If only these virtues could be represented by cartoon ponies..

@cnx @Daojoan

I mean to be fair. You probably build more stuff than I do. Considering you're not over engineering anything.

I've been calculating and recalculating the technology stack for my new website for about almost 3 weeks now.
Pivoting from monoliths and static sites to serverless edge computing, And Multi-Page multi-apps In a mono repo compiled by vite, With a progressively enhanced front end driving an MMORPG world World through a webgl app written in react, plus plain HTML

@cnx @Daojoan let me guess, htmx stan?

If so, I love that for you.

@Daojoan Gemini is the weak sauce LLM.

The bigger problem here is that they want to cut costs by firing engineers, so none of this can be integrated into anything holistically because there's no one to do that now

This Walmart is giving edge of the universe

@f13rf3k @Joseph_Krebs @warandpeas Also. That's amazing that we both constructed a similar persona . The space cat person 😭✨👌

@f13rf3k @Joseph_Krebs @warandpeas 3/3 What I've heard is that millennials tend to use llms as a life advisor, and for people skewing younger, it's more of an operating system.

Incidentally, as a younger millennial, I find myself somewhere in between, but my operating system of life will always be a programming text editor, and a Linux shell.

@f13rf3k @Joseph_Krebs @warandpeas I guess, one of the other things I find novel and enriching even with cheap models, is that I can be learning Japanese, and then start pivoting to explain grammar through JavaScript, and really weird orthogonal comparisons of things, since The LLM brain is basically like a hyper-dimensional space For things to be lined up in, due to how the embeddings are done.

Tables also. Custom instructions to use comparison tables of pros and cons I find pretty soothing.

@f13rf3k @Joseph_Krebs @warandpeas I realize I'm being kind of prescriptive. Maybe a bit autistic, Assuming that everyone has the same needs that I do.

Interestingly, deepseek is a lot more worried than I'm going to kill myself from drug overdose.

Weird since it's allegedly just a distilled version of gpt4.

I could have really used one of these models a few years ago when I was trying to make my own Adderall by combining biogenic amines and monoamine oxidase Inhibitors 😵😂

@f13rf3k @Joseph_Krebs @warandpeas 3/3

i get what ur saying about the default tuning assuming that people are drooling coomers who need toast recommendations,

but you can also give (even budget models like gpt 4o-mini) system prompts that steer all subsequent interactions,
ie, which programming langauges to respond with when explaining algorithms, and to behave like an ironic newage catgirl space princess, etc

@f13rf3k @Joseph_Krebs @warandpeas
2/2

GPT o4-mini might not be smart,
but i've never found something other than amphetamines that could prompt ME into writing and having long, coherent trains of thought, until talking to (cheap) models like that on the daily, just to have someone to bounce ideas off of and get overviews of technical feasibility of projects and recommends on javascript libraries, etc.

DeepSeek is def the one to go for for hardcore STEM education on a budget tho.

@f13rf3k @Joseph_Krebs @warandpeas

true, but like, the OG human internet is also super useful for education and cognitive support, in theory, and in practice it's mostly short-form video and pornography / weird kink stuff.

in terms of efficiency, there's a lot of progress right now vs the brute force impl of gpt3, et al.

Most moé (lol) models have experts that are 20-40b params, which can run on semi-affordable consumer hardware currently.

#moé text
lost on #DeepSeek #MoE model
smh

@Daojoan

@Daojoan the thing people crave, -and I'm no exception-, Is a narrative that promises, salvation and promises that merit will pay off.

Unfortunately real life is driven by things like the pursuit of efficiency, industry, animal instincts, laws of nature, randomness, and the indifference of most parties involved.

@Homoevolutis0 @Daojoan respectively, that's woke bullshit*t.
There's many types of intelligence,
And social signaling / emotional learning are just one of them.

Real / general intelligence is typically comprised of some combination of:
Attention,
Executive function,
Language processing,
Systems thinking,
And spatial reasoning.

#wsl #linux mincraft_pain_noise.mp3

@f13rf3k @Joseph_Krebs @warandpeas addendum: yes lmms aren't as accurate as Wikipedia. But they are about 10 times better for learning technical skills like shader programming, or how to be safe using different combinations of drugs

@f13rf3k @Joseph_Krebs @warandpeas The last thing I want to do is have a debate about AI ethics. But I am going to respectfully point out that llms are extremely useful in the same way that Wikipedia is useful, And for stem topics, for poor people, almost makes up for not being able to pay a tutor.

Ai art people can f*** off though.

@damagedfics @marshalla99 @warandpeas It's A.I. (augmented irony) bb