Fully Autonomous AI research: Data to paper with ChatGPT?

In this video I share with you recent reports on fully autonomous AI research with data to paper generation.

▼ ▽ Sign up for my FREE newsletter
Join 19,000+ email subscribers receiving the free tools and academic tips directly from me:

................................................

▼ ▽ TIMESTAMPS
0:00 – the tweet
2:21 – errors in code
5:19 – literature gap
6:52 – hallucinations
8:47 – P hacking
9:31 – copilot
10:58 – wrapping up

................................................

▼ ▽ RECOMMENDED ACADEMIC TOOLS AND SERVICES

AI TOOLS

▼ ▽ RECOMMENDED YOUTUBE TOOLS AND SERVICES

▼ ▽ Socials for shorts and reels

[geoffreyanderson]

Mr Stapleton's knowledge is perfect for helping make a training set for a paper critic agent. The agent can listen to transcripts of this youtube channel and look for and use his advice directly to improve its capabilities. Mimicry is a form of learning. People like Mr Stapleton are suited to contributing to the growth of paper critic AI agents. It would preferably require yet another AI agent to automate the training of a paper critic agent in the next phase (just reading his youtube video transcripts for the purpose of learning his ability to identify known weaknesses in known scientific papers) but before that happens in phase one, humans like him can show good examples and counterexamples manually with good prompt engineering.

[yasha.hartberg]

It would be interesting to dive just a bit deeper and see if ChatGPT really wrote from the data it was given or whether it sneakily went out and plagiarized results from other papers written from the same data set. As you mentioned, its findings weren't in any way novel. It's at least plausible that it reverse-engineered its paper from others out there.

[me]

I would LOVE to see this applied to more complicated systems like in engineering, chemistry, physics, and maybe even mathematics. Unfortunately I don't see this going very far as the datasets get bigger and with more moving parts. We'll have to find ways to consolidate variables, find bridges between various seemingly unrelated concepts, and other ideas that I probably can't even imagine right now.

[iddqd_]

🎯 Key Takeaways for quick navigation:

00:00 📝 Researchers claim AI generated papers autonomously from data.
01:23 🧹 AI helps clean and analyze large datasets effectively.
03:15 📊 Researchers used a methodical process of prompts to generate the paper.
04:24 🚫 AI-generated paper lacks novel insights and could struggle with peer review.
06:02 📜 AI-created citations have issues, and paper references are outdated.
07:39 💡 AI's potential in research, like chemistry tasks, is promising.
09:04 🕵️‍♂️ Concerns about AI potentially facilitating P-hacking in research.
10:09 👥 ChatGPT can assist as a co-pilot in research but needs human oversight.
11:19 📧 Sign up for the creator's newsletter for more information on academic resources.

Made with HARPA AI

[nondescriptnyc]

I wonder if chatbots’ inability to handle citations/references may be related to many journals limit access. If that is the case, with open access publishing becoming more normative, this issue should resolve itself…

[AiLatestNews]

Fascinating topic! 📚🤖 The idea of AI generating research papers from data is mind-blowing. Can't wait to dive into this video! 🚀

[geoffreyanderson]

This project's weaknesses can be addressed and built upon. Make another AI agent that criticizes the novelty of any paper, and make the original agent keep working to find and synthesize novel topics until the critic agent(s) is satisfied, that is, they can't find sufficient coverage of the candidate topic. Reuse of the proposed paper critic agent is evident too -- the journals can use it for their own purposes, and so can academic advisors.

[jwjwj]

Thanks for the great video. I've seen the future and it's not here yet.

[snipe]

I wish I could see if it used plugins like Scholar AI, which is great at finding the resources necessary for writing literature reviews.

[DreckbobBratpfanne]

I wonder if a web search ability itself could already help a lot "search for new papers in this area and use the knowledge to improve your work" eg

[shinysheep]

I'd like to show your videos to colleagues who might be interested. Since their English is not up to the speed of your speech:
any ideas which AI I could use to transcribe your videos? And which AI to get an abstract of the content?

[BenTheBSC]

Great video as always - a small editing error at 10:42, just a heads up

[nineteenfortyeight]

Omg 😭 There was already so much crap out there. Now there's so much more.

[avaraportti]

The difference between ChatGPT and the average academic is that ChatGPT is quite a bit less of a formulaic NPC

[dave]

These AI videos are making me anxious. Where will all of this lead to when the technology improves? Is research and academia becoming a thing of the past?

[geoffreyanderson]

Checking each and every paper reference is a boring job, perfect for automation.

[kategnidenko]

Pappers written by AI will be peer-reviewd by another AI, and read and reused by some other AI. We won't need scientisats and their publications, the system of publications will turn authonomous

[imranq]

Why write a paper when only the AIs will read them

[efkanturan]

Seriously this guy always click baits us. Stop pretending in the intro you found something amazing that will help us all out and then let us down later as usual.

[nadine]

So in conclusion: Artificial intelligence isn't all that intelligent. :D