筛选结果
13 篇排序
2:59:44EN
Cognitive Revolution "How AI Changes Everything"· 2026年6月3日播客Two Endangered Languages Broke Transformers. Nested Learning Gave Models a Sleep Button.主持 Nathan LabenzGoogle researcher Ali Behrouz tells Nathan Labenz how stacking update frequencies — not just layers — and an offline sleep phase could give language models genuine continual learning.
2:12:00ENCognitive Revolution "How AI Changes Everything"· 2025年5月15日播客Gradient Descent at Test Time: Titans Pushes LLM Context to 10 Million Tokens主持 Nathan LabenzA Cornell PhD student's neural memory module holds about 70% accuracy where GPT-4's scores fall off a cliff.
1:18:32ENCognitive Revolution "How AI Changes Everything"· 2024年9月20日播客A DNA Model That Can Tell Which Genes You Cannot Touch主持 Nathan LabenzStanford's Brian Hie walks through hybrid architectures, a genome-scale language model trained only on DNA, and an antibody that binds 25 times tighter than nature.
1:42:16ENCognitive Revolution "How AI Changes Everything"· 2024年7月4日播客267 Papers in Seven Months: Albert Gu Explains the State Space Model Revolution主持 Nathan LabenzThe Mamba co-author walks through five years of recurrent-model work, why selective compression beats a growing KV cache, and what Mamba 2 trades away to train faster.
1:19:24ENCognitive Revolution "How AI Changes Everything"· 2024年4月5日播客A Gibberish Prompt Unlocks What GPT-J Plans Three Tokens Ahead主持 Nathan LabenzNortheastern’s Bau Lab shows that a 40,000-parameter soft prompt can read a model’s future words, and that about ten attention heads carry the task itself.
1:20:44ENCognitive Revolution "How AI Changes Everything"· 2024年3月30日播客Seeing Like a Mamba: Vision, Rotting States, and a Cell Model in 90 Days主持 Nathan LabenzNathan Labenz and Jason Meaux survey the vision papers, long-context experiments, and biology work that followed the original Mamba release.
2:30:06ENCognitive Revolution "How AI Changes Everything"· 2023年12月22日播客Trained on a Thirtieth of GPT-4's Tokens, Mamba Beat Transformers on Home Turf主持 Nathan LabenzIn a solo emergency episode, Nathan Labenz argues that Albert Gu and Tri Dao's selective state space model is the first architecture since 2017 that deserves to share the Transformer’s throne.
39:02ENCognitive Revolution "How AI Changes Everything"· 2023年8月10日播客AI’s product advantage is consistency. Its real weakness is getting conned.主持 Nathan LabenzNathan Labenz updates the Cognitive Revolution scouting report: jailbreaks, resume automation, Med-PaLM’s 40% radiology win, and what might follow Transformers.
2:13:07ENCognitive Revolution "How AI Changes Everything"· 2023年7月27日播客Half an Hour to a Circuit in GPT-2 Small — and a Quadratic Wall Before GPT-3主持 Nathan LabenzArthur Conmy walks Nathan Labenz through ACDC, the algorithm that automates the most tedious step of mechanistic interpretability: finding which transformer components actually do a task.
1:32:22ENCognitive Revolution "How AI Changes Everything"· 2023年6月29日播客A Million Bytes and 256 Symbols: Meta's MEGABYTE Wants to Throw the Tokenizer Away主持 Nathan LabenzLili Yu of Meta AI explains a multiscale transformer that predicts million-byte sequences, skips tokenization, and aims to model any file as raw bytes.
2:00:19ENCognitive Revolution "How AI Changes Everything"· 2023年6月6日播客1.5 Million GPT-4 Fairy Tales Let a 33-Million-Parameter Model Out-Reason GPT-2主持 Nathan LabenzTry this fill-in: Lily likes cats and dogs. She asked her mom for a dog. Mom said no, so she asked for a…
1:16:52ENCognitive Revolution "How AI Changes Everything"播客Mamba After 90 Days: What 30 Papers Reveal About State Space Models, Hybrids, and Mixture of Experts主持 Nathan LabenzNathan Labenz and AI scout Jason Meaux survey the first wave of Mamba-inspired research, from mechanistic studies and synthetic-task tradeoffs to hybrid architectures and MoE variants.
1:03:10ENCognitive Revolution "How AI Changes Everything"播客Pause Tokens, Analogical Prompts and Ring Attention: A Tour of the Transformer's Next Upgrades主持 Nathan LabenzNathan Labenz and Trey Kollmer walk through five recent papers on pausing, backspacing, prompting, streaming and long-context training, and sketch how they might combine into a more capable model.