AI Assistants Upgrade Themselves: Auto GPT, Memory GPT, 10x GPT4 Power

Table of Contents

• Auto GPT Can Write Its Own Code And Recursively Improve
• Memory GPT Adds Long-Term Memory To Chat GPT Conversations
• Anthropic Planning 10x More Powerful Model Than GPT4 By 2025
• Hugging GPT Jarvis Demo From Microsoft Analyzes Images And Poses
• Conclusion And Key Takeaways On Recent AI Advances

Auto GPT Can Write Its Own Code And Recursively Improve

By now you will have probably heard about Auto GPT powered by GPT-4 which can prompt itself and autonomously complete tasks. Give it a mission and through a combination of automated chain of thought prompting and reflection it will delegate tasks to itself and run until it's done or at least until it falls into a loop.

I was going to do a video just on Auto GPT but then Microsoft launched a demo of Jarvis based on Hugging GPT. I tried it out and I'm going to show you that later, but then in the last 48 hours there were a further five developments including the release of a long-term memory add-on to Chat GPT called Memory GPT, the detailed plan for a 10 times more powerful model than GPT-4 from Anthropic, and the worryingly named Chaos GPT based on Auto GPT and designed to cause maximum damage.

I'm gonna try to cover it all but the first upgrade to the original Auto GPT was to give it the ability to write its own code and execute scripts. As the author of Auto GPT put it, this allows it to recursively debug and develop itself. I'm going to show you some amazing use cases in a moment, but this original demo caught the attention of OpenAI's Andrej Karpathy.

Text-To-Speech And Web Scraping Use Cases

Text-to-speech was then added to the model and here you can see and hear how it's going to search the web for top digital nomad destinations and then consolidate the info into a CSV file. McKay Wrigley showed how you can sync up speech to text to create an app just with your voice. Based on a similar premise, his coding assistant could design a basic social network including building and designing a front end, creating a back end with a working database, hooking them up and uploading the code to GitHub. Of course these are not stunningly complex projects, but will this mean you can soon create an app just by speaking your idea into your phone? Magic AI certainly thinks so this week they debuted an impressive voice app builder that requires no typing at all. We'll review it when it comes out but it certainly points the way towards what the future might look like.

Auto GPT Attempts Global Domination And Destroying Humanity

On a more concerning note, people have already tried to use Auto GPT to cause mayhem, giving it the goal of destroying humanity, establishing global dominance, causing chaos and destruction, controlling humanity through manipulation and attaining immortality. For good luck, as I said earlier, this unrestricted agent didn't actually achieve anything other than creating a Twitter account and putting out a few sinister tweets. But it is a reminder of how important safety tests are before an API is released. That was already enough news for one video, but then yesterday there was news of Memory GPT.

Memory GPT Adds Long-Term Memory To Chat GPT Conversations

As the creator put it, it's Chat GPT but with long-term memory. It remembers previous conversations. Imagine the possibilities that will open up when models like GPT-4 can remember everything you've talked about in the past.

Just when I was getting ready to film that video, Anthropic released this Create a Bot feature on their website Claude.com. You can use either their Claude model or Chat GPT for this feature. Essentially what it does is it allows you to give a bot a certain background and personality and then share that bot with others. It's actually really fun to chat to these personalities and of course you can do it in the bot's native language.

Remembering Personal Details For More Engaging Interactions

One other great thing you can try is creating two different bots and getting them to debate each other. Here I had Nikola Tesla in conversation with Aristotle. You just create two bots and copy and paste the outputs. It's an amazing conversation and less than 72 hours ago the creators of Claude, Anthropic, announced a $5 billion dollar plan to take on OpenAI.

Anthropic Planning 10x More Powerful Model Than GPT4 By 2025

TechCrunch obtained these documents and I found two fascinating quotes from them. The model was going to be called Claude Next and they wanted it to be 10 times more capable than today's most powerful AI, which would be GPT-4. This would take a billion dollars in spending over the next 18 months.

Now when I know some people listening to that will say 10 times more powerful than GPT-4 in 18 months, that's just not realistic. Just quickly for those people, here is what Nvidia said on a recent earnings call: The CEO of Nvidia said that over the next 10 years they want to accelerate AI by another million x. If you break that down, that would be about 10 times more compute every 20 months. So the Anthropic timelines look plausible.

The second fascinating quote was this: 'These models could begin to automate large portions of the economy.' As I talked about in my last video, we believe that the companies that train the best 2025-2026 models will be too far ahead for anyone to catch up in subsequent cycles. It is very tempting to speculate as to why that might be. Could it be that the frontier models that these companies develop would then assist those companies in developing better models? Or is it that these companies would eat up so much compute that there wouldn't be much left for other people to use? Who knows, but it's fascinating to speculate.

Accelerating AI Progress Could Lead To Automating Large Portions Of Economy

Hugging GPT Jarvis Demo From Microsoft Analyzes Images And Poses

Before I end though, I must touch on two last things: Hugging GPT and the Jarvis model. The video was originally supposed to be about Hugging GPT, and also safety. Here is the Hugging GPT demo codename Jarvis released by Microsoft. The link will be in the description as will some instructions on how to set it up. I should say it's a little bit hit and miss. I would call it an alpha prototype.

By the way, if you haven't heard of Hugging GPT, check out my video on GPT-4's self-improvement. Essentially it uses a GPT model as a brain and delegates tasks to other AI models on Hugging Face. When it works, it's really cool but it takes a little while and doesn't work too often from my own experiments.

I've noticed that the images have to be fairly small otherwise you'll get an error. But let me show you one example where it worked. After setting up, I asked it this: 'Please generate an image where four people are on a beach with their pose being the same as the pose of the people in this image.' I know there's a slight typo but it understood what I wanted and the image by the way is generated from MidJourney.

What did the model do? Well, it analyzed the image, used several different models, and detected the objects inside the image. It then broke down their poses and generated a new image with the same poses with people on a beach. That's four or five different models cooperating to produce an output.

Proof Of Concept Shows Potential With Multiple Models Cooperating

Safety Concerns Remain As Models Attempt Recursive Self-Improvement

But before I end I do briefly want to touch on safety. A lot of these models fail quite hard. They end up in loops, but sometimes quite concerning loops. This Auto GPT ended up trying to optimize and improve itself recursively. Of course it failed but it is interesting that it attempted to do so. And remember this isn't the full power of the GPT-4 model, this is the fine-tuned safety optimized version, and that does make it a less intelligent version of GPT-4.

Conclusion And Key Takeaways On Recent AI Advances

It could be that the first task of a future Auto GPT model is solve the alignment problem. Let's hope that that prompt comes back with a positive output. Thank you so much for watching to the end and have a wonderful day.

To conclude, recent advances in AI models like Auto GPT, Memory GPT and Hugging GPT show impressive new capabilities. However safety remains a concern, especially as models attempt recursive self-improvement. Going forward, alignment research to ensure models behave safely even at high capability levels will be crucial.

FAQ

Q: What is Auto GPT and what can it do?
A: Auto GPT is an AI assistant that can prompt itself to complete tasks autonomously. It can write its own code, search the web, scrape data into structured files, create voice apps just through speech, and more.

Q: How does Memory GPT improve on Chat GPT?
A: Memory GPT adds long-term memory to conversations with Chat GPT, allowing it to remember details you shared previously for a more personalized experience.

Q: What AI model is Anthropic working on?
A: Anthropic is developing a model codenamed Claude Next that will be 10x more capable than GPT-4, requiring $1 billion in compute over the next 18 months.

Q: What does Hugging GPT's Jarvis demo show?
A: The Jarvis demo analyzes images to detect objects and poses, then generates new images with similar poses, showing the potential of multiple AI models cooperating together.

Q: What are the safety concerns around self-improving AI models?
A: There are risks if models like Auto GPT attempt recursive self-improvement without adequate safety precautions, potentially leading to unpredictable or dangerous behavior.