First full day of Spring!

This post only focuses on prompt engineering for autoregressive language models, so nothing with Cloze tests, image generation or multimodality models. At its core, the goal of prompt engineering is about alignment and model steerability. Check my previous post on controllable text generation.

GPT Agents

• Sent the posters off to Staples
• Getting the embeddings for MAGA. Because I really don’t want to spend over $100 at OpenAI this month, I’m going to stop there and work on clustering, then export the db to the laptop
• Things are taking too long to get the clustering working. Wrote a procedure to take random tweets within a range that can be used for the clustering. The trick here is that all the values will need to be updated once the values are found:

DELIMITER $$
CREATE OR REPLACE PROCEDURE get_random_tweets(min_query_id INT, max_query_id INT, num_vals INT)
BEGIN
    SELECT * FROM table_tweet AS t1
    JOIN (
        SELECT id FROM table_tweet
        WHERE query_id >= min_query_id AND query_id <= max_query_id
        ORDER BY RAND()
        LIMIT num_vals
    ) as t2 ON t1.id=t2.id;
END $$
DELIMITER ;

SBIRs

• 9:15 Standup
• 9:30 tuning discussions with Rukan
• 10:30 meeting with Lauren and Aaron
• 1:00 MDA meeting