Like the rest of us, @gigi is feeling the yuck of slop.
No LLM slop ever did anything close to this to me, nor will it any time soon. No matter how “refined” it supposedly is. The late-stage industrial slop that is served to us as “content” right now is an insult to our human nature; an insult to free expression; to Truth, and Beauty, and Goodness. To being itself.
Gigi bought a new keyboard. A mechanical one that works with his phone. As someone who ends up typing on mobile more and more frequently, I'm wondering if I shouldn't do the same. However, Gigi seems to be looking for quiet and calm to do his typing and usually if I have quiet and calm, my laptop isn't far away, so maybe I just think he picked out is cool.
I find Gigi's take particularly interesting because I believe that he has really embraced AI tooling and vibe-coding, yet he, like all of us, finds slop abhorrent. His answer is this:
Less prompting, more typing.
But I hope he writes more on the subject. I'm not satisfied with just saying we all need to try harder. Perhaps it has always been like this: there are people who are willing to try hard to create something, people who are willing to struggle -- and then there are people who just want to be people who have created something, and aren't particularly concerned with the creation. Slop just means more people in the latter group can reach their goal.
I'm fascinated by this idea that slop is somehow connected to a lack of effort or a desire to get something for nothing. I haven't found the right way to phrase it yet, but I suspect the real problem with slop is here.
Gigi does a nice job of pointing out some examples of writing that is not slop. Of people who did labor over their thoughts. But I don't know if it's enough.
My friend asked my help yesterday about apricot farming from seed. I was in a train for nearly 5h yesterday evening so I decided that despite not having stable internet connectivity (cross-border travel is meh for mobile internet) I could still help.
Now... there's tons of slop here. But my friend liked the process doc - it's just a 2-pager - and I really liked the quality of the audit reports. Because of prompt queuing all this happened async and it simply goes as fast as I can review it, except for the comparison, that took tons of time because it was going over everything - which I like also. I wrote about 2-3 pages in prompt. The result is about a 100 pages of "slop" documentation, including process reports and validation results. But it is exactly what I asked for.
Is this lack of effort? Or efficient use of time to get something going? Automating away a week or 2 of research? I'm not publishing anything - I don't like publishing things that everyone can do themselves - but I'm pretty confident that I can get higher quality results than telling my friend to just ask GPT. Whether or not this was electricity well spent will be known if and when those trees start producing food.
@remindme in 8 years
That is impressive.
Sounds to me like, with some of your help, the bot can figure out how to start an orchard. It just can't actually start one. It can't go and feel the soil, or prune the branches.
I think the analogy to writing would be: it has the formula to write, it has the knowledge to write, but does it actually write content?
My stance is that it probably doesn't, since writing, to me, is the alchemy of coherently transmuting one's experience into words.
Whether or not the microprocessor is experiencing the world is another thing entirely...
Right so nothing here is original. It is all derived information. The original data from this work will form over the next 8-10 years (hopefully) while my friend grows those trees from seeds.
The idea I keep having is that we have the tools to kickstart anything. To research anything. In more depth and breadth than we'd allow our own time to be spent on. And to use AI heavily in the design phase but much less so in the execution. The energy cost of letting some AI monitor plants 24/7 will be insane anyway. Not worth it, trees don't need AI to grow, they need fertile ground, water and sunlight. In this case, pruning
The anti-slop side resonates with me. And he had me up until he typed "walled gardens."
I don't think we should hate "walled gardens" or other centralized systems, because they make sense. They found efficiencies. I think we should hate people's dependence on them. The phrase is often used as a buzzword to put the blame on the system/network/platform that is walled in, but doesn't really account for the fsct that they are all opt-in. No one ever forced me to use them, and the fact that I did, I don't think was coercion, but a strong tribal-primitive type of suggestion.
I am not saying I don't think the operators/proprietors of these systems are not malicious. But can we not be free by refusing to use them?
Maybe this is what Gigi means. I just don't like calling them "walled gardens" because actual gardens are good places where I think you should spend time, walls or not.
No, not raw. Edited; well thought-out -- McCloskey is really making me re-live the beauty of well-combined words, clearly non-slopped...
#1534753
Bro is FUNNY! <3
Good memorable days
When using LLMs for coding, I only get results that I like and trust when I slow down to my level of comprehension. If I don't, it becomes this blob of text that I have no context on. I do not feel proud of it. I cannot trust it.
I'm doing a project with LLMs now. It's been going for a month. I barely made any meaningful progress. I'm just trying new things, prompting, waiting for things to finish. But no real output.
Yes, I explored some things, but I maybe understand 10% of it because I have no time to review, I need to keep the velocity up. I need to keep prompting.
Sometimes I wonder if I should have just written it all by myself, slowly, mostly using the LLM as a search engine. I think I would have ended up in a better place.
I feel like we are all being lied to by AI companies. They dumb us down. We are too lazy to do hard things now because it feels not modern to do your own work.
But the thing is - with LLMs, you are getting the same results as everyone else. It's the worst. A complete monoculture.
I will keep using them because I am still worried that I'm just not using them right. Maybe there's a chance. Maybe I should just prompt better. Just one more prompt. Then one more. The cycle continues.
I feel like a sucker. I hate it.