1). Avoid all the nannying moderation,

2). Set persistent roles for user and AI,

3). Control determinism,

4). Use 8,192 tokens by default,

5). Use 32,768 tokens by special request,

6). Use Whisper

7). Top_p nucleus sampling

8). Number n of competitions

9). Stream on/off

10). See logprobs

11). Best of

12). Legit bias

13). Set max tokens

14). Track / report tokens

Lots of things.