What are the models useful for?

As best I can tell the only thing would be some kind of GMail-like auto-reply which a human consciously edits, and which has little consequence if it's wrong.

Are the models useful for customer service? Like reading a manual or knowledgebase and then answering a customer's questions about a product, and troubleshooting problems? Like about your Android phone, which lots of people have trouble using?

That would be a trillion dollar business. As best as I can tell that's completely beyond GPT-3 and requires a huge breakthrough, which may or may not happen.

GPT-3 involves zero fine-tuning or customization to any purpose. It’s not attempting to be a product, but rather a platform people can use to explore the possibilities of products. With fine-tuning it could do all sorts of domain-specific things that it can only passably do now. We suspect this to be true because we’ve seen how other systems behave with and without fine-tuning.

you are sidestepping the question above - what can it be used for? What kind of fine-tuning is required? Because to me it looks like 'fixing catastrophic errors' not fine-tuning - as anywhere where the text can deliver value, it has to be correct.

AWS/Azure are a platform, and they work reliably. What kind of platform 'sometimes works'?

Language models can help the disabled communicate better and faster (fine-tune on their online output, then the BCI can offer better continuations, so the patient does not have to type as much).

Fine-tune on educational materials, and a language model could be a 24/7 assistant to students.

Fine-tune on psychology or medicine data, and a language model could tell you of all medicine interactions, or act as a better ELIZA, and "socialize with" - and "support" people with depression or trauma.

Fine-tune on etiquette and social norms, and autistic people could ask questions without being ashamed.

These are some positive use cases. There are also neutral use cases (seeding a new social platform with autogenerated comments), and negative use cases (SEO spam, fake reviews, scaling up disinformation campaigns).

Edit: not complaining, but no idea why this was downvoted. Would be helpful to state why, so I don't make this mistake in the future (I am optimizing for useful replies).

I did not downvote, but I think it's because you kinda missed the crux of my question is - what is fine-tuning, like what do you have to actually do?

Has anyone done it and demonstrated that it can be done on GPT to get useful and reliable output for some domain? If we exclude fake news/ twitter-bots or something equally silly.

> what is fine-tuning, like what do you have to actually do?

For GPT-2 it is as easy as creating a corpus you want to fine-tune on, adding new unseen tokens to the dictionary, and run a finetuning command. https://huggingface.co/transformers/v1.2.0/examples.html#fin...

> Has anyone done it

Many have. AI Dungeon is a great example. Others are experimenting with humor generation, short story synopsis generation/creative writing, and supportive chat bots.

Because when people say “AGI is near, just look at GPT-3,” it’d clear that we’re in a really good version of Searle’s chinese room. The lack of understanding is the important point.

I don’t recall any strong argument that Searle’s Chinese room can’t be an AGI, just that it can’t be conscious.

Certainly consciousness and understand are central to Searle’s argument. However from my perspective (particularly someone who is critical of all the recent DNN advances as being some harbinger of AGI), if all you’re doing is looking up replies from a dictionary, then you have no capacity to generalize, learn new things, adapt, empathize, have memory that isn’t already pre-computed and pre-allocated.

Now there are ways in which thru fine tuning etc that you can take GPT-3 and have it adapt in some way, thru such adaption or “context/attention networks” give it “memory,” in practice neither of these looks anything like AGI because right now these pages of this Chinese dictionary don’t have any relationship to each other. It becomes clear both in this article (you shouldn’t need some “correct” prompt to get it to make sense), and clear with long form generation that there isn’t a deeper understanding of the meaning of these words. I will say it’s very impressive what it can do when it does make sense, shockingly so, but we are very much in the position where we didn’t expect “chinese” to come out of the machine at all and thus we are projecting onto its outputs an anthropomorphisis that is unwarranted.

To me you are also just a very good version of a Chinese room.

How would you show me that you are not?

You can look up my comment history and find a continuity and a perspective mired in personal experience that a machine cannot (at least currently) replicate.

That proves that you are not GPT-3, but you could still be a Chinese room.

You may be even a biological human and I still won't know if you are a Chinese room.

Maybe since birth you are just sort of sleepwalking? I wouldn't be surprised if there were fully functional people without qualia.

If I am, I’m the best Chinese room created yet.

I guess you are :) . I will tell you a secret. I suspect I'm a Chinese room too ^^ ...

There are also people with so called total locked-in syndrome i.e. people who can not do any voluntary movements even with their eyes, so there is no way to communicate with them. How would you decide if they are conscious? It may be literally matter of live and death for them. How should we know if they even want to live in their condition?