- Neural Nonsense
- Posts
- The Counter-Revolution Has Begun
Hello world, this is Neural Nonsense! The daily AI newsletter that's shorter than Napoleon, but just as ambitious.
We have got some smashers today:
The rise of GPTZero and the counter-revolution
What Reid Hoffman is looking out for in 2023
A $50K AI Hackathon
Let's get into it 👀
The Counter-Revolution Has Begun 🏴☠️
My faithful listeners! we had our first reply, bring out the confetti 🎊
How do you think AI, GTP-3 especially, can be policed within large organisations, so creative and content-specific roles, don’t use it to do their work for them? Do you think companies will have to track the use of the technology and immediately begin implementing policies and guardrails?
Good question Ben.
Well, we'll need to be able to catch people in the act first. And that's where things get tricky. Can we catch people?
Some folks, like Maggie Appleton are calling for "authored by humans" tags, whilst teachers are getting bamboozled by students. We are beginning to see big push back against generated text, and the millions of people it's going to disrupt.
The counter-hype of AI is in full stride.
The problem is, there's no way to know for sure whether a piece of text was written by a robot or a human. Until now.
Introducing GPTZero, the app that can detect whether an essay was written with ChatGPT. How does it work? Well, ChatGPT writes in a certain way, with a consistent "burstiness" and perplexity - and GPTZero can detect these patterns.
But there's a BIG problem with this approach (and all approaches like this). The people using ChatGPT don't want to get caught.
My thoughts? The genie is out of the bottle here, folks. The idea you can "fingerprint" text is doomed to fail, because as Charlie Munger would say "follow the incentives". And the incentive for people to use this 10x their productivity are way too great, and so they'll find clever ways around any type of detection.
All it takes is for a new model to vary the perplexity slightly, or for a human editor to thesauruses some of the words, and we're back to square one. Sure, OpenAI might have to start labelling text as AI-generated, but that won't stop people getting "creative".
Especially when doing nothing means leaving money on the table.
🔎 Everything Else
You can now generate SQL queries in seconds using AI, and even better, you can get the AI to explain it to you after. No more hocus pocus [link]
Microsoft have put together a free 12 week course for beginners, all about AI [link]
The recordings of the $50K AI Hackathon are out! [link]
What big investors like Reid Hoffman are looking out for in 2023 [link]
Okay, this is pretty cool - Building a language model from scratch [link]
🎨 DALL-E of the Day
This is what happens when one of my favourite artists starts experimenting with Midjourney V4 - Major Cyberpunk vibes [link]
Reply