Anthropic Mythos - Too Dangerous? Hype vs Reality

In 2019, OpenAI said that GPT 2 was too dangerous to be released. Now, we’re seeing similar sayings with the new Anthropic model, Mythos.

The justification for why it is partially fulfilled today: they predicted that GPT 2 could increase the number of AI slop in the internet, and a few years later its impossible to use the internet without consuming slop.

But it was largely hype and marketing. Nobody will release something truly dangerous for 20 usd / month for anyone to buy.

Those frontier labs do a lot of hype and promises, and they do deliver to some extent. Sometimes its hard to differentiate noise from actual data, but we’ll see in the near future if Mythos is actually dangerous.