Turns out, SOTA llms are great at finding and fixing bugs in code. Too bad companies like Anthropic are doing everything they can to stop these reprehensible acts of safety.
They heavily implement guardrails where you can ask about anything cybersecurity related and Anthropic constantly talks about the risk of openly releasing models without aggressive AI safety rules, while only letting a small group of organizations like Apple have access to them. Their own fear mongering around this safety stuff with Mythos led to Fable being briefly export controlled by the US government, despite Fable already being kneecapped for security research.
Turns out, SOTA llms are great at finding and fixing bugs in code. Too bad companies like Anthropic are doing everything they can to stop these reprehensible acts of safety.
What is the source for the claim that “companies like Anthropic are doing everything they can to stop” “finding and fixing bugs”?
I assume they’re referring to Fable’s guardrails?
They heavily implement guardrails where you can ask about anything cybersecurity related and Anthropic constantly talks about the risk of openly releasing models without aggressive AI safety rules, while only letting a small group of organizations like Apple have access to them. Their own fear mongering around this safety stuff with Mythos led to Fable being briefly export controlled by the US government, despite Fable already being kneecapped for security research.