Latent SpaceFREE

Andon Labs founders discuss building frontier evals for Claude models from Haiku to Mythos. They introduce VendingBench, a new evaluation framework for assessing AI capabilities. The conversation highlights the challenge of creating lasting benchmarks that remain relevant as models improve.

DockerFREE

Docker explains hardened images, a strategy to reduce container vulnerabilities and attack surface. Security scans often reveal hundreds of CVEs originating from base image packages like shells, compilers, debug utilities, and libraries, not application code. Hardened images address this by minimizing included components, leading to fewer known vulnerabilities and a more secure software supply chain. This approach helps teams focus on application-specific security rather than managing extraneous package risks.