mlpowered releases podcast deep-dive on what's inside LLMs
mlpowered · x · 2026-10-19
Following his interpretability explainer series (models plan ahead while outputting one word at a time, internal concept representations, world understanding needed for hard predictions), mlpowered released a companion podcast covering the topics in more depth, inviting listeners to weigh in.
More from Research
- Agent reproduces black hole images from EHT raw data in 2 hours for ~$7 — fofrAI · 2026-10-19
- The mysterious 'Chinese spirit neuron' may trace back to a Dream of the Red Chamber translation — Sauers_ · 2026-10-19
- Why native multimodal models won't win it all: they can't compress visual context — gabriberton · 2026-10-19
- 4-year bitter-lesson evolution of MLLM vision: from LLaVA crops to native-resolution images — gabriberton · 2026-10-19
- Data World paper: RL on synthetic data environments adds 20.9 points on DABstep-Hard — omarsar0 · 2026-10-19
- NVIDIA open-sources Structured Data Models library unifying tabular foundation models in PyTorch — lmoroney · 2026-10-19