TLDR Web Dev · MISCELLANEOUS
TRAINING IS NOT THE SAME AS CHATTING: CHATGPT AND OTHER LLMS DON'T REMEMBER EVERYTHING YOU SAY
Summary from the TLDR Web Dev newsletter
A common misconception about LLMs like ChatGPT is that they learn and memorize everything said to them during conversations. This is inaccurate as these models function as stateless functions, processing each conversation independently without retaining memory of previous interactions. Features like ChatGPT's "memory" are actually prompting tricks that insert relevant information into the context of ongoing conversations.