Conversational AI based on LLM’s, of which ChatGPT is the most prominent example, are the new hotness. There is Cambrian explosion of startups and products announced around either there own AI models or via apis to others models.
I will leave it to others to become the burgess shale of AI startups.
I expect general services based on “my set of data” as stored in a folder, or in a collection of documents like Microsoft OneDrive, Google Drive, Dropbox, Apple Cloud and others will be released soon to allow natural language queries of those data sets. This could be a new kind of conversational excel for the business people. Probably with just as many duplicate and bad ad hoc applications (needing a new generation of developers to fix?). I hope these queries will be repeatable, explainable and debuggable.
But what i really want are tools that allow summarization, validation, and explanation services.
- summarization or “TL;DR as a service” (Too Long; Didn’t Read) of an article, a book, or the current understanding of a question.
- Validating content, showing the truthfulness or falseness to varying degrees of depth.
- Possibly the opposite of summarization, an “Explain it to me like I am 5” as a service.
All with links to sources, and resources.
And all of these above with the options to do so in the written or spoken language of my choice, and the option to visualize different parts.
“TL;DR” as a service and “Explain it to me like I am 5” as a service could exist as a not just as a more open tool (point it at a url or pdf), but could also additional services build into sites (as a additional paid service?) for example a New York Times or Wikipedia run service for the content on their site.
It should go with saying that Privacy and Trust will be important in and of these services.
This is the sort of thing I’m talking about : What if ChatGPT was trained on decades of financial news and data? BloombergGPT aims to be a domain-specific AI for business news.
The article closes with the speculation :
But I believe there will be, in the longer term, openings for smaller publishers here, especially those with large digitized archives. Imagine the Anytown Gazette training an AI on 100 years of its newspaper archives, plus a massive collection of city/county/state documents and whatever other sources of local data it can get its hands on.
I wonder if the Wall Street Journal chatbot is in the works? If not a ChatWSJ, then expect all of Rupert Murdoch’s media properties to do moral panic hit pieces very soon.