Magical Search

by
on

I’ve watched the world of LinkedIn enthralled with GPT-Chat for weeks. There have been a plethora (my word for the Three Amigos fan who visit this humble blog) of articles on this being world-changing. After Microsoft released the Tay application and the internet corrupted it in 60 seconds, I give the folks at Open AI immense credit. A standing ovation even. I’ve been a user of Open AIs service for the last year plus, have built entire models on TensorFlow and PyTorch using my own writings, and continue to toy with this bot. First, there is nothing new here. Repos like Hugging face and others have been tinkering on this for eons. Google. Microsoft. But the presentation is remarkable and sets this apart. It’s like a magic trick. My personal bot is the home magician doing card tricks for their kid. And Chat is David Copperfield.

Here are a few random thoughts, positive and negative:

  • I could see this being the future of search. The results are far more helpful than what I find on Google; grant the dataset is older. The challenge is the business model and corresponding cost. When I built my model, I found it would suck all the resources on my machine locally, sometimes bringing my older Mac to its needs. Frankly, I’ve never seen a machine run this hot, and my own models took days to run. Before anyone shouts from the hills, you are running on a local machine; math still makes this problematic.
  • Surfacing a hyperlink is far different than coming back with a true answer. It’s a problem now but will it be in the future? The cost of computing is a race to the bottom. If you don’t believe me, review AWS, Google, and Azure storage costs for the past five years. Eventually, the perceived usefulness will reach the Mendoza line, a baseball term, and tilt in its favor. In the early 2000s, Content Management Systems were considered useless, slow, and impractical. Now, WordPress powers 40% of the web while pure HTML sites are now popular with writers and bloggers who don’t want to track cookies or leverage ads. Yes, technology moves fast. The winning cost of the hyperlink model will fade.
  • And then there is what it doesn’t know. I wanted Chat-GPT to tell me how many pitches John Smoltz had thrown in the greatest postseason game ever pitched before Bobby Cox pulled the plug. It was 106, give or take, which was hard to find. But GPT-3 had no record of the game or knowledge that he had thrown a single pitch in a World Series. This is blasphemy for baseball fans. Jack Morris refused to be taken out and tossed a complete game shutout across ten scoreless innings. That’s an impossible feat. Yet, our hero chatbot didn’t know. An easy fix, I’m sure. But, oh, I almost don’t want to ask it anything else. Grant, Google doesn’t do much better, again took me a half hour plus dredging through links to find an answer.

That being said, there is the awesomeness (if that’s a word):

  • For small businesses, I think this thing is a beast of a time saver. Write me an ad. It gives me an option. Write a summary. The paragraph is shortened. Research a topic. The AI comes back with an answer. Frankly, it’s the personal assistant for the lonely and downtrodden I’ve always wanted. Just don’t talk to it too much; you’ll think it’s alive. Either way, this product is a winner for entertainment alone. More importantly, it’s also useful.
  • Should writers be wary? For those who are writing simple copy, probably. But for those big-on ideas without an immense amount of time, I’d look at these as tools. When I use my models, I find they are often more troublesome than the corresponding value. But now and then, an idea might emerge, so what’s wrong with that? Often, I find these snippets amazing and wonder how the machine found the connection. What word choice led to that? And why? That is truly fascinating. But no, it’s not the end of the world. Life goes on. Don’t believe the worrisome articles being bantered about.
  • Is it accurate? Yeah, to a degree. I think it tries hard. When I asked the technology to write a short post on Hemingway, I ended up with tales of the famed writer walking across lands he had never visited. Kudos for creativity. More kudos for telling me like it happened. Here, OpenAI has improved immensely. Months back, OpenAI spewed nonsensical babble in their demos. Still, if you’re writing anything of value, I’d check the source. Fakery is hard to put to bed.
  • For a parent, this machine is fantastic for homework. Don’t know how many electrons are in Carbon? The tech even told me where to find an answer, which was better than diving through links for a solution.

Not Revolutionary but a Step

Lately, I’ve heard future tales that Google might be dead. If the company does nothing, possibly. I could see Microsoft dropping the engine into Bing, which might level the playing field. Big leaps in tech do take time. And then they just happen. Personally, I find searching for anything worthwhile problematic on the web these days. Anything is better than the current model.

Still, we move onward. High props for creating a helpful tool. And no part of this article was written by a chatbot. That’s too trendy these days, so I’ll continue to push in my own stubborn ways. Keeps me out of copyright prison until that’s worked out.

Notes:

  • OpenAI.
  • There are many tools out there using OpenAI’s API. You can also build your own.
  • Playing with old technology. This post is also highlighted in the article above. What’s missing is the commentary around copyright. Google was also considered revolutionary at the time but danced around the issue by referring users to the page. OpenAI bypasses this entirely by taking the data, creating a transformer, and then espousing an answer. In building my own transformer, an immense amount of tech is required to learn. The content doesn’t matter. So, when the machine writes a passage on famous magicians who is the copyright holder? We m right be debating this for years, which is probably why the company started as a non-profit.
  • Speaking of making money, that is no longer the direction as Microsoft is about ready to drop ten billion into the company.
  • Almost all of my pictures on this site were taken by me. However, the magic picture headlined by Henry Kellar above comes from the Library of Congress. They have a site with thousands upon thousands of free images. And since the chat feels a bit like magic (but it’s more of a prediction sleight of hand), I thought the poster was appropriate.

And Chat didn’t know John Smoltz, but the AI did know Kellar:

Yes, I’m familiar with Harry Kellar (1849-1922), who was a well-known magician and illusionist in the late 19th and early 20th centuries. He was born in Erie, Pennsylvania and began performing as a magician at a young age. Kellar traveled extensively throughout the United States and Europe, performing his magic acts for large audiences. He was known for his elaborate stage shows and illusions, including the “Vanishing Lady” and the “Levitation of Princess Karnac.” Kellar was also a member of the Society of American Magicians and the Magic Circle.