The problem isn’t LLMs

The problem is always people. In this case, the loudest llm shills happen to be the same bitcoin/nft shills. Ive found its a mildly useful tool slightly better than google thats overly expensive.

the foundation of a working gen ai is stealing

WTF?

I have a problem with people turning their brains off while scrolling on their smartphone or watching TV as well. Tunring your brain off just kind of a bad thing in general.

Right now, other than goofing with images, the only thing I use AI for it as a research and learning tool. I have far too many thing I need to understand before attempting to have it do any code for me.

i take time off of X/Twitter every now and then to keep my sanity. I applaud you for keeping it together and not sending mixed signals to your followers by switching sides like many and doing vibe coding sessions while dissing on ai. you sir, I strive to follow

Could it be the case that juniors are conflicted the most from this technology? Prompting at this point is producing output better than they ever could on their own, and at the same time they don't want to be meat proxies. How should they deal with this dilemma?

Is this from a dev stream or do you have a full video?

from where you bought that green cursor, I want one plz 👀

What do you think a good application of LLMs would be?

Low stakes work where correctness is provable, the margin of error is extremely high, and the cost of failure is very low

@pegvvin Do you think an expert like SebAaltonen using AI for his work falls into one of these categories? IIRC, he has claimed a greater than 2x boost in productivity.

He seems to post a lot about correcting horrible decisions "made" by these systems. Anyways, I do not disagree that you can sometimes see 2x productivity increase with a spot measurement, e.g. when throwing together a prototype where the details are easily predictable (this fits the scenario - low stakes, high margin of error, low cost of failure). Again I think the most productivity increases (for the same quality of work) are in scenarios where you can (a) prove correctness, (b) have a high margin of error (the bar is low), or (c) the cost of failure is very low. Certain styles of game prototyping can fit into that (though certainly not *all* styles of game prototyping). The question is, what is the multiplier when actually shipping highly valuable things? So not *spot* measurements. Have we seen incredible games or other software being made in half the time? No, not really. We have seen a lot of prototyping. The most rapid of such prototypes are, literally without exception, gigantic disasters technically, appear extremely janky, and are deep within the uncanny valley artistically. The more sensible usage of the tools creates better results, but are still much slower than the often-claimed 2x, 4x, 8x, 10x, 50x numbers. Is it faster or slower to take those prototypes to high quality, if they were very rapidly generated? And then, furthermore, is that increase in speed (if any) worth the costs in all cases? e.g. degradation of human familiarity with the solution, the cost of using the systems, the degradation in human fulfillment / happiness, any atrophy in skill over time.