GPT-4 Turbo KeyNote

1. long context: context length 128k, more accurate over a long context
2. control
   • JSON Mode
   • Function calling, call many functions at once
   • Reproducible output with a seed parameter
   • log probes
3. world knowledge
   • retrivial over documents or databases
   • knowlege cut-off: 2023-08
4. more modelities
   • DALLE-3
   • GPT-4-turbo-vision
   • TTS