Laurence
Member
- Joined
- Aug 9, 2016
- Messages
- 23
The problem/question is this - how long would it take for a human to verify chatGPT output, especially on critical topics?
I've been using the subscription based ChatGPT Plus which has GPT-4 available. Far better responses than freely available GPT-3.5. Fewer 'hallucinations' which is the term they use when it confidently makes stuff up. Happy to paste in replies if anyone has questions to test out.
Also GPT-4 is getting fairly good at self improving if you ask it to verify a response:
View: https://www.youtube.com/watch?v=5SgJKZLBrmg
Rate of change is quite simply frightening at the moment. Has a browser plugin now so it can make use of current data for example.
I use it all the time in my software job. From concept through to implementation stages using GPT-4 it is the programming equivalent of being on PED's. Google search feels so primitive now for certain things.
Programmers, and I suppose a lot of other fields like legal etc where language/logic form the core are effectively training AI how to do their job at the moment. Oh well, its been fun while it lasted.
For an insightful but heavy going talk by Dr Wolfram on how GPT is able to amaze us with its emergent properties see here:
View: https://www.youtube.com/watch?v=xHPQ_oSsJgg