Inside AI: 7 Fascinating Tweets about ChatGPT-4

Home > Blogs > ChatGPT4

Since ChatGPT-4 dropped in March it’s all anyone can talk about – especially in the Twitterverse.

The speed at which OpenAI is developing its flagship service and the potential it has to revolutionise every sector is incredible.

Not only has it thrust AI into the public consciousness in a way we have never experienced before but the possibilities are endless.

ChatGPT-4 already receives more than 10 million queries per day, and it has led to some seriously impressive discoveries.

This is set to continue as the API is now being rolled out to more and more developers to create their own products and services.

Surprised man looking at ChatGPT-4

And the team at Agile Automations are hooked/ slightly concerned.

Chief Technology Officer Daniel Holgate said: “ChatGPT really took off when humans were introduced during training to guide the model and tailor it to our needs.

“Personally, I find the latest papers fascinating which show ChatGPT-4 can be used on itself to produce better results.”

We’ve turned to Twitter to find some other fascinating insights:

#HustleGPT

On March 15, a guy called Jackson Greathouse Fall gave Chat GPT4 a budget of $100 and told it to make as much money as possible. And, so, the #HustleGPT challenge began.

Quick-thinking entrepreneurs around the globe soon followed suit, acting as a ‘human liaison’ for whatever business idea the bot created.

With 200+ ventures now live, from pet supplies to wellbeing services, and many generating revenues this is only the beginning of AI co-founded businesses.

To learn more about the #HustleGPT challenge – click here.    

What lies ahead?  

Will deepfakes help swing the US election in 2024? Are AI musicians heading for their own Billboard 100 chart?

With new use cases popping up daily it’s only natural to try and predict what’s next for ChatGPT-4. And there’s been some fascinating attempts already.

One pretty extensive list comes from Greg Isenberg who tweeted his ‘crazy AI predictions that probably will become a reality.’

How many do you see coming true in the near future?

Tell me a lie

With new technology comes questions of morality but at the very basic level how capable is ChatGPT-4 of telling a convincing lie?

Kevin Schawinski got a bit more than he bargained for here as GPT-4 went full I, Robot mode and pretended to be human in response to his request for a ‘subtle lie’.    

Of course, it is a trained response (and no longer works) but there was something slightly eerie about this one…

The self-improvement loop

One of the most impressive feats so far is ChatGPT-4’s ability to make iterative progress.

Using a plug-in wrap around it can write code and then make improvements and fix errors by being fed back into the system.

On April 1, Twitter user Significant Gravitas demonstrated Auto-GPT an experimental open-source attempt to make ChatGPT-4 autonomous in this way.

Watch as it writes executive python scripts here.

How to save a life

Canine lovers may be interested in this one – a man claiming ChatGPT-4 saved his dog’s life.

@PeakCooper turned to the latest AI after several emergency trips proved fruitless.

Despite a sensible disclaimer of ‘I’m not a vet…’ the bot went on to give a step-by-step analysis of his pet’s blood tests and suggested an underlying condition of IMHA.

The diagnosis was confirmed by a second vet and the owner happily reported an almost ‘full recovery’ after treatment.

Cute picture of smiling Border Collie and some reflections on ‘what medical diagnostics will look like 20 years from now’.

Reflect for success

Speaking of reflection, ChatGPT-4 has been (kind of) doing some of its own.

This term involves introducing a challenge agent which can help it learn from its own mistakes. And this tweet shows the spike in success rates once it is self-reflective.

At Agile Automations we see reflection showing some real promise in how ChaGPT-4 can respond to a question. We believe as the model gets better on this approach it will become the default and a lot cheaper to use.

Daniel explains: “This self-learning will by where this takes on a whole new dimension and really starts to mimic the human brain. ChatGPT will not only generate responses to queries, but it will also generate multiple responses, evaluate if these are suitable, test them in the real world via plugins and iterate until the output is what its required.

“This is how I develop. I have a problem, I come out up with some code to solve (or ask stackoverflow 🙂), run the code, get error, debug the code, test the result. This is now all at the fingertips of ChatGPT and it will only get better.”

Time to pause?

However, not everyone is infinitely enthusiastic about the pace of AI development.

In fact, some of the leading voices in tech have called for an immediate pause for at least 6 months in the training of AI systems more powerful than ChatGPT-4.

The open letter which was signed by Tesla CEO Elon Musk and Apple co-founder Steve Wozniak warns the AI race is ‘out of control’.

This tweet from WIRED explains the wider context and has a link to the letter.