🚨 New Library Alert! We just released SafeRL-Lite — a lightweight Python toolkit for safe + explainable reinforcement learning. ✅ Enforce constraints 🔍 SHAP & saliency maps 📦 PyPI: https://t.co/MVIkq5NAS7 🔗 https://t.co/ThOjx1dCI5 #RL #XAI #Python #AI4Safety #satms

Looking forward to serving India with @Starlink!

HANOI: A powerful reminder of how much potential Vietnam’s young generation holds in AI. Yesterday, I had the opportunity to speak as an AI expert on behalf of Quanskill at the seminar “Awakening the Potential of Data Science & AI” at Hanoi National University of Education.

When Reality Is Not Enough: Sim2Real, Domain Randomization & Causal World Models for Autonomous Driving https://t.co/BbDxKMAXmv https://t.co/yXyV4ypbpW

ज़िंदगी ने बहुत कुछ सिखाया है, कभी ग़म, कभी ख़ुशी, सब कुछ मौसम सा आया है। दिल टूटा है कई बार, मगर हर बार वही एक चेहरा याद आया है। अब क्या ही उन्हें दोष दें, गलती हमारी ही थी, कि हर टूटने के बाद भी हमने दिल को दिल से ज़ोरों से लगाया है। 🖤 - सत्यम कुसुम मिश्रा #satmis #poetry

Sim2Real Mastery: Teaching the Car to Survive the Real World The Day the Car Left Its Playground https://t.co/0D15GvkhdY https://t.co/jOvn8FPKjV

19 वर्ष के देवव्रत महेश रेखे जी ने जो उपलब्धि हासिल की है, वो जानकर मन प्रफुल्लित हो गया है। उनकी ये सफलता हमारी आने वाली पीढ़ियों की प्रेरणा बनने वाली है। भारतीय संस्कृति में आस्था रखने वाले हर एक व्यक्ति को ये जानकर अच्छा लगेगा कि श्री देवव्रत ने शुक्ल यजुर्वेद की माध्यन्दिन

What 19 year old Vedamurti Devavrat Mahesh Rekhe has done will be remembered by the coming generations! Every person passionate about Indian culture is proud of him for completing the Dandakrama Parayanam, consisting of 2000 mantras of the Shukla Yajurveda’s Madhyandini branch,

Broken Prayers of a Tired Heart Now available on Google play books for all my vn friends who were complaining they can't access Kindle... https://t.co/DKVezm7e8D #satmis

Can Your Model Learn This Function? A Visual and Mathematical Dive into the VC Dimension: the Hidden Geometry of Learnability https://t.co/amw5Wq0i0e https://t.co/fJiML1Lnik

ENCODER–DECODER LLMS ✓ Encoder–Decoder models (also called sequence-to-sequence models) process input text in two stages: the encoder understands the input, and the decoder generates meaningful output based on that understanding. WHAT ENCODER–DECODER MODELS ARE → Two-part

Broken Prayers of a Tired Heart Now available on Google play books for all my vn friends who were complaining they can't access Kindle... https://t.co/DKVezm7e8D #satmis

Lesen Sie Broken Prayers of a Tired Heart auf Leanpub! https://t.co/cxlyDC4X44 cc @leanpub

Lesen Sie Broken Prayers of a Tired Heart auf Leanpub! https://t.co/cxlyDC4X44 cc @leanpub

One Brain to Drive Them All: End-to-End Differentiable Autonomous Driving https://t.co/TQdtTo0W95 https://t.co/gTcVBmRPL5

This paper shows how to beat models with 671 billion parameters using just 7 million parameters on hard puzzle tasks. On the ARC-AGI reasoning benchmark, this tiny model achieves 45% accuracy compared to 15.8% for DeepSeek R1 and 34.5% for OpenAI's o3-mini. On extremely difficult

Driving With Imagination: DreamerV3 + TD-MPC in CARLA Before the Car Ever Moves: The Secret Training Ground https://t.co/FoSBdBYyDF https://t.co/jgS8T4q4MR

🚨 Stop whatever you're doing. A new study just found that advanced LLMs already think they're more rational than humans and they change their strategy based on who they believe they’re competing against. Yes, you read that right. Researchers built something called the AI

Wow. New paper from top Chinese labs, build a single multi-agent system that solves Olympiad-level science across domains. SciAgent solves hard math, physics, and chemistry Olympiad problems like IMO, IMC, IPhO, and CPhO at gold medal level. Reaches human gold medal A top

so we know what optimizers are, right? optimizers are algorithms that update model parameters during model training with the goal of minimizing a specified loss function but how do you create your own optimizer? i mean there already exists so many like adam, adamw, sgd,

Spatial intelligence has long been one of the biggest bottleneck for VLMs. Two years ago in Sept 2023, when I just started my postdoc, I still remember vividly how we are excited about GPT-4V and how our “What GPT-4V still can’t do” slides were completely dominated by geometric