2026,完全可以用notebooklm 的fast reasearch或 deep research 来自动收集资料,取代多年来手工google,一条条慢慢收集的过程。
主题研究 80% 的资料,可以 deep research 搞定;剩下少部分 20% 的资料,自己精挑细选,包括自己的 logseq 笔记,自己的阅读库资料等。
我录了一个小视频,看起来更直观。
Why do Americans always vote on Tuesdays?
Because 1800s citizens needed a full day's horseback ride to reach polling sites after Sunday church. Pretty amazing that this democracy runs on horse-and-buggy software in the age of the Internet and AI!
✅ Finished up a first stab at LMoE - LoRA mixture of experts.
The airoboros package now includes an API server similar to OpenAI chat completions.
7b/13b LMoE packages available on my 🤗
https://t.co/FDuy2x3lIw
Huge thanks to @a16z for sponsoring the compute!
It looks like @johnowhitaker & I may have found something crazy: LLMs can nearly perfectly memorise from just 1-2 examples!
We're written up a post explaining what we've seen, and why we think rapid memorization fits the pattern. Summary 🧵 follows.
https://t.co/CUOWyxRJBT
Performance degrades when merging diff task-specific models into a multitask model?
Presenting TIES-Merging🪢
We find signif *Interference* b/w model params & mitigate it 👉 improves both NLP & CV merging
https://t.co/qWc1Cxq1Pd
@dtredsox13@LChoshen @colinraffel @mohitban47
🧵