News for hackers. Stuff that matters.

A comment on T0* – Series of encoder-decoder models trained on a large set of different tasks. Parent comment

Re:T0* – Series of encoder-decoder models trained on a large... by jstx1 on Monday October 18, 2021 @11:14AM (#28906249)
It's an expensive autocomplete, not an information retrieval system or a question-answering machine.
[ Parent | on HN ]
Re:T0* – Series of encoder-decoder models trained on a large... by tiborsaas on Monday October 18, 2021 @11:35AM (#28906551)
What's the difference? Answering a question can be considered "autocomplete".
[ Parent | on HN ]
Re:T0* – Series of encoder-decoder models trained on a large... by RyEgswuCsn on Monday October 18, 2021 @11:43AM (#28906671)
Parent surely meant spellcheck autocompletion.
[ Parent | on HN ]
Re:T0* – Series of encoder-decoder models trained on a large... by tiborsaas on Monday October 18, 2021 @12:58PM (#28907725)
I doubt it, that's clearly exceeded by these language models. Calling it just an autocomplete - because it can mean a lot of things people are familiar with - is a way to downplay their significance.
[ Parent | on HN ]
Re:T0* – Series of encoder-decoder models trained on a large... by 6gvONxR4sf7o on Monday October 18, 2021 @12:07PM (#28907038)
Not true. Take a look at the paper and benchmarks. The point of the thing is that it does well on a number of NLP tasks by being an expensive autocomplete. As people demonstrate in this thread, it still has significant flaws.
[ Parent | on HN ]