Expert profile
Founder & AI Product Leader
Yash Thakker is a Generative AI expert with over 12 years of experience in product leadership and technical strategy. As the founder of explainx.ai, he has taught over 300,000 learners and built AI platforms serving millions of users globally. He specializes in Agentic AI, Multimodal RAG, and the intersection of LLMs with consumer hardware.
ORA lets open-weight DeepSeek-V4.1-Flash operate robots with no VLA: 69.2% RoboTwin, 81.4% LIBERO-Pro. What is real and what is not.
LeanGraphSearch uses Mathlib's dependency graph to retrieve Lean 4 premises, with up to 12% better recall and open code. How it works, limits.
LIBERO-PeRM tests personalized robot manipulation. pi0.5 scores 97.6% on tasks but about 24% on motion and skill preferences. Data and code are MIT-licensed.
Magnitude (Apache-2.0) claims 2.19x faster decode than Ollama on M4 Pro and 2.17x on DGX Spark. Setup, agent support, benchmark caveats and what to test.
MAI-Gym from Alibaba Tongyi-MAI unifies GUI agent evaluation and RL training. What is released, the 9B results, and what is still unverified.
At TechCrunch Disrupt, Mark Wahlberg said AI should enhance films, not write them, and he is open to an AI likeness carrying on his legacy. Context.
Mercor's benchmark map covers 924 of 18,797 O*NET tasks. Why 90% of US working hours go unmeasured by AI benchmarks, and the study's caveats.
Meta's commercial terms, effective Oct 30, let it crawl sites tied to business accounts and train AI on them. What is reported, what Meta denies, what to do.
MLExpert is an open Apache-2.0 35B MoE agent that lifts MLE-Bench Lite medals from 37.9% to 72.7%. How it works and how to run it.
Nemotron 3.5 Bolt 74B-A9B: Star Elastic compression, 2-bit experts, LiLiCorr drafter, 84 tok/s on RTX Spark vs Qwen3.6 and Gemma 4. Checkpoints and caveats.
NVIDIA EdgeWise: a local Qwen3.5-9B agent delegates critical steps to a cloud LLM, cutting cost up to 62% on SWE-ReBench and 34% on Terminal-Bench 2.1.
NVIDIA and AI4Bharat built 2.8T synthetic tokens across 22 Indian languages with an open pipeline. How it works and what the gains are.
