Set up a monthly contribution to the UNHCR
- Adding a bit on beauty for diversity injection
- Start on the GPT and Embedding interfaces. Prompt the GPT with something like “Once upon a time there was” and set the number of times to run and the number of tokens. Split on sentences (r”\.|!|?”) and get the embeddings for each. Then cluster and extract topics (Using EmbeddingExplorer pointing at a different db). Build maps!
- Continue fleshing out the Twitter embedding app
- Ok, what I really wound up doing was getting threading to work on TweetDownloader and fixing an interesting bug in the sampled day method. When I wrote it, I assumed that the number of tweets per day are reasonably constant. Not true. So as a bit of a hack, I moved the endpoint of the query to include the entire day and use REPLACE INTO rather than INSERT. Much better results so far. Will work on the other stuff tomorrow.
- Need to read this carefully. I like the fact that it uses the MinGPT: Transformers are Sample Efficient World Models
- Deep reinforcement learning agents are notoriously sample inefficient, which considerably limits their application to real-world problems. Recently, many model-based methods have been designed to address this issue, with learning in the imagination of a world model being one of the most prominent approaches. However, while virtually unlimited interaction with a simulated environment sounds appealing, the world model has to be accurate over extended periods of time. Motivated by the success of Transformers in sequence modeling tasks, we introduce IRIS, a data-efficient agent that learns in a world model composed of a discrete autoencoder and an autoregressive Transformer. With the equivalent of only two hours of gameplay in the Atari 100k benchmark, IRIS achieves a mean human normalized score of 1.046, and outperforms humans on 10 out of 26 games. Our approach sets a new state of the art for methods without lookahead search, and even surpasses MuZero. To foster future research on Transformers and world models for sample-efficient reinforcement learning, we release our codebase at this https URL.
- Delivered the quarterly report.