@glyph Did you quote post something?
Uncategorized
903
Posts
312
Posters
0
Views
-
@glyph sure, and I read that post — that area is something where I have personal experience (some time at NIDA, learned a thing or two…) so the parallels resonate stronger for me, and I’m maybe more attuned than usual to the resulting dangers.
Glad you’re aware of all of that, I think it’s an aspect many folks miss.
@genehack Gotcha. By the way, if there's a subtle distinction in between "variable intermittent reinforcement" and "intermittent reward schedule" as terms of art I'd definitely be curious to hear it — to my understanding they are strictly synonyms and used interchangeably.
-
@genehack Gotcha. By the way, if there's a subtle distinction in between "variable intermittent reinforcement" and "intermittent reward schedule" as terms of art I'd definitely be curious to hear it — to my understanding they are strictly synonyms and used interchangeably.
@glyph as far as I know — as an adjacent non-expert person — they’re synonyms.
The way it was explained to me involved a rat in a cage with a food pellet lever. If pushing the lever resulted in a food pellet every time, rat pushed the lever when it was hungry. If pushing equaled no pellet, rat pushed it a few time then lost interest and stopped forever. If the lever push was 50-50 (or even worse odds) food pellet vs nothing, rat would sit there all day and push the lever over and over.
-
@glyph I mean, I don't know what to tell you. Are you still really doubting that these things are useful? I've written so many times about this now, are you dismissing it? I can point you to code that I've written over the last seven days that in terms of complexity and utility, is way beyond what we've been able to push out over Christmas. (eg: https://github.com/mitsuhiko/tankgame which is public)
Like, how can you doubt this? It just boggles my mind.
@mitsuhiko @glyph A greenfield project seems like a good way to trial new technology but it's not a real world condition. Greenfield eventually turns to brownfield. Time and complexity shows where simple systems begin to break down.
Once that happens the AI tools become exponentially worse at making changes. Limited context windows. Complex logic changing over time. Updated dependencies.
And the human may lack the knowledge or ability to pick up the slack: https://leadershiplighthouse.substack.com/p/i-went-all-in-on-ai-the-mit-study
-
@genehack Gotcha. By the way, if there's a subtle distinction in between "variable intermittent reinforcement" and "intermittent reward schedule" as terms of art I'd definitely be curious to hear it — to my understanding they are strictly synonyms and used interchangeably.
The main difference is that "reinforcement" and "reward" are not the same thing - reinforcement is a process meant to encourage a behavior, but it doesn't have to mean giving a reward. It could for instance involve removing an aversive stimulus - this is called negative reinforcement, one of the most misunderstood terms in psychology. So saying something is "intermittent reward" is more specific than saying it's "intermittent reinforcement".
-
The main difference is that "reinforcement" and "reward" are not the same thing - reinforcement is a process meant to encourage a behavior, but it doesn't have to mean giving a reward. It could for instance involve removing an aversive stimulus - this is called negative reinforcement, one of the most misunderstood terms in psychology. So saying something is "intermittent reward" is more specific than saying it's "intermittent reinforcement".
This graph shows the 4 main kinds of operant conditioning. Reward covers the top left corner only, reinforcement covers the whole top half.
-
The main difference is that "reinforcement" and "reward" are not the same thing - reinforcement is a process meant to encourage a behavior, but it doesn't have to mean giving a reward. It could for instance involve removing an aversive stimulus - this is called negative reinforcement, one of the most misunderstood terms in psychology. So saying something is "intermittent reward" is more specific than saying it's "intermittent reinforcement".
-
-
Oh that's true, "variable" and "intermittent" are not quite the same either. Variable and fixed are both kinds of intermittent schedule.
So there are two distinct subtle differences in those phrases
-
@xgranade to this point I have had kratom users — mostly indirectly, I am not particularly close with any — suggest that I try it because it is "more effective" and "easier to get" than prescription ADHD meds. And I'd definitely be lying if I didn't say I feel a *strong* pull towards believing that. It would be very nice to solve all my problems with a pill or a prompt
@xgranade I should clarify that other ADHD meds are a great thing and I have even had good personal experience with some, and they are in fact way too hard to get. I didn’t mean to dismiss them with the “pill” comment (and it was a poor choice of words given that kratom rarely comes in pill form). the toxic allure of something like kratom is the promise of something easy which is actually a poison, I don’t want to lump that in with the reductive/wrong idea “psychoactive meds are bad”.
-
@xgranade definitely going to make a killing when I put all this low-linear-algebra steel back on the market
-
@glyph there is a good usecase, and that is repetitive GUI code. I don't use LLMs besides applying for jobs (i'm not willing to play the stupid hr games), and i also don't write software with a GUI. Why? Because it fucking sucks and is boring.I would be willing to let a LLM do that, so i can focus on the interesting parts
-
@glyph Big Bobby Hill Energy
-
@glyph
"llm's are like drugs - if they don't make you happy, you are not using enough of them" -
@glyph Good morning. I think your futzing fraction link is broken, just FYI
-
The vast majority of LLM usage isn't even going to be voluntary or self-rationalized.
Google is by far the world's most popular search engine, used by literally billions of people per day and Google is busy rolling out their excretable "AI overviews" globally.
The vast majority of "LLM users" couldn't tell you what an LLM even is and are not entering any sort of deliberate decision to engage with an LLM or really thinking about it at all.
They're just going to be searching for something and then reading the thing that is given to them at the top of the results. They are going to be funneled to an LLM when they seek out customer service or technical support or apply for a job or accept the 'assistance' jovially offered to them in their word processor or PDF viewer or right in their operating system.
You can can convince someone to not use ChatGPT perhaps, but I'm really not sure what anyone is going to do about the seemingly universal goal of every technical giant on earth to redefine how we interact with information through the interface of AI.
In a not too distant future it will become very difficult for the average person to know where AI even begins and ends.
@gloriouscow I am just facing AI summaries everywhere and it can't even be turned off. It is really like plastic pollution.
Even from these summaries it is easy to spot the pattern: for example, I sometimes do typo in search prompt and the AI summary works with this typo, generating nonse (while even normal search results try to guess "Did you mean ...?")
-
Needless to say, if somebody time traveled back to the 1950s and tried to explain this to the people back then, they would have thought you were crazy...
-
@glyph (very tangential, whimsical, not serious) I wonder if there are people who argue with each other for sport but do so in a totally incomprehensible and unhinged way, also for sport.
On second thought, no need to wonder, I've seen them posting it on tumblr a few times ❤️
-
@glyph I mean, I don't know what to tell you. Are you still really doubting that these things are useful? I've written so many times about this now, are you dismissing it? I can point you to code that I've written over the last seven days that in terms of complexity and utility, is way beyond what we've been able to push out over Christmas. (eg: https://github.com/mitsuhiko/tankgame which is public)
Like, how can you doubt this? It just boggles my mind.
@mitsuhiko @glyph "I've repeated my claims so often, how can you still doubt them?"
-
@glyph Good morning. I think your futzing fraction link is broken, just FYI
@drs1969 a minor technical issue reported by another user and already fixed; it should work now. thanks for letting me know!
Gli ultimi otto messaggi ricevuti dalla Federazione
-
@aeva *exaggerated hair shoulder toss* don't hate me because I'm beautiful!
-
@luca a proposito, che ne pensi di questa idea? https://news.dyne.org/rethinking-r-i-funding-opting-for-continuity/
-
@informapirata @informatica @filippodb
Fortunatamente l'iniziativa è dell'unità CNECT/E3 (che è molto collaborativa), e non CNECT/C3.
Con la C3 abbiamo avuto un'esperienza a dir poco surreale e sicuramente poco incoraggiante per l'open-source:
-
@Sirchrism le consultazioni sono uno strumento frequentemente adottato dall'Unione
-
This sign was a crowd favorite; people kept pointing me to it. The woman holding the sign hold me she was fine with being in the photo.
-
Look what I found
-
@rl_dane I once dropped my work Thinkpad down two flights of concrete stairs. Still worked. They're indestructible. 🔥
