I work with language models day to day and got tired of trusting benchmark numbers that never matched what I saw in practice. This site is where I write down what I tried, what worked and, more often, what did not.
No newsletter, no comments, no tracking.