Posts by ythos@dotnet.social
 (DIR) Post #ASzg8GoEktH7tMYSG0 by ythos@dotnet.social
       2023-02-18T12:54:22Z
       
       0 likes, 0 repeats
       
       @sgirlprivacy I would like to stand up for Clippy here. Clippy was programmed to try to detect particular situations and offer carefully determined assistance. Clippy was limited (and people poked fun as a result) but useful.ChatGPT strings together words in statistically plausible fashion and is as likely to be randomly harmful as useful. In fact the likelihood of harm is probably greater than the likelihood of usefulness, from what we've seen.COME BACK CLIPPY. WE LOVE U.
       
 (DIR) Post #Abf4hhrIw5QPdRCgi0 by ythos@dotnet.social
       2023-11-10T11:06:13Z
       
       0 likes, 0 repeats
       
       @tante And yet, the opportunities they offer for learning are amazing.For example - LLMs are really good content summarizers - if you write an abstract and get the LLM to boil it down to a couple of paragraphs, and it *doesn't* produce what you'd expect, your abstract is probably not emphasizing the things you thought it was, or it lacks clarity or contains contradictions.You can use the fact that LLMs are based on statistical consensus to explore that consensus.