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 - Feb 26 - The Pentagon wants War Claude, every benchmark collapsed, and a solo founder hit $700K ARR with AI agents

·1 hr 50 min·5 clips

Anthropic's red lines are no autonomous weapons and no domestic mass surveillance.

1. ThursdAI focuses on the February 26 AI news cycle and the episode title points to the Pentagon, collapsed benchmarks, and Ben from Pulsia crossing $700K ARR with AI agents. 2. Alex Volkov hosts, with Jan Peleg, Wolfram, LDJ, Ryan Carson, Carson, and Nader Dabit joining as co-hosts or guests for the news rundown. 3. The episode asks what AI news still matters when model releases, agent tools, and government pressure are all accelerating at once. 4. The panel opens with Anthropic's reported ultimatum from the Pentagon over Claude and military access by Friday. 5. They say Anthropic's stated limits cover autonomous weapons and domestic mass surveillance, while the government may use the Defense Production Act if Anthropic refuses. 6. The discussion notes that the report comes through Axios, not an official Anthropic or government statement. 7. The episode then turns to Anthropic's report naming DeepSeek, Minimax, and ZAI in alleged distillation attacks against Claude. 8. The hosts compare the reported account counts and note that DeepSeek is named first even though Minimax and ZAI are described with larger numbers. 9. They also discuss the timing of DeepSeek's next model and the broader pressure Chinese and U.S. labs are putting on each other. 10. OpenAI's GPT 5.3 Codex appears as another major release, with pricing cited at $1.75 per million input tokens and $14 per output tokens. 11. Ryan Carson says Codex is strong for writing code but too literal when explaining code back to humans. 12. The panel says OpenAI's Codex works well in open-claw for GPT-5, but not as well with GPT-5.3 Codex specifically. 13. Alibaba's QN 3.5 Medium is described as a 35-billion-parameter model with 3 billion active parameters. 14. Liquid AI's LFM 2.24B is presented as a 2.3-billion-active-parameter model that is fast for local inference. 15. The benchmarks corner covers METR's long-horizon task work, including Opus completing a 14.5-hour task. 16. The same section says RKGI2 is reported at 97.9% and SWE-bench Verified is being retired because its gold solution is widely known. 17. The tools segment covers Cloud Code's one-year birthday, Devin 2.2 with browser use and self-fix, Cursor's cloud agents, and Perplexity Computer. 18. The episode's tone is fast, crowded, and conversational, with the hosts interrupting each other and reacting live to breaking AI news. 19. Listeners tracking AI policy, benchmarks, and agent tools will get the most value. 20. People wanting a slow, single-topic interview will probably skip it.

As heard by us

Fast AI news with a sharp take on what benchmark wins actually mean.

This weekly AI news show keeps pace with a field that changes too fast for tidy coverage, and it shows how much judgment goes into deciding what deserves airtime. The benchmark segment is the strongest part.

Read the full review in PlayNext →

Why you'd press play

You want the week’s AI chaos filtered into a fast, expert-led recap.

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