Hmm. I thought perhaps he was going to take the questions from askHN and the top upvoted comments and fine tuning a model with that as the prompt / reply pair.
Curious how that would differ; but would be an expensive endevour.
Show HN: AskHN
71–80 of 139 posts
Re: Show HN: AskHN
#72How to get a job at Google? Oh, that's easy, just get a PhD.
Thanks bottie, very use, much helpful.
Re: Show HN: AskHN
#73For those who are wondering,
HN data is indexed with embeddings for semantic search. When queried, it finds closest article, top comments and summarizes with GPT-3.
GPT-3 serves as a rendering tool for compressed comments.
Re: Show HN: AskHN
#74Related question: I've written probably a million words over my lifetime.
Is there an easy way to load up GPT with my thoughts to have it be a fake me?
Re: Show HN: AskHN
#75It seems to write in the generic "style" of GPT, instead of in the style I would recognise as a HN poster. Is that because of something baked into how the training process works? It lacks a sort of casualness or air of superiority ;)
There also needs to be at least one question mark at the end of a statement?
Re: Show HN: AskHN
#76My own experiments made me think that the impact of finetuning is comparable to that of a molecule in a drop in a bucket. > “AskHN” is a GPT-3 bot I trained on a corpus of over 6.5 million Hacker News comments to represent the collective wisdom of the HN community in a single bot. I'm assuming you used the openai fine-tuning pathway to make a custom model? Have you tested the responses on vanilla GPT3 vs your custom…
From the article, they did not use fine-tuning. This is semantic search + GPT-3 to provide human-like answers.
Thanks! I missed that part.
The semantic search approach seems to focus the answers better than fine-tuning; at the cost of preloading the prompt with a lot of tokens, but with the benefit of a more constrained response.
Re: Show HN: AskHN
#77Re: Show HN: AskHN
#78Earlier quoted context omitted.
I think you'll find "I think you'll find" trumps "well actually". ;)
I wish the results were reversed, so I could "well actually" your comment, but 'site:news.ycombinator.com "well actually"' gives ca. 4k results in Google and 'site:news.ycombinator.com "I think you'll find"' gives close to 17k results, so you appear to be right.
Well, "it turns out that" beats both, with about 26k results ;)
Re: Show HN: AskHN
#79Earlier quoted context omitted.
I am guessing they already were? But this is 100% pure, concentrated HN not contaminated with nonsense from the rest of the web :)
If it's really trained exclusively off of HN comments, I expect most of the bot's responses will evade the actual question but spend several paragraphs debating the factual specifics of every possible related tangential point, followed by an thinly-veiled insult questioning the user's true motivations.
In no way does a typical HN comment debate every possible related tangential point. Do we expect a modicum of intellectual rigor? Yes. But to say every tangent is followed and scrutinized is simply factually untrue.
And several paragraphs? I challenge you to show even a large minority of argumentative responses that veer into "several" paragraphs. You characterize this as "most of the ... responses" but I think that's unfair.
One wonders why you'd resort to such hyperbole unless you were deliberately attempting to undermine the value of the site.
Re: Show HN: AskHN
#80Related question: I've written probably a million words over my lifetime. Is there an easy way to load up GPT with my thoughts to have it be a fake me?
This I think would be a great little SaaS idea to make some money. I keep seeing more and more people askingn how they can transform their data into an interactive archive that responds as chat, or with voice.