Watson was a marketing exercise designed to sell a bunch of disconnected text and image processing libraries pulled together by consulting services. It did not function as advertised.
At one point we worked with a large energy company that was basically sold something LLM-like (large-scale indexing and searching/querying of documents) in 2016 or so. IBM had a team of 90 people doing full-time data ingestion for something like 26,000 documents. We got asked to do a counter-product in two weeks, which was literally just a TF-IDF search and some smarts around ingesting different types of documents. Both solutions performed approximately equally, except one cost something in the order of $185m and one cost $40k. Watson continued running for about a year until an external data science contractor realised they could query Watson for highly confidential board meeting notes, and it would provide full previews into the documents. The project was shuttered shortly after.
At one point we worked with a large energy company that was basically sold something LLM-like (large-scale indexing and searching/querying of documents) in 2016 or so. IBM had a team of 90 people doing full-time data ingestion for something like 26,000 documents. We got asked to do a counter-product in two weeks, which was literally just a TF-IDF search and some smarts around ingesting different types of documents. Both solutions performed approximately equally, except one cost something in the order of $185m and one cost $40k. Watson continued running for about a year until an external data science contractor realised they could query Watson for highly confidential board meeting notes, and it would provide full previews into the documents. The project was shuttered shortly after.
Alas, nobody gets fired for hiring IBM.