Post ATdAcRLQIz5dOD76jg by dfrancis@mstdn.social
(DIR) More posts by dfrancis@mstdn.social
(DIR) Post #ATcsnwikBvikiAMIJk by TedUnderwood@sigmoid.social
2023-03-15T02:46:47Z
0 likes, 0 repeats
Well, after spending about eight hours talking to #GPT4: my review is: smart, good at brainstorming, capable of understanding irony, but sooo basic. It wants everything to end with hugs and learning.
(DIR) Post #ATctkDrWcKWgffFp7A by paddy@sigmoid.social
2023-03-15T02:57:19Z
0 likes, 0 repeats
@TedUnderwood Have you tried out Stanford Alpaca? It's not nearly as smart, but also doesn't have "guardrails", which makes it more interesting in some cases.
(DIR) Post #ATctpPSquO7K6OBfMW by TedUnderwood@sigmoid.social
2023-03-15T02:58:16Z
0 likes, 0 repeats
@paddy Not yet. I also want to try Claude.
(DIR) Post #ATcvgzoaaXPs8xTD8K by TedUnderwood@sigmoid.social
2023-03-15T03:19:09Z
0 likes, 0 repeats
It really does confabulate less, though. Pretty accurate replies, including largely accurate bios for academics, with titles of books, etc.
(DIR) Post #ATcwEdz0GXDAHKNyyW by hackerfriendly@mas.to
2023-03-15T03:25:14Z
0 likes, 0 repeats
@TedUnderwood Try asking it which model it's running. It insists that it's "based on OpenAI's GPT-3, which was released in 2020." 🤣
(DIR) Post #ATcz3lIv0NBHf6CgqW by budgibson@me.dm
2023-03-15T03:56:52Z
0 likes, 0 repeats
@TedUnderwood Right now, you’re testing it against stuff you know and can verify, which is only natural. I’ll be interested to see where you think you can get it to do work for you (the model saves you in total task effort or could foreseeably do that).
(DIR) Post #ATd6X3bsireEVtzQIK by TedUnderwood@sigmoid.social
2023-03-15T05:20:36Z
0 likes, 0 repeats
@budgibson oh, definitely coding; I just used it for that tonight
(DIR) Post #ATdAcRLQIz5dOD76jg by dfrancis@mstdn.social
2023-03-15T06:06:22Z
0 likes, 0 repeats
@TedUnderwood It's trying! :)