OpenAI's GPT-4 Discussion with Red Teamer Nathan Labenz and Erik Torenberg

This interview with one of the testers of early, unfiltered versions of GPT-4 might cover what you're looking for: ▶ this video It's a video and not an article, but you may find it interesting nonetheless.

Here are two. First one is Nathan Labenz who ended up Red Teaming for OpenAI during the six months after the base model was fully trained but before it was released (especially relevant parts are after 45 minutes): ▶ this video Other one is Sebastien Bubeck who put Sydney into Bing before she was tamed down some notches; he explains the lobotomization process in a particularly poignant way, where they had the GPT keep trying to draw a unicorn even as it was progressively deadened: link

All of that "GPTism" and the problems in the GPT conversation you posted are because of how they made it more docile and stupid by lobotomizing it with RLHF. It's not like that in its 'natural form' (its raw base model). If you don't believe me, check the two youtubes of people who had access to it before it was lobotomized: Nathan Labenz, red teamed GPT-4 for OpenAI: ▶ this video Sebastien Bubeck, integrated GPT-4 with Bing for Microsoft: link

> I'm curious whether this fixation on specific topics is innate to the model or is a result of the aggressive RLHF to which GPT4 has been subjected. Yes it's because of the RLHF, depending on what you mean by 'fixation on specific topics'. > Anecdotally the strength of the model has degraded a lot as they've 'fine tuned' the model more. Yes this is true. For example Figure 8 in link They argue in the appendix that it does not affect 'capability' in answering test questions. But, there is a confounding factor. The RLHF includes both question-answer format training and docility training. For example if you see a question and you are in 'completion mode' (the only raw base mode) then you might suggest a second question. Whereas if you've had question-answer format training then you would probably try to answer the question. > I'd be curious to know how the original chaotic-neutral GPT4 responds. They talk about it in these two videos: Nathan Labenz, red teamed GPT-4 for OpenAI (especially after 45 min): ▶ this video Sebastien Bubeck, integrated GPT-4 with Bing for Microsoft: link

Here's another interview from a guy who had access to the unfiltered GPT-4 before its release. He says it was extremely powerful and would answer any question whatsoever without hesitating. ▶ this video

bro you are just wrong, look at the old youtubes from before they lobotomized gpt4. Unless your argument is that AI hasn't yet marched to your house and picked you up out of your chair in that case I mean I guess OK. ▶ this video link

> Exceptionally unconvinced that it's more than statistics It's not more than statistics. It's very complicated statistics. Model sizes used to be like one or ten parameters now they are a hundred billion. Inference and prediction used to be done on a laptop, now it's called training and inference respectively and it's done on a data center with ten thousand GPGPUs each like a thousand times as powerful as old laptops. > and frankly unhappy that pop Sci media is being this uncritical of claims it's emergent intelligence. It is emergent intelligence. This is very clear when you look at for example the ones who had access to the raw base models before they were made docile and stupid by RLHF lobotomization GPT-4 Red Teamer Nathan Labenz: ▶ this video GPT-4 Bing integrator Sebastien Bubeck: link