Comment

Comment by 0xDEAFBEAD

I think "unwilling to divulge company secrets" is the best explanation here.

We know that OpenAI does a staged release for their models with pre-release red-teaming.

Helen says the key issue was the board struggling to "effectively supervise the company": https://nitter.net/hlntnr/status/1730034017435586920#m

Here's Nathan Labenz on how sketchy the red-teaming process for GPT4 was. Nathan states that OpenAI shut him out soon after he reached out to the board to let them know that GPT4 was a big deal and the board should be paying attention: https://nitter.net/labenz/status/1727327424244023482#m [Based on the thread it seems like he reached out to people outside OpenAI in a way which could have violated a confidentiality agreement -- that could account for the shutout]

My suspicion is that there was a low-level power struggle ongoing on the board for some time, but the straw that broke the camel's back was something like Nathan describes in his thread. To be honest I don't understand why his thread is getting so little play. It seems like a key piece of the puzzle.

In any case, I don't think it would've been right for Helen to say publicly that "we hear GPT-5 is lit but Sam isn't letting us play with it", since "GPT-5 is lit" would be considered confidential information that she shouldn't unilaterally reveal.

parent 0xDEAFBEAD 2 years ago

Discussion

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton