythm. Gemini 3 hype, browser-built operating system demos, World Labs' RTFM, and Sora's longer video limits all get their share of attention, and the panel stays strongest when it stops at the headline and asks…

Read the full review in PlayNext →

Why you'd press play

You want a fast AI catch-up built from the week's actual demos and releases.

Read the full recommendation in PlayNext →
Listen to the show on