The AutoGPTs Are Coming

Hello Everyone,

As large language models become multi-modal, they also begin to take on more functions of chain-of-prompting, agency and auto-GPT like capabilities. As LLMs go from billions to trillions of parametres many of us expect GPT-5 like LLMs to display AGI-like abilities.

This is both exciting and extremely daunting for how humanity relates to A.I. One thing is for certain, it will enable A.I. to do a lot more of the tasks of white-collar and knowledge workers in general. I explored this topic recently here.

Multi-modal chain of thought reasoning could lead to unexpected AGI developments. As OpenAI’s APIs among other things would become interoperable and LLMs would be able to use more specialized LLMs to figure out things in a more coherent way. This will likely take a few years to figure out but could lead to a drastically different world.

The GPT stands for Generative Pre-trained Transformer, and there is some debate about whether it is a General Purpose Technology GPT. Even Microsoft Researchers seem to hint that GPT-4 is a significant step towards AGI.

✨ From Our Sponsor

Outsource card data management. All the benefits, none of the headaches.

Collect card data, send it to processors or partners, and store it as if it's in your database while satisfying up to 95% of the compliance requirements that come with PCI.

Learn More

Another way to view this is that Generative A.I. will lead to a functional general purpose operating system (OS), in likely just a few years time. That is, A.I. is about to get way more useful, well beyond anything ChatGPT is capable of doing today in 2023.

Let’s take a closer look at Auto-GPT, for example.

Andrej Karpathy is someone who seems very bullish on AutoGPT and the capabilities of LLMs to become more living actors in the world at large. He was head of A.I. at Tesla and recently joined back with OpenAI.

Read his Tweet Thread on AutoGPT

(Are Tweet embeds again blocked on Substack?)

• In short, Auto-GPT (An Autonomous GPT-4 Experiment), a free program demonstrating how LLMs like GPT-4 may be used to develop and handle various activities independently, like writing code or developing business ideas.

• Concepts and experiments like AutoGPT might be the next big step in AI. Karpathy appears bullish and recently said "AutoGPT is the next frontier of prompt engineering".

• AutoGPT is the equivalent of giving GPT-based models a memory and a body. You can now give a task to an AI agent and have it autonomously come up with a plan, execute on it, browse the web, and use new data to revise the strategy until the task is completed. Think about it, this could be leading somewhere.

Here is AutoGPT on Github: https://github.com/Torantulino/Auto-GPT

As how we interact with the coming AGIs improves, we’ll likely all have non-invasive brain-computer-interfaces (BCIs) that will make ambient computing finally a reality. While rudimentary, experiments like Auto-GPT (which is a GPT-4 model), however, points to a future further than chatbots, by promising to be able to do tasks like market research all by itself, write snappy headlines and generate entire blog posts without much prompting.

LLMs will impact white-collar tasks likely much faster than robots will impact the jobs of blue-collar workers. In the story of automation this is not always what we had assumed would happen. Our entire vision of the future of work needs to be adjusted to the current reality of R&D in A.I. and the likely outcomes in terms of AGI that has more agency and is able to do more complicated things. There’s likely even to be a period where there is a Machine Rebellion against the Machine Economy. This will also depend on how the AGIs of tomorrow are used.

Startups like OpenAI and Anthropic, have big plans in how this comes about. It’s not clear if they will even be able to remain to be independent companies or get swallowed up by Microsoft and Google, respectively. China will also catch up with more regulatory scrutiny.

X-Corp (formerly Twitter) is even planning to make a ChatGPT rival. So has Alibaba, and scores of others. The anticipation for GPT-4 leads to the rushing approach of GPT-4.5 and GPT-5 but all of this pails in comparison to what AGI becomes, perhaps even in the next five to twenty years.

Even with something like Auto-GPT, when the model is provided with an identity, a role/task, objectives, and details about what it is supposed to do, it seeks to complete the task “autonomously” by using a framework that allows it to “reason and act” to do so. Smart prompting helps LLMs overcome their inherent limitations in areas such as the “content window” and “math problem-solving.” This suggests that we are moving towards a stage of artificial general intelligence in human civilization in the not-so distant future.

Andrej Karpathy’s Tweet on this has more than 1 million views. People appear curious if we are really heading to an AGI moment for humanity and what it could mean for them.

The NASDAQ 100 is up 18% so far in 2023, purely based on A.I. hype. Without all of this excitement, with a looming recession, things would feel very different. Now ChatGPT is a great tool that’s only five months old, but it’s what’s coming that’s the good stuff.

What do you think?

After I wrote this piece, I realized AutoGPT is actually trending on Twitter.

Related Reading:

1. Towards the Generative A.I. Operating System (OS) [Link]

2. Speculating on Multimodal LLMs and GPT-4 [Link]

3. GPTs are GPTs: An early look at the labor market impact potential of large language models. [Link]

4. Dawning of the Machine Rebellion [Link]

5. ChatGPT and Generative A.I's impact on our Jobs [Link]

6. Will Larger LLMs and ChatGPT Really Disrupt our Jobs? [Link]

I will continue to write about the potential manifestation of AGI and its impact on us, our jobs and our society and civilization.

Thanks for reading!

Disclosures about Our Sponsor

Basis Theory, is the compliance engineering platform. They are a compliant and developer-friendly platform for securing, using and managing the data that matters most to you. “Unleash the power of sensitive data faster”. Grateful for their support and what it means to me.

I may in the future sometimes feature Sponsored-Ads on free articles. Let me know if this is okay.