My team at Google DeepMind is hiring a student researcher (NYC, 2027) to work on how we evaluate sign language AI. Deaf and hard of hearing researchers are especially encouraged to apply, as are fluent signers. Get in touch: forms.gle/Gwfmp1cKGQ8oN2…
Chris Dyer
@redpony
I live in London. 🔴⚪ I am a researcher at DeepMind and work on AI for sign languages, among other things. I play the cello.
1.1K Following 10K Followers 477 Posts
Latest posts 5
Amazing find. Living in the future. But I’m not sure I’m more surprised that people or machines didn’t find it sooner. Both are sort of surprising.
Hi all, I have a new result I'm excited to share. It turns out that *all existing implementations* of softmax and cross-entropy (that's JAX, PyTorch, textbooks, etc) have a subtle bug --- two bugs, actually. A simple rewrite fixes both, and eliminates "slingshots" when grokking:View quoted post on X
First-price auctions are simple and (provably) good, and this is good news. Impressively thorough results from @_onionesque @jacobeisenstein and friends
Check out our paper on whether first-price auctions can efficiently allocate queries among strategic AI agents (in a decentralized serving-time market):View quoted post on X
arxiv.org/abs/2610.20587
Single colab experiments still show us important things in 2026.
Wrote a little Colab for "Planning to Learn".View quoted post on X
Runs in ~3 mins on a free T4 GPU:
- Exact PG loses to cross-entropy, even on accuracy.
- The horizon loss beats both.
- Planning CE even beats CE on its own NLL metric.
Nothing hidden... Run it yourself!
colab.research.google.com/drive/1nuACbgy…
I'm so happy to be able to announce the first general purpose sign-language-to-text translation (SL2T) model from my team that's out today and powering a new ASL input feature on @Android phones (and don't worry, we're not stopping at ASL).
0:18