Comment by tikkun | Hacker News Reader

Comment by tikkun | Hacker News Reader

Parent Full thread

I made a summary of this article, because I found it a bit hard to read. I used GPT-3!

Here's the summary:

The author is saying that people who write about productivity are, in a way, the devil, because they convince people to do things they hate. The author is also saying that these people are not literally the devil, but are just as bad.

The article argues that humans are not "naturally" lazy, and that the feeling of being a "lazy piece of trash" is actually a result of the unconscious mind doing its job. The unconscious mind is only able to communicate with the conscious mind when there is a problem that needs attention, which can make it seem like the unconscious is lazy. However, the unconscious mind is actually responsible for solving a lot of problems without the conscious mind even realizing it.

The author is talking about how people who are successful often have to eat a lot of frogs (do a lot of things they don't want to do), but that doing too many things you don't want to do can lead to burnout. The author suggests that people should stop and think about what they really want to do, and not just do things because they think they should or because other people are doing them. The author also talks about how some people never get to experience true love because they're stuck in bad relationships, and how this can be just as bad as eating too many frogs.

This is a shockingly coherent summary. The accuracy leaves a bit to be desired, but this is perfectly usable for distilling things you don't have time to fully evaluate.

I felt the same. I read the full article. Then, I returned to HN. I was surprised to see the top post was talking about GPT-3(!). Then, I read the GPT-3 summary. I think: "Hey, not bad! This could be a real tool for everyday use!"

I had no idea GPT-3 could be used for this, this accurately.

I suddenly feel like I need a browser extension to add such a summary to every article I read. Especially for less newsy, more long-form "conceptual" pieces where the author often doesn't make their actual real point until halfway through.

As someone who has not "used GPT-3" what does it mean to say that you used it to make a summary? Just dump the full text of the article into a tool and request a summary output?

I have only seen GPT-3 used when it takes a prompt and then generates more related text, I didn't realize you could run that process in reverse.

The main interesting thing about GPT-3 is that prompt engineering turns it into an incredibly general tool. This summary was likely generated with a prompt something like:

Here's an article about x: <text of sample article>

A quick summary of the article, focusing on the main relevant points and keeping critical detail: <sample summary>

as an example, then duplicating that with the real article and text.

That's the "few shot learning" from the original paper. It could also be that it's good enough at summarizing specifically that you don't need an example, just the right framing prompt around the article text. Either way, that kind of prompt engineering is how you get "text completion" to perform basically any text processing task, generate code or play tic tac toe, so on

Not OP but I've used GPT-3 quite extensively and can imagine that's exactly what they did.

You can give it anything - or nothing - and ask for any kind of output.

The relevance and accuracy of what it gives you can vary lots. You can give it examples of the types of things you want back etc. You don't necessarily have to give it examples either, just ask it a question. It normally performs better on semi-complex tasks when you give it examples though.

> I have only seen GPT-3 used when it takes a prompt and then generates more related text

This is a fully general technique. Any repeatable problem can be rephrased as a prompt + related technique via the following structure:

  Problem: <Example of Problem>

  Solution: <Example of Solution>

  <...maybe include one or two more examples>

  Problem: <Real Problem>

  Solution: <Ask GPT-3 for the solution>

If you don't have one hour to meditate, meditate for two hours.

If you find the article hard to read, consider listening to it. It will take longer but it will be easier. The author's reading of the article has a little extra humor too.

Is there a firefox extension that can give me a GPT-3 summary of any given page? That would be super useful.

If GPT-3 can coherently summarize a lengthy article with plenty of metaphors, and even explain those metaphors, does that not constitute "understanding" (that many people claim it cannot possess)?

Reply on news.ycombinator.com