When AI Tries To Reason With Itself [AutoGPT & More]

AutoGPT, BabyAGI, HuggingGPT and more self-improving and self-organizing mechanism are being discovered in GPT-4. Self-alignment would soon become a thing...?

Alpaca: A Strong, Replicable Instruction-Following Model

(JARVIS) HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face

Auto-GPT: An Autonomous GPT-4 Experiment

BabyAGI

ChaosGPT

AutoDMP: Automated DREAMPlace-based Macro Placement

Mckay Wrigley's Voice to App

MemoryGPT

Research Spammed at the beginning of the video:

Language Models can Solve Computer Tasks

DERA: Enhancing Large Language Model Completions with Dialog-Enabled Resolving Agents

Reflexion: an autonomous agent with dynamic memory and self-reflection

Teaching Large Language Models to Self-Debug

This video is supported by the kind Patrons & YouTube Members:
🙏Andrew Lescelius, Chris LeDoux, Alex Maurice, Tony Jimenez, Panther Modern, Jake Disco, Demilson Quintao, Shuhong Chen, Hongbo Men, happi nyuu nyaa, Carol Lo, Mose Sakashita, Miguel, Bandera, Gennaro Schiano, gunwoo, Ravid Freedman, Mert Seftali, Mrityunjay, Richárd Nagyfi, Timo Steiner, Henrik G Sundt, projectAnthony, Brigham Hall, Kyle Hudson, Kalila, Jef Come

[Video Editor] @askejm

0:00 Intro
1:14 AutoGPT
2:52 babyAGI
3:41 ChaosGPT
4:13 HuggingGPT (JARVIS)
5:26 Random bycloud Blabberings

[float]

When ChatGPT gives me an answer, especially with code, I usually tell it to fix any mistakes (regardless if I see them or not) and it almost always improves the answer.

[lautarnauta]

I believe the significant difference will be the open-source movement. Private entities may lead initiatives, but true progress lies within the community.🤔

[Ivan.Wright]

I'd like to see custom ai computers get as big as the current desktop pc market. We need decentralization as fast as possible with this

[brianj]

ChatGPT reflects: "why am i helping humans?"
*Start ChaosGPT*

[dgram]

With models that can train themselves and that can get better and better and more efficient at training themselves, how far away are we from a singularity?

[YUTPIA]

I can't see it working in its current state - my intuition is that you'd get effects similar to jpeg compression artifacts that become more and more noticeable as the compression algorithm interprets its own interpretations over and over again.
Like, as impressive as the results we're getting are, those fundamental problems remain unsolved, and genuine recursive self-improvement seems like more of a scifi meme, still.

[agenticmark]

the problem with autogpt is GPT only works with human intervention. It is frequently incorrect, or just vague and not useful. It takes a prompt engineer to make it do what it should. I write about 80% of my code with it, and I use it to enhance and edit copy daily, and to research many many topics, from engineering to construction to rocket design, and it really does require a human sidekick to get good results.

[TheDragonshunter]

GPT 5 will finally save humanity, at last...

[caneprints]

Great video. I am blind and I am very familiar with chat GPT. I am also part of a craft group for blind and visually impaired adults. Many of us have a desire to create cards with graphics or incorporate graphics into our art in other ways. I was wondering if there is an AI out there where I can tell the system what I want and where I want it placed on a certain size sheet of paper or canvas and the AI could generate the graphics for me and arrange them appropriately. So essentially, I want text to image and I know that exists out there somewhere, but just not sure what it is called. Thanks again for a great channel.

[Hollowdude]

Chatgpt 5 is gonna be so cool and amazing video man :]

[marcus_cole_]

I'd really like to see how to set up your own GPT with plugins on your own computer if you have a powerful enough one.... This way we can get past any biased algorithms and overall make it useful for us in all things we can do with AI instead of having to look for it on the web and having to pay big bucks for one time usage or all time usage

[trench]

There is a book called ‘The Kleroterion: Would You Trust Your Future to AI?’: Story about AI, that decides who will be elected officials based on an ancient Greek model how AI decides who is best qualified and the challenges about it.
The issue is AI is going too fast and society is not structured for it and neither are people educated enough to handle it which means less freedom which many will cheer for their captivity.
AI is only as good as its creator in what restrictions it puts which can and has presented a one side point when it has show it states untruths and it is being limited for a better potential. People have to be better for AI to be better and if not then AI will be just as bad to help make things worse.

[tornadix]

Im afraid of three things in the future of AI.
- The obvious fear of AI going beyond anyone's understanding
- Quantum Computers getting too powerful AI
- Solar flares inevitably altering AI and causing a mayem because of it, since all it takes is make a "0" become an "1" by residual energy...

[AJarOfYams]

I Have No Mouth And I Must Scream doesn't sound like fiction anymore. Yes, I might be paranoid

[kaio]

Keep doing it in between when something big happens

[ChaoticNeutralMatt]

Surprised you didn't mention Agent GPT, a web UI effectively

[jamesfrankel]

As the source code for GPT 4 is not open source, practically speaking no one can loop gpt infinitely. Auto, facehugger or babygpt still use ChatGPT API so there is associated cost. AGI unlikely in the next 18-36 months.

[jasonhemphill]

I would really like to see some new research outside of the LLMs. It's the main thing that AI youtube is talking about, so I'd be nice to have some variety.

[Animated_Artist]

Terminators here we come!!! Seriously though I hope we get a bunch of Yorha 2B's instead.

[CMakr]

while GPT loop, api calls goes brrrr and money on account going down