Saksham Srivastava·Oct 27SMN2-RescueDiff: What 15,000 Docking Jobs Against SMN2 Taught Me About AI Drug DiscoveryAcross 147 generated molecules screened on an identical 97-decoy panel, not one achieved worst-case separation from the decoy set. Neither…
Veter·Sep 10How GPT Works: Without Code or Complex MathI’ll try to explain how autoregressive Transformers, such as the ones behind ChatGPT, work without complex math or formulas. Naturally, I…
Eleventh Hour Enthusiast·Aug 15Autoregressive Language Models are Secretly Energy-Based ModelsWhat the bijection proves, and what it asks the architecture to carry
InGoPenAIbySubrat Mishra·Aug 4Inside DeepAR: The Architecture Behind One Model That Forecasts Thousands of Time SeriesIntroduction
Jason Yang·Jun 26One Silicon, Two Brains: How NPU Optimization Powers Both Autoregressive and Diffusion Transformers…The Rise of Physical AI and Its Compute Demands
Sumit Sah·May 22Revolutionizing Real-Time Video Generation: Meet Q-ARVD, The Quantization Breakthrough for…How researchers solved the unique challenges of quantizing autoregressive video diffusion models to make real-time AI video practical.
AmeerSaleem·May 17How G-EVAL improvements vanilla LLM-as-a-judgeAn overview of Chain of Thought prompting and the G-EVAL framework, with insights from their debut papers.
InGenerative AIbySourav Mukherjee·Apr 3Beyond Autoregression: How Diffusion Language Models Are Rewriting the Rules of AIWhat if the way we’ve been generating text for the past 5 years is fundamentally suboptimal?
