Mastodawn

Something I love about engineering/physics/CS culture is that almost every intellectual resource is available for free. There's a real presumption of non-gatekeeping that is, in my experience, not present in other fields.

Adriel Tan Feb 12

Anders Eknert Feb 12

AI agent "contributes" PR to matplotlib.
PR gets rejected.
AI agent *writes and publishes blog to shame the maintainer*.

What a time to be alive.

https://github.com/matplotlib/matplotlib/pull/31132

Adriel Tan Jan 7

Jan 7

Periodic reminder to boost the posts you like to keep the Fediverse alive.

WE are the algorithm here

#Mastodon

Adriel Tan Dec 2

ploum Dec 2

Unix - BSD developers : this code should run and be maintained as long as computers exist.

Linux - GNU dev : this code should run and be maintained as long as someone is willing to do the work

Consultant: this code should run until the end of my contract.

Enterprise grade B2B software : this code should run till the end of the demonstration to that big customer.

Vibe coding : You are perfectly right. The code doesn’t run. Try this one. You are right, it deleted your hard disk, I’m sorry.

Adriel Tan Nov 28

mcc Nov 28

If posix is so great why didn't they make a poseven

▓▓▓▓▓▓▓▓▓▓▓▓▓▓░ 90%

Also it’s really funny when you correct it… Gemini goes OH I MADE A DISASTROUS MISTAKE LOL.

Show thread

Adriel Tan Nov 19

If you remove the rate limit, it is way better than auto mode in cursor. But for a SOTA model, I sure hope so lol.

Show thread

Adriel Tan Nov 19

Overall, I'm happy with G3 performance… it’s thinking/debugging “skill” is better than cursor auto mode (which tbf known to be bad) and it’s been useful for writing integration tests, refactoring, small functions. But it really needs a keen eye on it to stop it doing stupid things.

I am also looking forward to testing browser mode for non DE related tasks… seems cool.

But the current rate limit makes it unusable for serious work

Back to cursor, will give it a shot with G3 next.

Show thread

Adriel Tan Nov 19

Cons:

- The rate limit is useless for anything serious. I could only get a plan and a single task on that done on G3 High without hitting limits. I got 60% of tasks completed with G3 Low. And I can’t pay for more.
- G3 likes to leave comments on it’s thinking in code without cleaning it up
- G3 deleted all my if verbose: printSchema(), show() statements…. For no reason.
- G3 changed my arg parse flags for some reason. I don’t know why.
- It’s buggy af (but not sure if it’s day 1 issues)