Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: HackYourNews – AI summaries of the top HN stories

hackyournews.com

Post reply on HN
Print view View original

51–60 of 170 posts

Re: Show HN: HackYourNews – AI summaries of the top HN stories

#51

Post by remexre »

jarulraj wrote: post #45
remexre wrote: post #44

Earlier quoted context omitted.

Probably not the point, but shouldn't you be able to choose to sample only the tokens for "positive" and "negative" (they're both one token!) instead of (or in addition to) needing to put a request for model to restrict its responses in the context?

Interesting observation :) I guess this is the SQL query you have in mind that uses the LIKE operator: SELECT ChatGPT("Respond to the review with a solution to address the reviewer's concern", review) FROM postgres_data.review_table WHERE ChatGPT("Is the review positive or negative?", review) LIKE "%positive%" AND location = “waffle house”; From a query processing standpoint, both queries should have equivalent perfo…

mm, no, not unless you're doing some LIKE-specific optimization (and even then, I think you'd want "positive%").

So like, at the end of all the decoders, the model gives you an output vector; you multiply this by your embeddings to get your token probabilities, then you sample from them to choose a token.

Instead of sampling, you could just look at the probabilities for the tokens "positive" and "negative" and return whichever of those two is highest.

Re: Show HN: HackYourNews – AI summaries of the top HN stories

#52

Post by djha-skin »

As can be expected, we find that AI consistently chokes on lisp programs. Here is the description for number 8, John McCarthy's lisp program with a bunch of numerical conversion information in it:

> This file contains a collection of numerical facts and units conversions, as well as astronomical facts related to moving Mars to a more temperate location... The content is not particularly relevant to the current technology landscape or contemporary societal issues, as it seems to be more of a historical piece. It is not particularly innovative or groundbreaking, as it primarily provides basic numerical facts and conversions. The technical depth is moderate, as it provides detailed units conversions and constants. The clarity of the information is good for those familiar with Lisp, but may be less accessible to the general public. The content does not consider a diverse range of perspectives or voices, as it is focused on numerical facts and conversions. It does not appear to be research-backed or cite reputable sources. The impact of the discussed topic is minimal, as it primarily provides basic numerical facts and conversions. The timeliness of the content is unclear, as it does not specify when it was created. The engagement of the content is likely low, as it is a technical program rather than a piece of content that inspires discussion. The reliability of the source is uncertain, as it is a personal file created by John McCarthy. There is no self-contradiction in the content. The discussion quality is not applicable, as there are no comments.

Re: Show HN: HackYourNews – AI summaries of the top HN stories

#55

Post by jarulraj »

remexre wrote: post #51
jarulraj wrote: post #45

Earlier quoted context omitted.

Interesting observation :) I guess this is the SQL query you have in mind that uses the LIKE operator: SELECT ChatGPT("Respond to the review with a solution to address the reviewer's concern", review) FROM postgres_data.review_table WHERE ChatGPT("Is the review positive or negative?", review) LIKE "%positive%" AND location = “waffle house”; From a query processing standpoint, both queries should have equivalent perfo…

mm, no, not unless you're doing some LIKE-specific optimization (and even then, I think you'd want "positive%"). So like, at the end of all the decoders, the model gives you an output vector; you multiply this by your embeddings to get your token probabilities, then you sample from them to choose a token. Instead of sampling, you could just look at the probabilities for the tokens "positive" and "negative" and return…

Yes, it should be "positive%". That's an interesting LLM-level optimization.

Doesn't this token sampling optimization require using a locally-running model like Llama?

I am presuming that OpenAI doesn't provide direct access to token probabilities in its API.

Re: Show HN: HackYourNews – AI summaries of the top HN stories

#56

Post by ukuina »

JimA wrote: post #48

Great idea, I would also regularly use this with a few tweaks: 1) Main one - Separate stories more clearly. Maybe slightly outdent, enlarge, or bold the title, or add a little more space between stories so they don't all run together. 2) Some articles don't have content available for no apparent reason (Deciphering Mary Stuart’s lost letters from 1578-1584) 3) Filters would be nice (articles >rating, comment count, e…

Thank you for trying it out and the feedback.

1) I have increased the spacing between stories. Also, have you tried the Mobile view?

2) Looks like some websites are harder to parse than others. Some users have also reported similar issues with GitHub/Gists. Will investigate.

3) I've removed the rating for now. Will add filters to the roadmap.

4) I'm currently showing all 30 articles from the HN frontpage. Would you prefer fewer articles, but paginated?

Re: Show HN: HackYourNews – AI summaries of the top HN stories

#57

Post by ukuina »

entropie wrote: post #42

> The frontend is pure HTML+CSS. I wondered why there is a special URL for mobile, since this could be easy done by media queries. On a side note; I desperately need a dark mode and even installed an extension to make HN dark mode compatible.

The Mobile view honors your OS dark mode setting!

I need to read up on media queries, thanks.

Re: Show HN: HackYourNews – AI summaries of the top HN stories

#59

Post by ukuina »

rexpop wrote: post #49

> Overall, the comments provide a mix of technical insights, personal experiences, and opinions on the topic. Yeah, I'll bet. At first blush, I find this uselessly verbose. I get more out of just the headlines and I am already up to my eyeballs in plausible, zero-content boilerplate. I think our whole society has been since the 90s, frankly. Generally, filtering public comments through a language model reduces signal…

Haha, some of the summaries are quite bland, indeed. I'm working to add some character back to it!
Post reply on HN