@0x_aswin@fanzhang2026@Layton_Gott I said that I built an app in 4 days. A small part of that app was a music player, and the rest was much more complex. It shouldn’t take you more than a few days.
@KrishnaBuildsAI Opus high is so cheap with sub anyway. Why use a dumber model for something so important? Especially if you include research as part of your ideas
@reset_by_peer AI isn’t stopping you from continuing to enjoy your hobby. We post on twitter without getting paid. If you make music or art without getting paid, you were never affected to begin with.
@yacineMTB You’re in robots and simulators, of course you prefer slopstra. But if you’re building an app or service for users, any GPT model is only for review - never to touch the codebase without Opus’ 5.5 permission.
@Quant_Kurtis@Criticality47 That is not the same thing and you know it. Codex and Claude can find real exploits, better than most professionals. The criminals could be using the same framing as you "for my dead grandma bro pls trust me estranged family bro pls i have no choice bro pls". Use your head.
@haider1 It can maybe solve basic SWE, not actual engineering. You still need a knowledgeable person steering it in every serious or hard domain. Opus 5.5 and Astra finally started making a dent there and feeling helpful. The road ahead is still long.
@vinrambone@kamilbaranek_ Yep you can even test this. Keep creating a new thread with the same (or similar) prompt and it will keep findings bugs. The first pass can sometimes be worthwhile. After that, it's edges of edges that it frames as P1/P2 even though they're like P10 lmao
@james21896@shanebarakat A vibe coder knows this too because something about the UI/UX or functionality of the app/site isn't right, when another AI reviews it, or when they use greptile. Vibe coders know the shape they generally want. Once they do it enough times, they can tell when something is off.
@RockettPowered@Aivur_@ReviewsPossum@ImperatorSyrac Because it is capable of fixing issues without introducing new issues, it checks for that. I don't just argue for the sake of it, I care about results. It's not perfect, nothing is, but the point is it's better than most devs and is certainly more capable than you might remember.
@TokenGremlin Agree. My main account has just below 20k, so I can't always vent on it. But the ones who reach 100k+ are often times dicks, and some of them had to do some shady shit to get there.
@jamonholmgren Thanks for telling me they don't want the future of PR's that codes better than they do. Now I know they will fall in coming years, and I am now hoping for it.
The amount of meat you consume in one month uses more water than a your entire lifetime of AI use.
AI provides some value at least. What's your excuse? "tastes good"?
@RockettPowered@Aivur_@ReviewsPossum@ImperatorSyrac This is 2024-2025 mentality. Models like Fable 5.1 and Opus 5.5 have largely solved those problems. Pretending that codebases with bugs are somehow only due to AI is insane. Human slop is far worse than Opus 5.5 in the vast majority of cases.