Ever since Turing, the U in NLU was indeed ‘understanding’. Not sentience.
This is important. For more than 50 years NLU (natural language understanding) AI researchers have been trying to crack half-decent NLU like that sported by the disembodied AI, Hal, in Kubrick’s 2001: A Space Odyssey.
With LLMs (large language models) and GPT-4 we’ve pretty much got there.
It’s shocked us all how quick it’s come. And we know it’s got some ways to go.
But some are virtually insisting on consciousness in their testing of ‘understanding’.
However, we are the NLU researchers, and we define what we mean by the ‘understanding’!
Not psychologists and linguists and non-NLU AI resesearchers!
GPT naysayers
We have some very loud, over-the-top GPT sceptics, academics and ethicists that have IMO completely misunderstood the Turing-ian and Dartmouth-ian quest for NLU. (The latter is a reference to the 1956 workshop that essentially established the field of AI).
These GPT naysayers and whistle-blowers alike are coming from all angles, and they’re all arguably wrong and debunk-able, but it can be subtle:
- Claims the understanding in LLMs is ‘not real’ misses the point. To NLU researchers ‘understanding’ was…