🧵Responses to adversarial queries can still remain latent in a safety-tuned model. Why are they revealed sometimes, but not others? And what are the mechanics of this latent misalignment? Does it matter *who* the user is? (1/n)
@LingdongH@realmgyong@golan@GoogleAI@TensorFlow Hi there - in TF.js 2.0, the backends no longer come bundled with tf-core. So you'll either need to also include a backend script tag (e.g. <script src="https://t.co/9pZOwE7tzV">), or use the union package instead of tf-core (e.g. <script src="https://t.co/lLsWKDfV2n">).
🎭 https://t.co/aSWzRRCaiC ✨ is a new website for writing and playing with indie face filters!
Try playing with the presets or live-editing your own filter in the language GLSL.
ML for the Web and Beyond 🚀
TensorFlow.js is the key to training and deploying ML models in browsers or anywhere #JavaScript can run. Get up to date on all the newest features including MobileBERT and FaceMesh.
Watch #TFDevSummit now → https://t.co/pIahCuslnA
📢 Blazeface is now available in browsers with TensorFlow.js!
The model detects faces and facial features in real-time 👨👩⏰
Try it out now → https://t.co/Km66lDh5R6
Understanding UMAP - a high-level introduction to how the algorithm works, how to use it effectively, and how it compares with t-SNE.
https://t.co/Dt9Ap9G2Mc
We just released PoseNet 2.0 with TensorFlow.js, including a much more accurate ResNet-based model!
Try the live demo by @tylerzhu3, @oveddan, @greenbeandou, @dsmilkov, @ire_alva, @nsthorat → https://t.co/Dgz0kay40U
Learn more here → https://t.co/JDs5wIeByP