@isodoublet
"You're prompting it wrong!!" -- Steve Jobs, 2009
@InfiniteQuest86
Prime defends AI by saying it's good because of the way no one uses it. Literally everyone just generates code and trusts it. Which is what this article is pointing out is terrible. Prime is like no one would do that. You are wrong. They do that.
@TennSeven
The "bro doesn't even know how to spell skeptical" person is an idiot. "Skeptical" and "sceptical" are both correct. The latter is how they prefer to spell it in the UK.
@dgo4490
It doesn't t have to make money, just move money around.
@kavajawa
I wonder how many people voting for the skill issue wrote a single program more advanced than Fiz-Buz…
@gsgregory2022
At the 16:50ish mark. I think you missed the point. Yes an llm will just output an answer, however the point is that cursor a tool built on top of ai, doesn't do much more than implement the answer from the llm.
@McZsh
A week and a half ago, I had a review with one of our junior guys, he called me he is ready to commit, but the darn thing won't build. Fast forward an hour, I sit next to him and watch him do what he wants to achieve. Build process fails with a message of a missing brace, file and line give. We delve into the editor, the line has a red squiggly with a message "brace missing". I point at it and say "there it is". Junior copies the whole line into the prompt and asks "what is wrong with this code?" LLM generates half a novel of an answer , bottom line "missing brace" and returns the line, with a brace added, 7 letters in total longer. Copy, paste, build. Junior thanks me, full of himself. Real dev he his, right!? I stared on him for a full 10 minutes and then mangled him on his own code for a full three hours. It was total and utter horseshit. So, bottom line... the criticism of LLMs and their perceived vs their real world usage is totally correct. And I'll add a question: "Is AI generated crap without any barriers and best practices applied what we developers as a profession want to profess to the next generation, and to the customer, indeed?" (The original question is from Robert C. Martin, I just added the AI perspective)
@kavajawa
Best way to shut down any AI programming booster? “Show me the code!” For some reason they never show…
@pchasco
The biggest takeaway is that if vibe coders take over, we are all fucked. When you have to very explicitly spell out the requirements down to the smallest detail in order to expect a correct result, vibe coders have no chance in hell of creating anything more than throwaway AI slop.
@0xCAFEF00D
16:50 He keeps missing the nuance in this critique of _cursor_. It's not about the LLM just guessing it's way to the fact that you need to be more specific in the prompting within every query. It's cursor that should allow the LLM the space to validate this. The utility of cursor is severely deminished if it just tosses the problem to an LLM and naively accepts the output. It's a poor integration.
@k98killer
Cursor completely broke during an update and didn't have a rollback feature, so I had to uninstall and reinstall an outdated version to be able to continue working. The fact they downgraded paying subscribers to Free Tier rate limiting without explanation made me dislike Cursor substantially after having been a happy customer for about 6 months. I also found myself turning off the tab complete and not using the chat features because the suggestions were a net negative. This led me to drop Cursor for regular old vim in April. Haven't gone back since, but that is likely because I am not currently working on web apps, which is where Cursor shines.
@Dom-zy1qy
Net negative productivity happens when you use "Agent" mode and Sonnet's logits (the voices in its head) tell it to refactor the entire codebase, implement random design patterns, rename classes, methods, and update comments when you ask it to make a minor change to one file.
@Rallion1
Saying it's not a cursor problem because it's an AI generation problem is nonsensical. Cursor explicitly exists as a tool to add AI generation to your coding workflow. It lives and dies on the quality of the AI generation.
@lazyplayer1
wait did you finish that cursor shill tower defense game ? w fuckign lua btw
@nicholas6497
What I took away from this video is I will now call "bidirectional "skibidirectional"
@DePhoegonIsle
Sorry, that's a curser issue. Other IDEs will actively point out code that is either 'unreachable' or conditions that will always be true/false, given their location. IntelliJ does this, and I have to admit has caught several times where I'd had invalid/unnessary checks, that got there by accidental re-organizing and the logic wasn't entirely thought through. It is a partial LLM issue, sure.. but the IDE should know to spot the invalid/impossible/redundant/useless code/statements. Seriously. THe problem is that people blindly trust the code without thinking, and if your IDE isn't flagging useless states, impossible states, or flawed states..... it's a bad IDE.
@Ratstail91
I'm seriously considering switching to vim full time maybe neovim never looked into it. I'm so sick of AI.
@GorgyCL
Prime working for that Cursor money hard.
@bause6182
I'm waiting for the day when there will be a big IT disaster because a company in charge of an important project will use a vibe-coded AI library as a third-party dependency
@KelvinShadewing