Nov 8, 2025

• Reverse Engineering a Neural Network's Clever Solution to Binary Addition (2023)

  • The neural network effectively learned binary addition by utilizing a unique architecture that mimics a digital-to-analog converter, allowing it to sum inputs and convert them to analog signals.

• Study identifies weaknesses in how AI systems are evaluated

  • A recent study reveals significant weaknesses in the evaluation methods of AI systems, highlighting the need for more robust and standardized assessment frameworks.

• PyTorch Helion

  • Helion is a high-level Python-embedded DSL that compiles to optimized Triton code, enabling developers to create performant ML kernels without deep hardware knowledge, thus reducing technical debt and enhancing productivity.

• Oddest ChatGPT leaks yet: Cringey chat logs found in Google Analytics tool

  • Sensitive ChatGPT conversations have been inadvertently leaking into Google Search Console, revealing user prompts that were expected to remain private, raising significant privacy concerns about OpenAI's data handling practices.

• Analysis of Hedy Lamarr's Contribution to Spread-Spectrum Communication

  • Researchers.One is a platform that facilitates article submissions for a fee of $10, enhancing accessibility for authors and readers alike.

• [R] WavJEPA: Semantic learning unlocks robust audio foundation models for raw waveforms

  • WavJEPA is a novel audio foundation model that operates on raw waveforms, achieving superior performance in audio representation tasks with significantly less compute and training data compared to models like Wav2Vec2.0 and HuBERT.

• Humans have remote touch 'seventh sense' like sandpipers

  • Humans possess a form of remote touch, akin to that of sandpipers, enabling them to detect buried objects through subtle mechanical cues in granular materials, despite lacking specialized anatomical features.

• [R] GRAM: General-purpose Real-world Audio Model to efficiently learn spatial audio representations.

  • GRAM introduces a General-purpose Real-world Audio Model that effectively addresses the performance drop in audio foundation models when faced with real-world acoustic environments characterized by reverberation and noise.

• AI benchmarks are a bad joke – and LLM makers are the ones laughing

  • A study from the Oxford Internet Institute reveals that only 16% of 445 LLM benchmarks for AI use rigorous scientific methods, raising concerns about the validity of performance claims.

• Understanding prompt injections: a frontier security challenge

• Notion’s rebuild for agentic AI: How GPT‑5 helped unlock autonomous workflows