Recent searches
Search options
Yesterday I encountered a "wrong-on-the-internet" rando professing his excitement for "using machine learning" in #3dprinting to throttle speeds in the right places to avoid quality loss.
While completely not worth engaging with, I feel like this is a useful example to understand why this idiocy is so infuriating...
@dalias Can you tell us more about why machine learning isn’t a good fit for this? Though I didn’t even know throttling speeds could possibly improve quality. Knowing little it sounds possible? But what tipped you off that it would not work?
@futurebird It's not that it couldn't work, but that we understand the physical reasons when/why going too fast can give poor results, and can just apply that rather than trying to coax a model into figuring out the same things without underlying models of thermal transfer, etc. It's really low hanging fruit.
@dalias @futurebird You mean instead of just counting a zillion things and noticing a pattern, we should understand how things work and know ahead of time how things will go?
@geonz @futurebird There's actually a recent related example of great success with this in #3dprinting: MPC (model predictive control) of the heater, vs traditional PID loops. PID is a very general purpose control mechanism that can track a target, but actually modelling the thermal system (energy going into melt, energy lost to ambient air passively & via fans, energy input from heater) gives vastly better results that can respond to rapid changes in demand for melt in ways PID never could.
@dalias @futurebird MATHS for the win :)
@dalias @geonz @futurebird But thinking is so hard! /s
@dalias @futurebird so many people chase “machine learning” when what they have is a basic control systems problem
@dalias @futurebird and the reason “machine learning” seems appealing is that it offers the appearance of being able to solve the problem without figuring out what control systems problem they have or even that they have a control systems problem, I suppose
@kevinriggle @dalias @futurebird interpolation functions are for when you can't solve the problem analytically.
LLMs are very big interpolation functions.
@kevinriggle @dalias @futurebird
"Fuzzy Logic has entered the chat."
@RealGene @kevinriggle @futurebird Don't you love how the same bad ideas repeat generationally?
Sometimes I think the best inoculation against this stuff would be flooding the space with clips of the previous generation's version of the same hype, so that folks recognize it's been done before and looks laughably bad in hindsight.
@dalias @RealGene @futurebird people say that these kids (broadly construed) have read too much science fiction, but, having read too much science fiction, and having talked to lots of these kids, the pattern that’s emerging is that they haven’t read _enough_
@dalias @RealGene @futurebird (“well I read some Asimov, he’s not very realistic” darling that’s like saying you listened to “Lucy In The Sky With Diamonds” and it didn’t make much sense)
@RealGene @dalias @futurebird my rice cooker!
@kevinriggle @futurebird Yep. And often one with a plethora of existing expert literature on it.
@dalias @futurebird “getting engineers to read the literature” _has_ always been the actual hardest problem in computer science
@dalias @kevinriggle @futurebird
How easy is it for ordinary people to know about, access and apply existing expertise?
GPT models are a “click” away.
@dahukanna @dalias @futurebird We all _have_ expertise. Some of which is in saying “I don’t know and I need to consult an expert”. By offering a putative universal expert which is only a click away… it’s the early days of “Dr. Google” again before we got good at understanding search results
@dahukanna @dalias @futurebird except with ~all of the relevant authority and expertise signals thrown out and the information ground up and regurgitated as a pink knowledge product slurry
@dahukanna @kevinriggle @futurebird Well that's what a search engine would do, if they hadn't dismantled it to repurpose as a machine for scamming ppl...
.
@dalias @dahukanna imagine controlling access to knowledge at such scale and being in a position to economize it (selling premium levels of 'IQ" for AI models) - rather than primarily needing to use adds and behavior tracking on the back of granting access to that knowledge "for free".
@Virginicus @dalias @futurebird now you have two problems