Are Large Language Models Like GPT-3 a Hype?

Question

LLMs keep getting bigger and bigger, but I still don't understand the value they provide to businesses and individuals. There must be a reason for MS to purchase the GPT-3 license for $1b, and Meta, Google, OpenAI, etc. wouldn't spend millions of dollars just to train these huge models. Yet, I don't get it.IMO, "AI through API" is a bad practice that sets a wrong precedence for the incoming AGI. It's basically a black box wrapped around another black box. In addition, the cost of GPT-3 and other models may be prohibitive for small businesses and startups.What is going on here? I see some similar patterns between LLMs and cryptocurrencies: both are super hot and receive billions of dollars, but neither has proven its use cases. Given the crypto bubble burst, are we going to witness an AI winter in the next few years?

FrenchDevRemote · Accepted Answer

Zero shot classification on any random task seems like a pretty huge deal to me.
Copilot seems like a pretty clear use case that is very useful, Copilot is currently writing 10-20% of my code, sometimes more, sometimes I don't even have to read the docs or go on stackoverfow, for example I didn't know much about AWS S3, started writing a class with copilot, and autopilot +/- autocompleted a whole 150 line file, the only thing it got wrong was an uppercase in a table name(which wasn't mentioned anywhere and that it just guessed!!)
Automated description of pictures, seems like it would be pretty useful to blind people once it can be done reliably in +/- real time/on video.
I've been thinking about using GPT or Bloom to take my job search automation to the next level: filter companies that match my criteria(using more than simple metrics that can filtered with regular code), for example I want companies that works against climate change, with good remote policies/remote first, a focus on code quality, I want to avoid certain technologies etc...
generate cover letters by using the job description+data about the company+a summary of my resume(or even the recruiter profile) as input
If you think it have no use cases you just have no imagination

beernet · Answer

A big part of the business claim lies in the generalizability. LLMs are basis technology that, once trained, have applications in practically all sectors and businesses, because text and data of other modalities are everywhere. Massive scalability potential.
Just to name a very short selection of apps that are very likely to be tranformed by LLMs, or are in the process already:
* chatbots and everything conversational
* QA systems, customer support
* writing and grammar assistants
* Code generation (Copilot etc.)
* Translation
Each of these has a billon dollar market, and you might be able to solve them all with the very same model. That is the bet.

solomatov · Answer

I think LLMs are pretty useful already. For example, the gmail feature where it suggests you a phrase is likely implemented via language model, and is very widespread on other applications too.