https://www.youtube.com/watch?v=-MUEXGaxFDA
Pretty ranty, makes a “anyone who disagrees is wrong” statement
Mainly a complaint about quality, but those seem transient/historical complaints, but interesting to read the reactions in the comments…
Biggest complaint is the same argument about any type of mechanical reproduction and has always been proven wrong. Levels of creativity can always be injected.
Also bases a big argument about the inability for LLMs to invent anything new (already proven wrong in the fields of math, biology)
Training exploits labor:
- This is a valid moral critique
- Solvable by paying market rates, could build a better solution that people could choose to use
- Most training data will be synthethic/user collected, we may be reaching some escape velocity soon.
Quality:
- This is pretty temporary, most of the critiques largely out of date
- Complaints about low memory, inability to track characters, items etc largely will be going away
Vaudeville AI Dungeon
Existing NPCs suck, so LLM-based ones seems like a huge improvement
Complaints about hardware requirements / cloud-based generation not being a thing (wrong already)
Training sources/data
- Copyright vs Fair Use
- Social tradeoff
- Comes to the right conclusion on input though
AI can’t be new, can’t be as good as Earthbound? That’s going to be a hill to die on.
Ignores AI as systems