Speech without comprehension

Where can I be snide if not my own blog? So, the AI rage these days is about the wonderful language mimicry of ChatGPT and OpenAI. What a deep understanding of sentence structure provides, brought to you by the likes of Chomsky and other dolts north of the Charles, is that now computers can produce fine sentences and paragraphs with no understanding; except that the form of the question determines the form of the answer - with tidbits of internet-scraped content inserted in the appropriate blank spaces in the sentence structure.

Beautifully articulated prose with no understanding is what you get from Chomsky himself. Imagine believing that the transistor explains the software written on a computer! Or that internet protocols are all that is needed to understand internet content! Or that the meaning of "red" could somehow be embedded in grammar and syntax. Anyway, Chomsky's stupidity is now leveraged on the world and engineers are frantically trying to figure out how "large language models" are supposed to do things like order a pizza. They cannot. 

But the old "assistants" like Siri and Alexa work fine using keywords. Software assistants use keywords. My approach has always been to do better keyword matching, to include narrative pattern matching. 

Anyway, something genuinely ironic is beginning to emerge in the latest ChatGPT developments. They are starting to market "structured commands" as a kind of input template "to save the user having to re-type the same things over and over". I think that is not really what is going on. It seems obvious that the attempts to make ChatGPT work are going to be introducing keywords. Some poor bloke thinks all they need to do is structure the command, let ChatGPT do something with that input, then write a program that acts on the basis of what ChatGPT produces as output. The irony is that: it will not take them long before realizing that ChatGPT did not add and value and can be eliminated as an unnecessary intermediary between structured commands and structured output. 

In the end, Chomsky's work will have the lasting value of making computers seem literate. But for language understanding, there has been no progress from that direction. 

I am proud I went to BU: My university was south of the Charles River. We were phenomenologist in Boston, while they were logical positivists in Cambridge. It explains why they are still trying to write programs to discover reality over there.