We are excited to announce The Next Wave Podcast with Matt Wolfe and HubSpot! Learn more

AutoGPTs and Google

Written by Nathan Lands

Thought Q1 was crazy? It's about to get even crazier with "AutoGPTs"!

Google is in big trouble. Here's why:

With GPT-5 expected to complete training in December, it seemed progress might slow. Giving Google, Apple, and others a chance to catch up.

But now, thanks to indie hackers and a VC that can't code, a whole new world of experimentation and progress has opened up.

A few days ago, I read this tweet thread from @yoheinakajima, a VC based in Seattle. He has created an experimental "AI founder" that can do tasks on its own called babyAGI.

I talked with him yesterday on video chat and learned a lot.

His story is incredible. He went from idea to code. From code to paper. Paper to viral tweet thread. Thread to open-source project on GitHub.

All in 3 days. Without being able to code.  All created with AI.

He fired up a @Replit and discussed the idea with GPT-4. Then fed it documentation from @LangChainAI and @pinecone.

It then created some code which he pasted into a Replit. The code had a few errors. He just kept giving the code to GPT to debug until it worked.

Then had it create a paper for him. Then a tweet thread based on the paper.

Check out this AI-generated diagram illustrating the process.

One GPT-4 acts as a task manager, keeping track of tasks and priorities. While another GPT-4 serves as a task creator. A third GPT-4 executes the tasks, with results saved in memory.

Theoretically, this could make GPT-4 self-improve. And carry out business tasks by itself. @yoheinakajima is considering using it as a VC associate for his fund.

He even has it set where he can get text messages to confirm actions.

And he's not the only one with this idea.

Yesterday I discovered AutoGPT created by @SigGravitas. They're blowing up on GitHub.

@karpathy, the ex-head of Tesla AI, stated it could be the next major frontier.

And @Altimor thinks something like this could possibly be a path to AGI.

Based on a poll by @SmokeAwayyy, others seem to agree.

It seems too simple to lead to AGI. But, as @ID_AA_Carmack stated last year on @lexfridman, AGI will likely be surprisingly simple.

So, what does this mean for Google and others?

@NickADobos nailed it.

OpenAI might have already won. The flywheel of improvement is spinning too fast now. OpenAI has a killer culture + plugins + hacker community.

Meanwhile, based on this tweet from @ricburton, it sounds like Google's culture is...different.

So what is Google doing in response? Pushing for regulation.

@gregisenberg thinks it's a move to slow things down. I agree.

And pushing out products that are really bad in comparison to ChatGPT. But claiming that is on purpose, as noted by @buccocapital.

Google is increasingly full of ads and bad content, and its problem will worsen as people mass-produce AI content. When ChatGPT solves the 'confident bullshit' problem, I'll probably rarely use Google.

In conclusion, community-created experiments like babyAGI and Auto-GPT are a big deal. They allow GPT-4 to keep improving while OpenAI builds GPT-5.

OpenAI is further ahead than people realize. And Google is in big trouble. If they didn't have YouTube, I'd even say they might be screwed.

As a disclaimer, it's worth noting that some people aren't impressed with babyAGI and Auto-GPT, such as @ctjlewis. And the tech is still very experimental.

And as @icreatelife stated, based on technical definitions, we're not close to AGI. But I have a feeling by next year, this will be heavily debated.

And many smart people, such as my friend @bilawalsidhu, think it's too soon to count Google out.

And who knows, maybe they're a sleeping giant and they'll respond in a way that shocks and awes.  But I'll believe it when I see it.

Here's the link to babyAGI:

GitHub - yoheinakajima/babyagi
Contribute to yoheinakajima/babyagi development by creating an account on GitHub.

And link to Auto-GPT: https://github.com/Torantulino/Auto-GPT

I'm trying both of these today and will be sharing what I learn along the way, attempting to automate parts of my business. Or creating new business units.

Will it work? I guess we're about to find out. :)