Thomas Fairfield

They said they tried using a cloud frontier model, but it refused cybersecurity questions due to safety guardrails. Both Anthropic and "Open"AI are known for never releasing frontier models for local, open weight usage. HF resorted to GLM as it is a local open weight frontier model

Really interesting, especially with what we now know. Hopefully, this convinces the US Feds that open source models are a must. I can see the responsible party somehow twisting this to mean we need more regulation however.