Summary

OpenAI's introduction of the 'gpt-3.5-turbo' model significantly reduces costs, making GPT integration more accessible for developers and enhancing AI tool adoption in applications like the language learning app Litany.

Abstract

The recent announcement by OpenAI about the new 'gpt-3.5-turbo' model has sparked excitement due to its 90% cost reduction compared to the previous 'davinci' model. At just 0.2 cents per 1000 tokens, this development makes GPT integration far more affordable and appealing for developers. The author, who has been developing an RSS reader named Echo and a language learning app called Litany, sees this as a pivotal moment for AI integration into various tools. Litany, which aids language learning through spaced repetition, now includes a feature that leverages GPT to clarify synonyms during study sessions. While the technology is impressive, it is not without limitations, such as occasional struggles with context and multiple word meanings. Nonetheless, the author is optimistic about the future of AI in learning applications and is considering further AI integration, such as an article summarizer.

Opinions

The author views the cost reduction of the new GPT model as a significant opportunity for developers to incorporate AI into their applications.
Previously, the high cost of GPT was a major barrier to its widespread adoption, but with the new pricing, it has moved from being "absurdly overpriced" to "extremely expensive," which is a step in the right direction.
The author speculates that competition in the AI space may be contributing to the reduction in costs, as more entities seek to develop their own versions of GPT.
The integration of GPT into the author's app Litany is seen as a valuable addition, particularly for its ability to help users understand synonyms, although it is acknowledged that the AI sometimes misses the best synonyms and lacks full context understanding.
Despite its current limitations, the author believes that GPT, when it works well, is a "magical tool" and is eager to explore additional AI enhancements, such as automated article summarization.
The author encourages readers to engage with the content by clapping for the article and considering subscription to their RSS reader, Echo, which is available on both iOS and Android platforms.

Photo by Emiliano Vittoriosi on Unsplash

The Biggest Problem Behind GPT Has Been Solved

So this is a little late because I didn’t know if I was going to write anything about it. Because I was developing my RSS reader Echo which did not have any GPT integration planned. I thought about having it summarise articles for you but that seemed unnecessary. But then I went back to my app Litany which does use the GPT API. And I knew I had to talk about this.

I am of course talking about this press release

Introducing ChatGPT and Whisper APIs

Jeff Belgum, Jake Berdine, Trevor Cai, Alexander Carney, Brooke Chan, Che Chang, Derek Chen, Ruby Chen, Aidan Clark…

openai.com

It talks about how ChatGPT has an API now. This isn’t too exciting by itself. GPT-3 can do almost everything ChatGPT can do. But there is something else in this post that I’d like to highlight.

90% Cost Reduction

So the new model OpenAI is rolling out is called ‘gpt-3.5-turbo’ and it’s priced at 0.2 cents per 1000 tokens (about 750 words). This represents a 90% decrease in cost from the previous best model ‘davinci’ which costs 2 cents per 1000 tokens.

Integrating ChatGPT Into My App Litany

I wrote an app called Litany. It’s a language learning app similar to Anki. In fact the basic idea of the app came from…

medium.com

And I think this is a really great opportunity for developers to get into the AI game. The biggest problem with GPT previously was the cost: 2 cents per 1000 tokens. And it’s not too uncommon for a single query to use 500 tokens. So you could end up spending one cent per API request.

So a 90% reduction is a very welcome change. It’s not going to set the world on fire, it’s definitely still more than using a database, but it’s not that unreasonable. I’d say it moved from ‘absurdly overpriced’ to just ‘extremely expensive’.

I wonder if some of the rumoured competitors to GPT are bringing down the cost. It seems like everyone wants their own GPT now.

Now Is The Time To Get Started

GPT, when it works, is a magical tool. So I’m integrating a tool into my app Litany which allows you to ask it about various synonyms. I’d often confuse myself with various synonyms during reviews so I built a way to ask GPT about them. And this is how it turned out:

Very impressive. Although it’s not perfect yet. Sometimes it doesn’t pick the best synonyms. Although if there is a very obvious synonym it’ll find it. And it doesn’t really understand context yet. So a single word can have multiple meanings and it’s not great at finding them.

But for simple things it’s really great. And it’s really useful.

And this is in addition to the other use of AI: breaking down phrases:

This is really impressive too.

This is all done with GPT-3. I plan to move to GPT-3.5-Turbo soon, but not right now. When you create an account you actually get some credit for free and I still haven’t exhausted the free credits yet. But soon I’ll look into it.

Conclusion

It’s really easy to integrate GPT into anything learning related. Other things, not so much. But now I am incredibly tempted to try integrating AI into more things. Maybe I’ll experiment with that article summarizer. It can be quite useful to have an article summarised for you.

If you liked this article be sure to give it a few claps. It helps out a lot with the algorithm. Also consider subscribing, I made an RSS reader that makes it very easy to do. It’s available on iOS and Android.