Post B8JC8i2zHDTAiqACtE by ariadne@social.treehouse.systems
 (DIR) More posts by ariadne@social.treehouse.systems
 (DIR) Post #B8JAKxttxZhnSQd7se by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       also if you legitimately believe that LLMs and more specifically, these very large monolithic models created by the AI labs will lead to actual cognition, I'm sorry to tell you that we do not yet have an actual definition of human cognition, and, ultimately, prediction is only a small part of what cognition is.
       
 (DIR) Post #B8JAegG9G7akXIR9DE by cadey@pony.social
       0 likes, 0 repeats
       
       @ariadne One fun thing I did at an AI meetup where everyone was drunk was made an impassioned argument that nginx is a sentient entity. I won the argument because it made people drink heavily.
       
 (DIR) Post #B8JBAwsF1SQZKbdoG0 by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       in cognitive science, prediction accounts for an agent's ability to interpret sensory input and make quick decisions in the moment.  a cat or dog is able to predict that certain inputs lead to certain outcomes.  that part is the *boring* part.  "AI" has managed to duplicate this in various ways since the 1960s.and, importantly, that's *not* human-level cognition.  we humans have a prefrontal neocortex that allows us to reflect and refine our planning.  in other words, meta-cognition.there is *zero* evidence that LLMs, agents, etc. are architecturally capable of meta-cognition.  if your goal is so-called "AGI", all of this is a fucking dead end.
       
 (DIR) Post #B8JBZcnAp5TOHx2Mm8 by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       it is the same with vibe-coding.  "writing" code and "generating" code are two very different things."writing" code means following the principles of software engineering to plan out the design and implementation of software."generating" code means generating code that is based on some other design input.the current monoliths can generate code, but as bunix demonstrates, they can't generate good design.
       
 (DIR) Post #B8JBwgk6IkhDCWb6WG by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       anyway, the "AI" labs are staking their large valuations on two major areas: software engineering & computational bioinformatics, that these processes can be entirely handled by agentic workflows.for software engineering, i rate the agentic thing a solid ~maybe.  but it is going to be expensive with current architectures.for bioinformatics, i don't know enough about it that i feel confident saying either way.
       
 (DIR) Post #B8JC8i2zHDTAiqACtE by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       a larger question is: if software engineering becomes entirely agentic, and generated code cannot be copyrighted, then what does this mean for copyright in general?i'm not a lawyer, but it seems like the entire tech industry kinda lives based on the concept that copyright is still a thing
       
 (DIR) Post #B8JCFLoiKIr5nsecqG by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @dalias correct, which is why i compared token prediction to the cognitive processes of lower animals
       
 (DIR) Post #B8JCUOgTwrdv2Wyj2m by xyhhx@social.treehouse.systems
       0 likes, 0 repeats
       
       @ariadne wait theyre making ai dogs now too???
       
 (DIR) Post #B8JCXQLkDElF4eiooy by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @xyhhx what?
       
 (DIR) Post #B8JCdzMWwuDpOxDh3I by xyhhx@social.treehouse.systems
       0 likes, 0 repeats
       
       @ariadne ai labs... labradors... ai labra.. im sorry
       
 (DIR) Post #B8JFRHxCoZSJ9z0grI by bloognoo@retro.pizza
       0 likes, 0 repeats
       
       @ariadne The company i work for has kinda shrugged and declared if they work with AI they don't own their own code any more, and so i think we're relying on our industry being niche, the legacy code a systemic nightmare only the biggest ai models can touch at all and brand recognitiion.I used to believe ideas were worthless and implementation was key. Inow believe implementation is worthless and all we have left is systemic knowledge of arcane systems, and that will fall. Yay.
       
 (DIR) Post #B8JKANMfwZiemzg6pE by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @bms48 @Elizafox look i am not presently stoned enough for this subthread, can y'all tag me out
       
 (DIR) Post #B8JMNCRc3Z7QaqEKky by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       @yakmacker sure, i think i mostly agree that LLMs predict language in a similar way to other language predictors, including humans.  but that does not equate to cognition.