Coderabbit for PR reviews at work has both impressed me and made me aggregated by how incorrect some of the comments are. It’s like, it’s caught bugs that i would have missed even when looking very closely but also makes the same suggestions to over complicate chunks or suggestions that literally don’t work. Such as assuming the db schema when looking at a query and saying “that’s not what the column is called”.
So, that’s best I’ve experienced really, basically a PR check that’s able to find some really out there bugs but a lot of comments need to be ignored.
1.) A convoluted way to search for sources about a thing I know nothing about. I looked up how property deeds work and gemini gave me a rough summary and enough links to actually find out what I needed to know. For anything I do know about, just plain google suffices.
It works well for small programs and boilerplate. But you need to know what you are doing to guide it. They can very often get stuck in some rabbithole
I concur, it has gotten me in multiple wild goose chases when debugging. It is extremely confident especially when it’s wrong
It’s pretty good at fleshing out documentation. and if you’re naming variables properly it’s pretty good at gleaning what you’re trying to do and autocomplete on a small scale.
It is extremely confident especially when it’s wrong
This reminded me of a recent study I saw posted about the accuracy of AI and trying to remove hallucinations. One of the conclusions of which being, besides that it’s impossible to stop them from hallucinating, that the tests companies use to grade the quality of an AI and the expectations of users grade confidence in an answer higher than the accuracy of the answer.
I’ll be honest, it’s probably wasted more time than it’s saved me. I only trust it to format files and find where things might be used in the code base. So, you know, plain-language pretty print and grep.
Has there been any good consumer Ai products yet? I keep seeing all these products in search of a problem.
Coderabbit for PR reviews at work has both impressed me and made me aggregated by how incorrect some of the comments are. It’s like, it’s caught bugs that i would have missed even when looking very closely but also makes the same suggestions to over complicate chunks or suggestions that literally don’t work. Such as assuming the db schema when looking at a query and saying “that’s not what the column is called”.
So, that’s best I’ve experienced really, basically a PR check that’s able to find some really out there bugs but a lot of comments need to be ignored.
Yeah you need to review the PR review instead lol
I do, but it catches really weird bugs. And dismissing comments is definitely less work than figuring out the bug later when it shows up.
Better text to speech and speech to text, automatic image descriptions, better translations.
ChatGPT and the likes can also be used to access content without ads nor license
Also you can write the letters “AI” on a paper and rich people will give you money. They’ll want it back eventually but hey
They can ai deez nutz
Besides improving grammar in emails, no, not really
Do they have an AI program that filters out AI content yet? That might be good.
1.) A convoluted way to search for sources about a thing I know nothing about. I looked up how property deeds work and gemini gave me a rough summary and enough links to actually find out what I needed to know. For anything I do know about, just plain google suffices.
2.) Vulcan straight man for comedy routines.
A few developers I know are very impressed by Claude.
I, too, can copy and paste from StackOverflow.
Yes, but you’ll know if you’re copying from the answer, or the question.
Even if that were literally what it did, having a StackOverflow button would be pretty cool
May I introduce to you: https://github.com/drathier/stack-overflow-import
It works well for small programs and boilerplate. But you need to know what you are doing to guide it. They can very often get stuck in some rabbithole
I concur, it has gotten me in multiple wild goose chases when debugging. It is extremely confident especially when it’s wrong
It’s pretty good at fleshing out documentation. and if you’re naming variables properly it’s pretty good at gleaning what you’re trying to do and autocomplete on a small scale.
This reminded me of a recent study I saw posted about the accuracy of AI and trying to remove hallucinations. One of the conclusions of which being, besides that it’s impossible to stop them from hallucinating, that the tests companies use to grade the quality of an AI and the expectations of users grade confidence in an answer higher than the accuracy of the answer.
I’ll be honest, it’s probably wasted more time than it’s saved me. I only trust it to format files and find where things might be used in the code base. So, you know, plain-language pretty print and grep.
Is there any smaller, “semantic search”-oriented model, rather than all the coding agents?
I think Adobe’s new generative fill is supposed to be pretty good? It’s just a more complicated version of their much older content-aware fill.