Interestingly when I attach gpt-3.5-turbo-1106 and prompt the assistant how much context window it has to work with and which model its using it responds with 4000 tokens and GPT 3. Confused because 3.5-turbo-1106 has a 16K token context window and isn’t 3.5…
When I update it to use gpt 4 preview the supposed 128K context window and ask it the same question it says its using GPT 4 with 8K context window.
No sure whats real or fake here now.
You get a recycled answer.
You know from the announcement and reading model details that the pretrained knowledge of the newest model extends to April 2023, but if you don’t tell the AI it is that model, and the AI also doesn’t know the newest advances in OpenAI models, it will have a hard time answering.
If you want to waste system instructions on simply pasting information about models and prices from OpenAI pages, you can get an AI that can answer.
