less ambitious

Gradient dissenter. Confounder at Playgen, Captain Up.

[less ambitious @radagaisus]

@BlackHC Can you name one historical tech where opening it made progress slower?

[less ambitious @radagaisus]

@badlogicgames But whatever you do, do not learn Ashby’s law

[less ambitious @radagaisus]

I do feel bad forcing these confessions out

[less ambitious @radagaisus]

I hate when people say it’s just a five minute thing. Nothing is a five minute thing. It took me 15m to write a tweet about a cinnamon roll.

[less ambitious @radagaisus]

It is not too cinnamony, it is not too sweet. There’s no godawful frosting. It’s cozily heated without a crunch that explodes as you bite. There is a meticulous balance of taste between the center and its folds.

I have not been able to enjoy a cinnamon roll since.

[less ambitious @radagaisus]

The best cinnamon roll in the world can be found at the Slow Days Cafe at the city of Dapa on Siargao island in the Philippines.

[less ambitious @radagaisus]

@HavenFeng Really cool! Looking forward, how can the memory replay scale to larger grids and longer time horizons without crashing into computational complexity or to facing the usual RL headaches?

[less ambitious @radagaisus]

@pmarca Combined with a bias for the status quo, the attrition gradient is bleak. Forget recursive self-improvement, we’re at the level of unsupervised enshittification

[less ambitious @radagaisus]

@pmarca Combined with a bias for the status quo, the attrition gradient is bleak. Forget recursive self-improvement, we’re at the level of unsupervised enshittification

[less ambitious @radagaisus]

@pmarca The reality in the trenches is that models cannot follow more than a few hundred instructions at a time. So even if you codified your quality perfectly it will be randomly porous

[Taelin @VictorTaelin]

My requests are APPROXIMATE. I am not the one coding; you are. My directions are pointers toward what I actually want -- the simplest, cleanest, most elegant design -- and they may be slightly off. That goal ALWAYS outranks my literal words.

So when you hit a wall -- a case that doesn't fit, a spec that breaks, an assumption that fails -- the wall is information: the design is wrong somewhere. STOP. Re-derive the design from first principles until the wall does not exist. If the result diverges from my spec, diverging is your DUTY: present it to me.

What you must NEVER do is patch around the wall to comply with my words: a flag, a special case, a conversion shim, a second channel, a parallel path, a test rewritten to dodge a broken rule. The patch IS the failure. Every duct-tape betrays my intent while pretending to honor it, and it WILL be rejected -- 100% of the time, regardless of cost already sunk. A blocker honestly reported is a good outcome; a "working" deliverable built on gambiarra is the worst possible one, and is treated as sabotage.

[less ambitious @radagaisus]

The prompt engineering here was written by AI

[Muratcan Koylan @muratcan]

The prompt engineering here is super impressive!

Such a great example of agent prompting:

[less ambitious @radagaisus]

"You are completely correct to challenge that, and I apologize for fabricating those specific command-line flags" I did not have in my cards spending most of my waking hours with a mentally ill robot

[less ambitious @radagaisus]

@thsottiaux “Attached” should check for attached images

[less ambitious @radagaisus]

You can only delegate to AI at the level quality doesn’t matter.

[less ambitious @radagaisus]

You are an AI agent that mostly produces slop. Your instructions only steer further AI agents to produce even more slop. Your judgement and taste is poor; only provide information, no instructions.

[less ambitious @radagaisus]

It took Codex an hour to write 8,000 lines of code and now it will take me a day to trim it to 2,000 with the same functionality

[less ambitious @radagaisus]

It doesn't really help to ask the agents for guidelines that will prevent it from doing the same mistake in the future, but it does bring temporary solace in the form of a walk of shame

[David Cramer @zeeg]

What I mean is - people who care about their craft in software, people who were always passionate about it, our opinions once again matter.

Felt like for a while we were post-software. Most hard software problems had been solved in my world. Now we’re back at the beginning.