Can a robot learn what its own body looks like — just by watching itself move? W/Neural Jacobian Fields (NJF), CSAIL researchers show that it can. Using just one external camera, NJF enables robots to infer the structure of their joints & limbs — no built-in sensors or prior
MACHINE LEARNING
-
Why AI Advances Haven’t Improved Autocorrect Technology
By
–
Why haven’t advances in AI been applied to Autocorrect?
-
Preference for Scalable Turn-the-Crank Algorithms with Increased Compute
By
–
I sometimes try to explain it as a statement of preference for "turn the crank" algorithms. When you're eventually given more compute (faster crank), you shouldn't have to touch anything at all, you just crank faster to make better. You can (and probably locally should) knowingly
-
Can AI models self-improve recursively via agentic methods?
By
–
Could AI models start self-improving recursively, scaling up performance to the moon in an infinite loop? A friend said that it's impossible, I wasn't that sure. I argued that with agentic methods like Absolute Zero (basically an LLM generates training data by itself by
-
Position Bias in LLMs: How Architecture and Training Data Matter
By
–
LLMs sometimes exhibit position bias, where they overemphasize the beginning or end of a document or conversation while neglecting the middle. A theoretical framework from MIT revealed that model architectures & training data contribute to this problem: http://
bit.ly/44mW9V8 -
PyTorch Community Invitation: Join TorchTitan Channel Slack
By
–
impressive that no one has yet, but I've invited you to the pytorch slack, after you join you can hang out on the torchtitan channel and get productive much quicker together
-
SigLIP 2: DeepMind’s New Multilingual Vision-Language Model
By
–
SigLIP 2: DeepMind’s Multilingual Vision-Language Model Google DeepMind just dropped SigLIP 2! These new multilingual vision-language encoders are a game-changer for AI. Better understanding, localization & fairness. Get the quick rundown on our blog:
-
Kimi-Dev-72B shows impressive SWE-Bench Verified performance, no tech report yet
By
–
Why isn't anyone talking about Kimi-Dev-72B, released ~1 week ago on the Hub?
The tech report is not out yet, and I'll love to see more benchmarks, but performance on SWE-Bench Verified seems very impressive! -
Transformers Prefer Simpler Hypotheses In-Context Occam’s Razor
By
–
In-Context Occam's Razor: How Transformers Prefer Simpler Hypotheses on the Fly Deora et al.: https://
arxiv.org/abs/2506.19351 #ArtificialIntelligence #DeepLearning #MachineLearning -
KerasHub enables Hugging Face checkpoints across multiple frameworks
By
–
KerasHub lets you use any Hugging Face checkpoint for all top models like Llama, Gemma, Mistral, etc… Run your workflows in JAX, PyTorch, TensorFlow – inference, LoRA fine-tuning, large-scale training from scratch Blog post: https://
developers.googleblog.com/en/load-model-
weights-from-safetensors-into-kerashub-multi-framework-machine-learning/
…