Does anyone know why instruction precision degrades over time?

GPT builders
6

I believe Custom GPTs currently use 4o model… It has a context window of 128,000 tokens so once you use more than that there is an issue with the conversation’s attention (ie what it can read in one request). The pricing page says that non pro users have only 32k context window so expect this might not be the full story.

The image displays a pricing plan comparison for Free, Plus, Pro, Team, and Enterprise options, highlighting "context window" sizes of 8K, 32K, and 128K respectively. (Captioned by AI)
The image displays a pricing plan comparison for Free, Plus, Pro, Team, and Enterprise options, highlighting "context window" sizes of 8K, 32K, and 128K respectively. (Captioned by AI)2060×252 11.8 KB

What models are my custom GPTs using? - #15 by _j.

Here are the stats for 4o
https://platform.openai.com/docs/models/gpt-4o.

When models change for GPTs they start to break/produce different results… This may also happen if there is a different dated version of the same model. You can’t choose or see this version with GPTs as far as I know.

The image shows a snapshot selection interface for GPT-4o, listing available versions with their dates, where the 2024-08-06 snapshot is highlighted. (Captioned by AI)
The image shows a snapshot selection interface for GPT-4o, listing available versions with their dates, where the 2024-08-06 snapshot is highlighted. (Captioned by AI)1390×311 30 KB

There is a new model 4.1 which has 1,000,000 context window however it is not yet used by GPTs

1 Like