Quote:
Originally Posted by cerulean
LLMs were trained on tightly controlled datasets by highly intelligent people, some of which are empathetic. That's why they lagged behind recent information, and were great to work with initially.
LLMs now are trained on rapid accumulation of data found on social media sites and through interactions with its users. Most people have decided to abandon empathy.
Overtime this likely spoiled the tight controls for the sake of keeping the datasets fresh with new information. No one wants to use an LLM that is 3 years behind on data.
|
Great so in the future AI will be just like a stressed-out female Tik Tok user. Wonderful.
Quote:
Originally Posted by mainstreammix
When you say AI which one? Claude is smarter than ever since it incorporated deep web searches.
|
Chatgpt. I use the pro version and the more I work with it on a project unless it hits right away and I ask for variations or refinements it gets worse and worse. So if I take a break, clear cache, etc it helps. But AI is fickle. For coding it's great but there are still issues with hallucinations (six fingers etc) and other miscues.
But hey, in two years it will be almost perfect. AI is still in its infancy.