Was the Mythos Ban Justified? (Good Idea. Bad Execution.)

The gist

1. Newport says Mythos was mostly marketing: earlier models found similar vulnerabilities and smaller models replicated its results, so it was no revolutionary leap.
2. AI safety guardrails have been jailbreakable since GPT-3.5 by using obfuscated prompts, so this is not a new or unique problem.
3. Newport argues AI companies should face mandatory pre-release government safety reviews, the same way standard consumer product manufacturers are held accountable.
4. Newport says trillion-parameter frontier models are built for brand hype, and narrower 50-billion-parameter models can do the same work more cheaply and safely.