What is "soyfacing over LLMs"?

Ahhh, thanks

Slop, as-in, "Look at what this LLM can code!"?

i post a lot of things that claude has said, because i see something that i think others should see. maybe it's too much? but when i started, ai was drawing with melted crayons and writing javascript with syntax errors. just stay buoyant out there 🌊

yes, sorry to say, I stuck with you this long but I unfollowed you today. bon voyage my friend

nominate, best authentic nostr post of the day..

I write all my code by hand. Pen and paper 📝 Then my secretary types it on her typewriter, we file a copy and only then is it carefully keyed into the terminal.

Misogyny-driven development

when it comes to image generation, it's all slop, as far as i can tell. i have found gpt-5 can do reasonable simple line art semi-realistic images (my profile pic now is this) but none of it is art. i think you can start from what the LLM does and massage it into art but most of it comes out slop. as far as technical slop goes, a lot of the popular models are shit. gpt is shit at code. gemini is shit at code, though i hear the latest is nearly as good. the only LLM that produces nearly solid technical documentation is claude, but you have to constantly watch what it's doing because it often will suddenly decide to do something completely the wrong way, can even happen it does the same mistake twice within one chat context (before a compaction). i created a design for an emergency shelter that can float, and one of my follows fed it into gpt to create an image and almost every minor detail of the image was contradictory to the text. it barely even caught the gist that the frame was a classic football/soccer style geodesic dome. getting these things to do work properly takes a lot of time and a very stringent vigilance, and most importantly, sufficient technical knowledge of the field. even then, they still make lots of mistakes but they are mostly of a small list of characteristic errors that pops up here and there. like saying it did something, when it did not, that's the most common one i see, and then using a method or design that you specifically previously said to not use. they don't have working long term memory, barely even short term memory. but they still can do stuff. art is not one of the things they can do. i was already hating on them 2 years ago. i had colleagues in my employment, even a boss, who was using some anime style slop as his avatar. in the shitcoin world, the slop is everywhere.

They have put a lot of guardrails on them over the last 6 months. Grok has help me destroy two NucBoxes. But, I could not have written Redacted Science without the old ChatGpt. 🔥 👆 Science is Redacted!

I like “old school”

Im with you brother. It's getting insane.

slop gobblers

You joke but I’ve been considering a soothing amber VT220 or similar attached to some shitty computer just to use command line utility and text editors. No LSPs or agents or whatever they call it.

cool

I enjoy AI and use it for various things but I agree, too much slop. AI should be used for enhancing our lives and our abilities, not to outright replace them.