Hacker Newsnew | past | comments | ask | show | jobs | submit | eoinbmorg's commentslogin

Is RAG the right tool for this? My understanding was that RAG uses vector similarity to compare queries (the extracted string) versus the search corpus (the PDF file) using vector similarities. The use case you describe is verification, which sounds like it would be better done with an exhaustive search via string comparison isntead of vector similarities.

I could be totally wrong here.


Some people define RAG as having to use vector search, others (myself included) define RAG as any technique that retrieves additional relevant context to help generate the response, which can include triggering things like full-text search queries or even grep (increasingly common thanks to Claude Code et al).


RAG is just "Retrieval Augmented Generation", vector similarity is one way to do that retrieval but not the only. Though you are right, there is really no retrieval step augmenting the generation here, more like just a validation step stuck on the end.

Though I imagine scenarios where the PDF is just an image (e.g. a scan of a form), and thus the validation would not work.


The notion of 'borrowing' an e-book, which has no intrinsic limit to number of times it can be checked out or when it must be returned, is a joke.


I do the same in the Ergodox QMK firmware, no app needed on the device. Just mentioning it since OP has the same keyboard!


The thing with cooking meth is that you don't have to be comfortable putting it in your body. You just have to be comfortable selling it to someone else who will put it in their body! Unfortunately, for many meth producers/dealers that bar is not very high.


Maybe at the top of the food chain, but a lot of lower level people are addicts working to support a habit by pinching from wholesale


"verifying" signatures is an inherently insecure practice anyway. It's trivial to copy a signature any many people (like me) have poor writing that varies each time anyway. It's more about showing intent than verification.


Not really.

If you signed something and cannot make it look reasonably like what I understand your signature to look like, I have a really big red flag telling me you aren't who you say you are. Then I then have the option of denying you, or investigating why you can't match your own signature, and looking at other available proof of your ID.

People's signatures may change a little over time, but there are clear muscle-memory related aspects that don't disappear. It's the most basic aspect of handwriting their is - your own name. Usually it's the first thing you even learn how to write. If you can't do that, then you definitely need to carry plenty of other forms of ID that don't need your signature. And even your driver's license has that. Outside of physical or mental damage, there is very little excuse in the world for not being able to duplicate your own signature satisfactorily.


Out of curiousity: are you Irish? These patterns match some of my relatives, particularly ending sentences with "filler" words and the "th" pronounciation. I never thought about these as things that need fixing since they are just part of an accent (not suggesting you shouldn't change them if you feel like you'd prefer a different speech pattern).


I'm not Irish in any meaningful way (a great-great grandmother). I'm an Australian from a regional area.

It's interesting that there's an overlap!


I got reasonable offers from 2/3 companies I went on-site with from Triblebyte. I found that their stable of companies had too many tiny startups with questionable (read: boring/dumb) product ideas. I wanted a company with 100-300 employees so not many of their partners met my criteria. I found their recruiting partner that works with you kind of annoying... I didn't appreciate them texting me/"checking in" all the time. It's bad enough to deal with the company recruiters doing this, I don't need another cook in the kitchen. Overall, I didn't think it was a waste of time but I don't think I'd use it again now that I know that they mostly represent smaller companies with advertising/fintech aspirations.


sitting on the floor requires a pretty high degree of hip and hamstring flexibility to avoid rounding the pelvis under and introducing poor posture. You might be better off taking breaks to focus on improving hip mobility and work up to a sitting desk option as a longer term goal.


Small counterpoint: many people who buy instruments want to buy the actual instrument they try out. Quality in guitars (and other instruments) is wide enough that one might play well, while another of the same model plays like a worse, cheaper instrument. Buying a guitar untested/online (even the same model you tried in store) has a large amount of risk that decreases with the instrument price, but not always and not as quickly as you might expect.


I don't know enough about Benford's law to draw conclusions, but is N=O(1000) a large enough sample size to expect it to apply? My intuition suggests that if county size does not span enough orders of magnitude then Benford's wouldn't apply because the distribution might lie in one of the "spikes" of the Benford curves, rather than reflecting the average. For example, if many counties had a population size starting with a "4", you'd expect more vote counts to start with "2" (presuming each candidate received close to 1/2 of the votes).

Hopefully someone with a stronger math background can expand in a more fluent way :)


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: