Maybe it's good to play games because it's a way to get a verifiable reward signal for one's thoughts.
19.03.2025 18:21 — 👍 0 🔁 0 💬 0 📌 0Maybe it's good to play games because it's a way to get a verifiable reward signal for one's thoughts.
19.03.2025 18:21 — 👍 0 🔁 0 💬 0 📌 0
100% agree on the fly UIs is the future. It will still take a bit for our models to be good enough at generating dense data (i.e. not text) but it will happen.
Is tricky to imagine and reason about the downstream effects, wish I could get a glimpse of the future!
I just learned that Torch ctc_loss calculates the wrong gradient (but when there was log_softmax before, it does not matter).
For the grad ctc_loss w.r.t. log_probs, it calculates exp(log_probs) - y, but correct would be -y. Some workaround: github.com/pytorch/pyto...
PS: First Bluesky post.
Observing the responses here and on twitter made me reflect, realize and act. There is a real difference, and I describe it (as well as some of the non-differences) here:
gist.github.com/yoavg/9142e5...