MyAdversary

Inspiration

responses from LLMs tends to be overly stochastic, hallucinatory, and surface (lacking depth).

What it does

challenges the response given and scores it for consistency

How we built it

prompt engineering in POE and bot agent function chaining.

Challenges we ran into

retries and context optimizations

Accomplishments that we're proud of

works with mixtures of experts

What we learned

importance of distillation

What's next for MyAdversary

model training, fine tuning, more models

Built With

• poe