> this is a RL problem where you have to balance the chance of an infinite loop (it keeps thinking there's a little bit more to do when there is not) versus the opposite where it stops short of actual completion.
Any idea on why the other end of the spectrum is this way -- thinking that it always has something to do?
I can think of a pet theory on it stopping early -- that positive tool responses and such bias it towards thinking it's complete (could be extremely wrong)
My pet theory: LLM's are good at detecting and continuing patterns. Repeating the same thing is a rather simple pattern, and there's no obvious place to stop if an LLM falls into that pattern unintentionally. At least to an unsophisticated LLM, the most likely completion is to continue the pattern.
So infinite loops are more of a default, and the question is how to avoid them. Picking randomly (non-zero temperature) helps prevent repetition sometimes. Other higher-level patterns probably prevent this from happening most of the time in more sophisticated LLM's.
> Any idea on why the other end of the spectrum is this way -- thinking that it always has something to do?
Who said anything about "thinking"? Smaller models were notorious for getting stuck repeating a single word over and over, or just "eeeeeee" forever. Larger models only change probabilities, not the fundamental nature of the machine.
Seriously, the product focuses on personal and independent content, primarily published on websites, but the app only works in iOS. That seems like a fundamental mismatch.
Beautiful, I wonder what kind of craziness would be possible with this, at scale. Whole buildings being printed and assembled block by block. Real life Minecraft, if you will
Blame! is a manga where in the future humans have robots that build, and are controlled by people with Net Terminal Genes. Something happens and those humans die leaving the robots building non-stop procedurally for eons. By the time our protagonist moves about in the world, its said the Megastructure reaches from Earth all the way to Jupiter.
There are serious efforts and working prototypes of printing houses. This works surprisingly well, allows construction in days instead of months, and shows a lot of promise. It’s a great rabbit hole to fall into!
> From my experience - don't even go looking for "most optimal ways" in this journey. The simpler, the more unstructured - the better. Otherwise maintaining the whole setup will take more than give.
> Did you come up with this yourself, or did you copy it from somewhere else? Because I would like to know more. This sounds like a very good approach.
I came up with the thought, but the exact term Value Plan, I've read somewhere, Andy Grove (maybe, unsure, I'll try to find a source).
Shaping something for the whole body would be difficult, because most puncture resistant materials aren’t very stretchy. You’d have to have the perfect shape for a specific person. It probably is possible, but likely to take a lot of work and be rather uncomfortable.
Doing some quick googling, I find sharkstop.co, which makes wetsuits to protect against shark attacks. They say the “placement of Shark Stop panels … have been strategically chosen to significantly reduce the amount of blood lost during a shark bite incident…” It sounds like they’re carefully choosing the location to make trade offs between protection and wearability, which I suspect is the right approach. We do the same thing with our booties, and only have stingray resistant material covering some percentage of the bootie.
Any idea on why the other end of the spectrum is this way -- thinking that it always has something to do?
I can think of a pet theory on it stopping early -- that positive tool responses and such bias it towards thinking it's complete (could be extremely wrong)