Playgent Labs

Reinforcement learning environments for real-world work

[Playgent Labs @playgent_labs]

We've open sourced our test set and RL harness!

🤗 https://t.co/9a2NSgwSOe
RL harness: https://t.co/2Q8ETVHeDr

Can you build the next IB agent? 👀

[Playgent Labs @playgent_labs]

🤔 How do frontier models perform on core investment banking tasks?

Introducing Playgent-IB-Bench, an investment banking benchmark we built with former M&A bankers from bulge-bracket and elite-boutique firms.

Writeup: https://t.co/hQGONrOhht

[Playgent Labs @playgent_labs]

We then built an RL harness on @PrimeIntellect that mimics the human setup. Agents work in a filesystem with 58 tools across:
- working with spreadsheets
- creating presentations
- reading PDFs
- editing Word documents

The environment enables learning a policy for long-horizon financial tasks using the same tools as human bankers.