4 ms·
That 200k context needs some proper testing. GPT-4-Turbo advertises 128k but the quality of output there goes down significantly after ~32k tokens.
by msp26 3y ago
That 200k context needs some proper testing. GPT-4-Turbo advertises 128k but the quality of output there goes down significantly after ~32k tokens.
- KaoruAoiShiho 3y agoRead the article, it's addressed with charts.
- msp26 3y agoI did but I want more independent testing than just QA performance by position.