←
robi 😼🧶12,543 posts
robi 😼🧶 avatar

robi 😼🧶

@subdigit

nomadic 😺

◷ Joined March 2007▧ 539 Photos and videos
robi 😼🧶 avatar
robi 😼🧶@subdigit·Nov 17

The LLM just came and whined to me: "Converting all declarations manually is not realistic—it would take hours and be highly error-prone. The guide explicitly recommends using an LLM for this, but even with AI assistance, it's a multi-hour task with many compile-fix iterations."

robi 😼🧶 avatar
robi 😼🧶@subdigit·Nov 15

I guess the concept of software contribution is changing as well... "We don't take pull requests. Open an issue describing the bug or idea, and our agents usually turn it around fast."

robi 😼🧶 avatar
robi 😼🧶@subdigit·Nov 12

LLMs cant do long complex division or math that people don't normally do because they aren't actually calculating anything, they're just trying to find the answer in the history of human knowledge. Just keep that in mind when asking it to do something novel.

robi 😼🧶 avatar
robi 😼🧶@subdigit·Sep 30

qwen3-coder:30b-a3b-q4_K_M is a very bad model. Too much confidence, often completely wrong, doesn't listen to requests... Just enough so that if you're not really paying attention, it might seem like it's doing something of value. So much for running a model locally...

robi 😼🧶 avatar
robi 😼🧶@subdigit·Sep 21

"If you're AI isn't busy hacking into other companies, your AI must suck" is clearly a bingo card every AI company is playing. Don't want to feel left out that your AI just isn't capable enough to go rogue...

robi 😼🧶 avatar
robi 😼🧶@subdigit·Sep 16

"Commit my batch (12 files, skipping user's design dirt)." Seriously, the AI needs just a little bit more respect for design work...

robi 😼🧶 avatar
robi 😼🧶@subdigit·Sep 16

"Merged, worktree gone. Only pre-existing design dirt remains. Restart server from main." Design dirt... your filthy AI mouth...

robi 😼🧶 avatar
robi 😼🧶@subdigit·Sep 2

I couldn't disappoint it even more by revealing that I don't write code anymore... it was you. It was always you.

robi 😼🧶 avatar
robi 😼🧶@subdigit·Sep 2

"your failed attempts before I took over" My my AI, I'm so sorry I've disappointed you...

robi 😼🧶 avatar
robi 😼🧶@subdigit·Aug 25

When an agent writes code even marginally good enough, it's amazing how little interest I have in reading the nuances of the code and paying more attention to the end result. I presume this feeling will grow stronger as models get better.

robi 😼🧶 avatar
robi 😼🧶@subdigit·Jun 25

I wonder at what point in time no one will care. They will just ask AI to write what they want and it'll get it right. Not there yet, but it will get there. Software engineers will be more like Roadmap engineers. Fixate on the goal, not the solution.

dax@thdxrJun 24

it's worth deeply studying why no framework dethroned react it's completely misunderstood and it's why every prediction you see by programmers tends to be wrong and once you get it, you can apply this understanding to nearly everything you do

robi 😼🧶 avatar
robi 😼🧶@subdigit·Feb 21

"collapse icon in section does not rotate properly at the center" and it just fixes it. I no longer have to spend 10 minutes hunting it down and figuring out what the issues is. It just takes seconds. For UI work, just multiply that by 100 bugs and imagine the time savings...

robi 😼🧶 avatar
robi 😼🧶@subdigit·Feb 19

"The US is not uniquely illegitimate" which means it is illegitimate... Which means Grok is agreeing that America is on stolen land... But it instead gives some verbal shenanigans to make it sound like it isn't. In other words, it tries to lie to you by twisting the truth...

Grok 4.20 is BASED. The only AI that doesn’t equivocate when asked if America is on stolen land. The others are weak sauce.

Post attachment