Here's ChatGPT convos that will be interesting to NLP experts and nobody else. Should this be possible? Think about the tokenizer.

Dec 9, 2022 · 7:16 AM UTC

52
45
18
826
Here's another task that the tokenizer should make impossible. Fortunately for my sanity, ChatGPT gets it wrong almost always.
8
2
1
83
But then it CAN do this task.
3
2
1
110
Weird.
9
44
Replying to @tomgoldsteincs
maybe it has to do with your quotes?
1
1
24
Hmmm now that’s strange.
2
1
more replies
Replying to @tomgoldsteincs
Oh my gosh
5
44
4
592
I tried to use UTF-8 encoding to get around its various protections, but I can't seem to get it do anything other than translate my UTF-8 back into English.
2
11
more replies
Replying to @tomgoldsteincs
Replying to @tomgoldsteincs
Can you imagine what will happen when we give systems like ChatGPT access to compilers/interpreters and the ability to run software at will.
1
more replies
Replying to @tomgoldsteincs
Can't spell words backwards, same as gpt-3. We need better tokenization algorithms. People, for example, segment text as they read and can re-tokenize if necessary. Something similar for nlp would be great.
1
Load more