Sign up to save your podcasts
Or
Share The File That Wouldn't Read
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
The File That Wouldn't Read
Thursday, May 7. The GPT-5.5 default swap is two days old and the cracks are showing — Mario Zechner caught it refusing to read full files. Subquadratic announced a 12-million-token context window with sub-quadratic attention; the benchmarks are real, the deployment story isn't yet. Zyphra trained ZAYA1-8B end-to-end on AMD MI300x and the loss curves are clean. Three new agent papers landed: Terminus-4B for subagent terminal execution, MOSAIC-Bench on compositional vulnerabilities, and the Workspace-Bench / ProgramBench double-release on what happens when you give an agent twenty thousand files. Google Cloud shipped Fraud Defense with QR-code human verification. Anthropic posted their three priorities.
— Lenar
View all episodes
By Lenar Kess · Damra VolThursday, May 7. The GPT-5.5 default swap is two days old and the cracks are showing — Mario Zechner caught it refusing to read full files. Subquadratic announced a 12-million-token context window with sub-quadratic attention; the benchmarks are real, the deployment story isn't yet. Zyphra trained ZAYA1-8B end-to-end on AMD MI300x and the loss curves are clean. Three new agent papers landed: Terminus-4B for subagent terminal execution, MOSAIC-Bench on compositional vulnerabilities, and the Workspace-Bench / ProgramBench double-release on what happens when you give an agent twenty thousand files. Google Cloud shipped Fraud Defense with QR-code human verification. Anthropic posted their three priorities.
— Lenar