30
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 06 Jul 2026
30 points (96.9% liked)
chat
8632 readers
204 users here now
Chat is a text only community for casual conversation, please keep shitposting to the absolute minimum. This is intended to be a separate space from c/chapotraphouse or the daily megathread. Chat does this by being a long-form community where topics will remain from day to day unlike the megathread, and it is distinct from c/chapotraphouse in that we ask you to engage in this community in a genuine way. Please keep shitposting, bits, and irony to a minimum.
As with all communities posts need to abide by the code of conduct, additionally moderators will remove any posts or comments deemed to be inappropriate.
Thank you and happy chatting!
founded 5 years ago
MODERATORS
I feel like there's an element of self sabotage in the "ai" space that has resulted from the amount of money poured into it. All of the major advancements have come from throwing more tokens and more compute at every single problem, but with every engineer working on the technology thinking about the problem in terms of "how do I solve this with the power of infinite money", there's nobody actively working on the alternative question of "how do I make what I already have work more efficiently". Or at least there are very few people working on the problem from that angle.
I work in tech and there are definitely a lot of engineers focused on efficient AI usage. Almost all of the best engineers I know are obsessed with efficient usage actually. Context lengths are limited and smaller contexts reduce hallucinations, so more efficient model usage also produces better outputs. The main challenge is getting high quality 'lean' context to describe your problem fully enough to the model, and there are many promising attempts at this for code problems which will eventually be reflected in other industries with huge knowledge bases that LLMs struggle with.
There is also tons of work on more efficient and smaller models. There are tradeoffs with smaller models in that they hallucinate more, can't generalise as effectively, and perform much worse. There is a technical limit on what they can model based on their size, though active research is still ongoing.
Reaching a certain context size for sure sets off red flags that cost and hallucinations are both about to multiply as the models start looping nonsensical tool calls which degrade with each iteration.
I understand why people get obsessed with trying to “one-shot” problems, right now I think it’s the developer equivalent of AI Psychosis, playing the lottery to see if you get lucky or receive a half-finished rewrite.