Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

[early draft] AI Is Not Designed For You (Patreon)

Published:
2024-11-28 16:06:43
Imported:
2026-06
Tags:
haven't archived this post yet. have a subscription? use the importer!

Downloads

  • 384715395.mp4

Missing 1 full-res photo, 1 video, as this post has not yet been imported.

Content

How long is it going to take until AI is useful, and why is everyone raving about this extremely mid tech?
Here's my theory!

(PR for comments/corrections here: https://github.com/0atman/noboilerplate/pull/77)

Thank you all so much for supporting me, I hope you enjoy this shorter video. I'm still working on my deep-dive 'second brain' Obsidian video, an extremely advanced look at my whole ecosystem of living and thinking inside plain text.

Happy Thanksgiving for those who celebrate, I'm fortunate enough to have American friends to hang out with tonight, so I'm expecting a food coma tomorrow (as I understand is traditional).

Thank you again,
Tris <3

Files

Previews only

Comments

Jason Doege

This is a really bad take and I think you made it for two reasons. 1) you appear to have an anti-capitalism bias and 2) you are an amazingly competent human being as demonstrated by being able to use and prefer NixOS and training yourself to use alternate keyboards, for just two instances. You seem to overlook the fact that many people, myself included, are obtaining great value from LLMs. You also completely discount all the smaller things AI does. For instance, I can verbally ask Apple Photos for all the photos of a family member by name having never tagged them as such and it returns them. Accurate voice recognition, inference based on various sources of information and image recognition are all AI-based and are deployed so generally and subtly that we hardly notice. No doubt there is a lot of irrational exuberance about AI, but there is also a lot of obvious capability left on the table that AI companies seem reluctant to roll out in general release. Here are some: * Incremental re-training and active memory * ability to write and execute computational code * ability to look up information for more accurate answers * bicameral reasoning (two models talking to each other tend to produce better answers) I use AI features all the time, probably without even knowing it, but I also use LLMs several ways, regularly, to great effect. I use it to summarize business meetings from transcripts having it extract salient points and action items. This lets me pay attention rather that focus on taking accurate notes. I use it to quickly determine how to represent an idea in code, especially in unfamiliar languages. I use chatGPT, the way I expect to be able to use Siri in the not too distant future, to make an information query verbally and get back useful information. It carries a conversation fairly well, now, too, accepting suggestions for refinement and elaboration. So long as you don't just believe the hype and just use it what it is good at, there is a lot of benefit to these things and they are getting better rapidly.

-g2

Welcome back from RSI. The phrases I keep saying are "AI is all A, no I" and "It's mainly automated plagiarism". I'd like to also add, does anyone working in tech remember the hyperventilating hype and flowing funds that burst about 2015-2016 for all things VR? Like cicadas, AI is the latest hatch of this industry brood.

noboilerplate
>>156684144

I don't think we disagree on the tech - All the things you've described are normal machine learning features that improve our tools all the time. Perhaps I should be clearer about the "hype" nonsense at the start that I disagree with, and mention the features that are great, what LLMs are good at? Oh, and yes, I absolutely have an anti-capitalism bias. :-D (Not anti commerce, anti-capitalism)

noboilerplate
>>156685374

Thank you so much! Hey, VR games are fun! But as facebook/meta found out, it's not an infinite money machine! :D

Dan MacDonald

Heh, every new idea is heretical by definition. New technology is always rough around the edges because it hasn't had the years of polish to make it the mainstream choice. ChatGPT didn't become the fastest product ever to reach 1Bn in revenue because it was failing to do things people wanted. Since inception Bitcoin has been a hedge against inflation and a way to preserve purchasing power in savings. Along the way it became the single best performing investment to hold for the past 10 years (even more so if you take its full history since 2009). These things have real utility and their market price represents the value people place on that utility. I suspect the takes in this video will not age well. That said, kudos to you being brave enough to put your opinions out there for people like me to judge. 🍻

Companion Technology

Totally agree about altcoins; though Bitcoin is worth understanding

Bastian Asmussen

Glad to have you back! :D First of all, I want to preface this by saying that I agree with most, if not all points here. The thing about decentralization is that it is always inherently more of a hassel to set up rather than spinning up an EC2 instance on AWS for your server. Yet, even so, for downloading files I truly prefer the decentralized torrenting method and I think it really is a shame it gets so much flack for being a piracy tool. I more see it as a way to aliviate the download servers from whatever it is I'm downloading (Linux ISOs). For the take on LLMs I cannot agree more, I tried out things like GitHub's Copilot and ChatGPT but found that the tool more got in my way than helped me out. I sometimes ended up trying to reason with the model rather than be productive. I think we are in an AI bubble right now, which means people love using and abusing LLMs. What I mean by this is people redneck-engineer the models to do things they aren't good at. Like hiring a blind person to identify faces by looking, using an LLM for something that isn't related to natural language such as complex mathematics or non-standard programming problems it won't go well for it. It doesn't have sentience, LLMs and the GPT model work on statistics: "what is the most likely word coming after this word?" There is no part of that which will reason about logic. As an experiment I gave it a rather complex Nix expression from my NixOS config and it hallucinated. I expect it to do so, because there isn't any documentation in mass for Nix like there is for, say, Python, and especially not for my custom expression. Making it reason about Rust lifetimes and lifetime elision rules will also just give bogus output at best. However, using an LLM for what it's good at will give some very nice results, say we give an LLM some documentation on an internal product our billion dollar startup uses and instructs it akin to the following "this is all you know, if it is not listed in the docs, you cannot answer." It'll probably work really well as a chatbot for our fake startup. :) Sorry for the long rant! Glad you're feeling better again. :D - Bastian.

Packtion

I think an even stronger comparison would be to the more standard tesla vehicles. Over a million Model Y cars sold with the promise that the car will be able to drive itself by 2018. Now we're being sold the idea that one day AI will be able to do all the thinking and innovating for us.

nicklredwood

I feel like you didn't complain about the 1919 decision _Dodge v. Ford Motor Co._ enough and how it's the entire reason the fiduciary obligation even exists