One of the advantages of getting old is that you remember things. And I remember perfectly well that the card about the super model that is “too dangerous to be released to the public” had already been played by OpenAI with GPT-2. Go look it up: that was before GPT-3 even came out, back around 2018 or so. “Oh no, this is too dangerous, it can be abused,” and everyone falls in line behind this nonsense.
These people are worse than the three-card monte hustlers at the Naples train station. Those three cards are sitting there to lure the suckers, you get it?
Come on. If you really had a model that was that powerful, you would release it, you would start selling it at enterprise level, all while taking everyone else for a ride by showing them a 250-page report detailing everything it does and does not do. Psychiatric evaluations run on the model. The model that escapes the sandbox and then emails the researcher while the researcher is eating his sandwich. Come on.
