New tech bad at old stuff

From the rise in popularity of ChatGPT and language models in general, I see two camps. Those who try and use ChatGPT as Google and find it gives bad results and thus think it’s useless. And those who start using it in new ways to do things they didn’t have technology for before and finds that it is super useful.

When new technology emerges, people have a tendency to try and use it for things we already do. The first TV programs were theatre plays. Marshall McLuhan talked about how it takes time before new technologies gets used for what the new technology affords, rather than just being used to do what we already can do. Mobile phones are not the same thing as the stationary phones in a household. Household phones did not allow for micro coordination, but required times and places to be agreed upon ahead of time.

ChatGPT and language models in general can do things we previously were not able to do. So we have a tendency to try and use it for what we already do. But what is interesting is not how well it can do things we already have good technology for. What is interesting is what new things it can do.

Previous tech could not generate a script for a book. LM can. They could not be used to write unit tests for already written code, or solve programming tasks. LM can. Current tech could not generate plausible text on a topic. LM can.

To evaluate the usefulness of new tech, it is not very relevant to ask how well it can do things we already can do, but to what new things it may allow us to do. And if this new thing is useful, then surely the tech is.

How bad it is at other things, become rather uninteresting. Don’t use it for that.

In closing, here is what ChatGPT thinks Marshall McLucan would say about this new technology:

One aspect of large language models that might have interested McLuhan is their ability to process and generate human-like language, which he would likely see as an example of how media and communication technologies extend and amplify human capabilities. McLuhan believed that media and communication technologies have the power to transform society and culture by changing the way people think, communicate, and interact with the world around them.

It’s possible that McLuhan might view large language models as a significant advancement in the development of artificial intelligence and as a tool that has the potential to transform how people communicate and interact with each other and with machines. He might also have seen the emergence of large language models as an example of how media and communication technologies are constantly evolving and shaping the way we understand and engage with the world.

Posted in Uncategorized

Doing Neural Networks

Once a year I’m teaching a course called Design & AI. During this time I get to play around a little bit more with machine learning.¬†As this technology is becoming increasingly available to people, I find it important for people to understand what’s going on inside them. Much of it seems complicated when presented in abstract terms, but when looking at it in practical terms through using modern tools it is not as complicated as one may think.

After being asked by colleagues what is involved in implementing and training neural networks, I gave a seminar on the topic this week called Doing Neural Networks. I forgot to record the actual seminar but made a video with the same content afterwards as people asked for it. So, feel free to watch the video below where I explain what neural networks are through code.

Posted in Uncategorized