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

🔥 Someone Trained an LLM in Space This Year (And 50 Other Things You Missed)- ThursdAI yearly recap is here!

·1 hr 50 min·5 clips

DeepSeek R1 was called “the first reasoning model in open source” and was said to have caused a $560 billion Nvidia loss.

Alex Volkov opens the yearly recap edition with a festive live welcome and brings in Wolfram Ravenwolf, Yam, and LDJ. The room treats it like the last regularly scheduled stream of the year. That matters, because the whole episode is built around how much 2025 managed to cram in. The panel keeps circling back to how some of the releases feel like they should have happened much longer ago. The tone stays celebratory, caffeinated, and a little unhinged. The recap starts broad before it drops into the actual launch pile. Google I/O gets handled more like an avalanche than a single conference. Gemini 2.5 Pro Deep Think is named as one of the big markers. Gemini 2.5 Flash is called out for going to GA with thinking budgets. Gemini Diffusion comes up as a strange, interesting detour, with the group noting its token-filling style and its speed claims for code and math. Google's agent story also gets attention through Jules and Project Mariner. Project Mariner is described as browser control via API, and the conversation notes that it started as a Chrome extension. Gemini Ultra and AI Mode in Search are added to the list of platform moves. The panel then pivots into open-source and benchmark talk. Mistral's Devstral shows up as a 24B developer model. SWE-bench Verified is treated like the benchmark everybody kept chasing higher all year. The room also moves through the Codex agent, the OpenAI/io/Johnny Ive deal, and Microsoft MCP in Windows. LM Arena's price increase gets its own mention. Nvidia and Emotron are part of the relentless release churn too. The whole episode feels like a live group chat trying to metabolize a year of AI news in real time. It never slows down long enough to turn clinical. Instead it keeps stacking names, claims, jokes, and quick corrections until the year starts to feel legible again. The holiday close is warm, brisk, and direct. The show asks listeners to subscribe, send one friend the link, and leave a five-star rating if they want to support the work.

As heard by us

A brisk year-end AI recap that turns a flood of releases into a readable map.

ThursdAI's year-end episode moves quickly through a crowded 2025 AI news cycle, with Alex Volkov and the panel revisiting model releases, agents, benchmarks, and platform shifts. The strength here is not just coverage but sorting the noise into something usable.

Read the full review in PlayNext →

Why you'd press play

Catch the 2025 AI recap that packs a wild year of releases into one live panel.

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