ChatGPT Opens A Research Lab…For $2!
❤️ Check out Lambda here and sign up for their GPU Cloud:
Guide for using DeepSeek on Lambda:
📝 The paper "Agent Laboratory: Using LLM Agents as Research Assistants" and "Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researchers" are available here:
📝 My paper on simulations that look almost like reality is available for free here:
Or this is the orig. Nature Physics link with clickable citations:
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Benji Rabhan, B Shang, Christian Ahlin, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Michael Tedder, Owen Skarpness, Richard Sundvall, Steef, Taras Bobrovytsky, Thomas Krcmar, Tybie Fitzhugh, Ueli GallizziIf you wish to appear here or pick up other perks, click here:
My research:
X/Twitter:
Thumbnail design: Felícia Zsolnai-Fehér –
Thank you for your work 🙏🏼
You are too kind, thank you so much!
I literally dreamed of this this morning😅
What a coincidence 😉
It needs a human in the loop…for now.
But extrapolate where this is going…two more papers down the line.
Will we need a human in the loop tomorrow?
God I hope not.
We could be AI pets, in a decade, if we’re lucky.
Agent networks?
I’ve been waiting for this 😊
As if we didn’t need any more evidence that AI is just a GOAP AI with settings created by LLM data, we now see them just putting AI into a game like The Sims or Rimworld where they behave like GOAP agsnts with chatbot functionality.
i would kill to see multiple ai playing rimworld, each one is tasked with their own interests first and their traits affects their personalities. now ai groups speak to each other in order to persuade one pawn to risk their lives for common good.
spread hundreds of random pawns across the map with no npcs and watch them form societies, wage wars and make diplomacy
Decentralized Virtual Organizations with Personal AI Assistants / Agents and Virtual Workspaces for research, development, collaboration, and deployment. 😎🤖
What a time to be alive!! 😀
OK. I want to test them with the ultimate question of life the universe and everything. Will they create Earth?
No because God is Creator
Why? The answer is 42. Don’t need AI for that one.
“The AI does not invent fundamentally new things unless it meets human intelligence” *for now*
Have to hope it doesn’t invent the new thing that kills all humans, or it will be, “what a time to be dead.”
This is a perfect example of how AI isn’t here to replace human researchers but to amplify their capabilities. The ability to rapidly analyze vast amounts of data, summarize findings, and suggest new directions makes AI an incredible co-pilot in discovery
Research assitants rip
Keep coping.
Why russian isn’t work?
🎯 Key points for quick navigation:
00:00 *🧠 The video explores the concept of using multiple ChatGPT instances to simulate a research lab with roles like professor and PhD student.*
00:29 *🎮 Inspired by previous experiments with video game characters, the idea is to give AI agents motivations and simulate interactions, like elections in a virtual town.*
01:28 *📚 The research process starts with a human idea input; multiple AI agents then collaborate to conduct research and coding, resulting in high-quality outputs.*
02:25 *🏅 The AI-driven research shows impressive results, surpassing previous methods on various tasks, albeit with limitations on certain languages.*
02:52 *💰 The entire process of utilizing these AI agents to perform research costs only $2.33 and takes about 20 minutes.*
03:44 *📂 Open science is highlighted, with the availability of code and papers for free, showcasing the future potential of AI in research.*
04:14 *🤔 AI suggestions are novel and exciting but often lack feasibility, highlighting the need for human ingenuity in practical research.*
05:05 *🌟 The synergy of AI and human intelligence is crucial, as shown by achievements like AlphaFold, which required a blend of AI and human brilliance.*
Made with HARPA AI
Faster to just watch the video with those key points…
I’ve been thinking about this idea for a few years now. One of the problem of AI is the context length. It does not have enough “memory” to make the output actually useful or long enough. And using and increasing the context length, just gives diminishing returns. My idea is to somehow hook up ten or even a hundred of small LLMs, to communicate with each other, and to have more focus on a specific part, and the thought is that you are technically increasing the context or memory, by using multiple AIs, as each has their own little memory segment, maybe even having a “wisdom of the crowds” effect. How you would let these communicate? I don’t know, but it would be interesting to see how it plays out.
“One problem of the AI is the context length” – not for long, bycloud made a video about Googles “Transformer 2.0” model which will significantly improve context length. At that point one might have enough context? Now if we were to have multiple of these high context AI with a specific task…
https://www.youtube.com/watch?v=UMkCmOTX5Ow
As for communicating: It is not unfeasible that one could have them evolve their own “language”. Such properties have emerged before, the question is if it is just junk or if they can actually use that to generate useful output in natural language. I suspect such a language would have dimensions not found in human languages if it is possible.
we in the matrix for sure the question is what are humans building for their captors
They can’t cure hair loss
Researching emergence could open some impressive doors.
And nothing of value was built lol. Still no actual product out there.
Next Sims game is gonna be nuts.
This sounds like a job for Minecraft GPT !