My PhD project is to work on language model capable of playing the answerer in a Twenty Questions game (reverse of Akinator). If you are interested you can play here: https://twentle.com It is meant as a party game to play on your mobile with the learderboard on a large screen (like Quiplash)
On the back-end is a GPT-3 model answering the questions with: never, rarely, sometimes, always or usually.
It's killing me that if I run out of time, I never get to find out what the word was. could you tell me what word three was for https://www.twentle.com/p/2358 ?
Hi, first author here. Yes, we evaluated 3 retrieval-based models, DPR, RAG and FID - check out the paper for the numbers (https://arxiv.org/pdf/2008.02637.pdf)
Thanks for your suggestions. It is updated every day at market close. Will add a dropdown to select between YTD/MTD/1Y.
Would you be interested in receiving a weekly update to your inbox?
Thanks, I will check it out.
There are a few python wrappers around Bloomberg APIs. For example: https://github.com/kyuni22/pybbg. It mimicks the Bloomberg Excel formulas (BDP, BDH, BDS). Quite powerfull. But you can hit the daily limit easily (500,000 data points per day max).
They released the Bloomberg Query Language (BQL) that you can use to make larger queries. But not all FLDS fields are available yet...
They also provide BQNT, a jupyter notebook where you can run BQL queries and share "apps" with your colleagues.