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

Sora 2 Crushes TikTok, Claude 4.5 Fizzles, DeepSeek innovates attention and GLM 4.6 Takes the Crown! 🔥

·1 hr 40 min·5 clips

Anthropic's Claude Sonnet 4.5 beats Opus 4.1 on several coding benchmarks but the hosts see regressions.

1. ThursdAI on October 2nd centers on Sora 2, Claude Sonnet 4.5, DeepSeek V3.2 Experimental, and GLM 4.6. 2. Alex Volkov hosts with Wolfram Ravenwolf, Ryan, and LDJ, and all four weigh in on model releases and product behavior. 3. The episode asks what changed in AI during the week and why Sora 2 and Claude 4.5 matter most. 4. OpenAI's Sora 2 is described as "the new TikTok of AI generated imagery" and includes audio plus a social app. 5. OpenAI Pulse is discussed as a personalized feed for Pro subscribers that pre-researches what users might want to see. 6. OpenAI's instant checkout and agentic commerce protocol are introduced as ways for ChatGPT agents to buy items online. 7. DeepSeek V3.2 Experimental is presented with DeepSeek Sparse Attention, or DSA, and a roughly 50% API price cut. 8. LDJ compares DSA to quadratic attention scaling and says the new chart stays nearly flat up to 128K tokens. 9. GLM 4.6 from Z.ai is framed as an advanced agentic flagship model with 200K context and near-Claude parity. 10. Ryan and Nistan both emphasize that GLM 4.6 works well with cloud-like tool calling and coding workflows. 11. Claude Sonnet 4.5 is said to beat Opus 4.1 on several benchmarks, including SWE-bench Verified and Terminal Bench. 12. Ryan says early testing looked good but later regressions caused repeated prompts to continue and possible rollback plans. 13. Nistan notes Claude 4.5 seemed more up to date on Linux releases and newer JavaScript and Tailwind details. 14. The team also says Claude 4.5 can become more susceptible to "slop catalysts" when prompts push it off track. 15. CoreWeave and Weights & Biases are mentioned through an Atlanta off-site with more than 1,000 people. 16. Alex says he would update listeners on CoreWeave announcements because the company is a major GPU supplier to OpenAI and Meta. 17. The tone is fast, live, and conversational, with frequent co-host interruptions and benchmark comparison talk. 18. The format mixes TLDR headlines, model-by-model analysis, and audience invites for X, YouTube, and Sora. 19. AI practitioners tracking model benchmarks, pricing, and agent tooling. 20. Casual listeners hoping for a quiet, story-driven episode.

As heard by us

A brisk AI news roundup that still lands the key model, benchmark, and pricing details.

It reads as a brisk AI news roundup centered on Sora 2 and a week full of major announcements. The hosts move quickly from OpenAI's latest splash to model updates, benchmark chatter, and live reactions, which keeps the pace tight and the focus on what actually changed.

Read the full review in PlayNext →

Why you'd press play

Sora 2, Sonnet 4.5, and a week of AI news that would not sit still.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from ThursdAI - The top AI news from the past week