A few weeks ago I wrote about DALL-E 2, a a new machine learning model by OpenAI that can create images from a scene written in words. In my blog post, I explained how DALL-E 2 introduces a form of generative AI, enabling users to create images from scratch or to alter existing ones.
Last week, OpenAI introduced ChatGpT, a machine learning model that interacts in a conversational way, based on Open AI’s GPT-3 language model.
You can enter a question and OpenAI’s chatbot will come back with a (detailed) response (you can access ChatGtP here). For example:
I enter my question about fixing my leaking kitchen tap and ChatGpT starts generating the answer to my question. Within a few seconds I’ve got a pretty comprehensive answer to my question:
I’m curious to see what the response is to my question about a new talk on systems thinking that I’m working on:*
* Yes, I know it feels lazy to ask a machine-based language model for input into a presentation. However, I can imagine that this will quickly become a real use case for presenters and students the world over …
Again, the response was generated in mere seconds and provides a comprehensive structure to a generic talk about systems thinking. Arguably, the points returned could easily apply to a presentation of any kind, with the exception of points 3 and 4.
Finally, I want to know about today’s weather forecast:
This is where I learn about some of limitations of the current version of Chat GpT:
Google doesn’t have to worry yet about competition from ChatGpT and applications built on top of it. (If you ask 1littlecoder, however, ChatGpT already poses a threat to Google, making it obsolete for programmers). There are some other limitations — built-in — the current version of ChatGpT:
- Beliefs or opinions — The model refuses to answer questions about opinions or beliefs.
- People, topics or current affairs — The model won’t provide answers to questions related to people or current affairs (see Qatar World Cup example below).
- Small prompt changes — Small changes in the prompts created by humans might cause big changes in ChatGpT answers. Changes in input by a ‘human instructor’ can have a big impact; depending on the input, it may not answer a question, answer it incorrectly, or answer it correctly.
By the way, Open AI is very open about the current limitations to ChatGpT:
Finally, these are some of the many use cases and opportunity areas where I can see ChatGpT fulling an important role:
- Generate content and copy — This is already happening with products such as Copy.ai and Jasper.ai.
- Write software — Put in code commands or questions and ChatGpT will generate the code. Stack Overflow has already banned AI generated answers on its Q&A platform.
- Provide customer support — Thinking of the bots I worked on at Intercom, applying ChatGpT will significantly reduce the workload currently placed on humans to train the bots that underpin self-serve customer experiences.
- Improve productivity — Whether it’s about generating quick answers (think tutoring!), plans or repetitive tasks, I can see numerous productivity apps emerging from ChatGpT.
- Create recipes — Will we see the end of cookery books?!
Main learning point: Wow, this is exciting! And scary at the same time. I know it’s early days and that there are all kinds of limitations to ChatGpT technology, but it’s easy to see its transformative impact of ChatGPT on a whole host of use cases and business models. Generative AI is here to stay!
Related links for further learning: