How do we make LLM understand the context of time

For example, if I ask LLM “What is the forecast for this year”. How do I make it understand that I am talking about 2023. This may be for last quarter, last month etc

You can finetune gpt-3.5 now. Maybe that will help.

1 Like

then i will need to fine-tune every month, quarter, year

I would just send the current date in the prompt somewhere…ie “Today’s date is 1/1/23…” or something.

1 Like

I see system prompts with current date appended at the end. Maybe that will help the AI to know the context of ‘this year’, ‘today’, ‘yesterday’, etc.

3 Likes

I think your answer exemplifies Schillaces Law #3

Code is for syntax and process; models are for semantics and intent.

Getting the date/time is very easy in Python. At least, I assume it is, I’d have to ask ChatGPT how, but I’ll bet it’s one package and one line of code.

Meanwhile, how does an LLM determine the date? You can figure out something very sophisticated, or just pass in the current date!

1 Like

I just solved it by adding in the prompt "For any questions related to date, remember todays date is {date} and passed the current date

3 Likes