[2204.02311] PaLM: Scaling Language Modeling with Pathways
6
公开标注数
6
参与人数
2026-07-20 10:04:51
首次 Whisper
本页的公开 Whisper
划选高亮2026-07-20 16:28:51
原文高亮摘录
“arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.”
Whisper 随想笔记
So this is how they let the community build stuff now, neat.
划选高亮2026-07-20 13:25:51
原文高亮摘录
“PaLM: Scaling Language Modeling with Pathways”
Whisper 随想笔记
540B params, wonder how much compute that took.
划选高亮2026-07-20 13:16:51
原文高亮摘录
“PaLM: Scaling Language Modeling with Pathways”
Whisper 随想笔记
Pathways finally making a splash in language models, nice.
划选高亮2026-07-20 10:22:51
原文高亮摘录
“Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning”
Whisper 随想笔记
Few-shot with that many params feels like cheating—still, the results speak for themselves.
划选高亮2026-07-20 10:13:51
原文高亮摘录
“Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning”
Whisper 随想笔记
I wonder if the 'remarkable performance' is just on benchmarks, real-world tasks might differ.
划选高亮2026-07-20 10:04:51
原文高亮摘录
“Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning”
Whisper 随想笔记
Scale really does seem to be the magic ingredient for few-shot, but 540B is just insane compute.
分享本页 Whisper
短链接
https://domwhisper.com/s/24fe2b2a7a6f嵌入代码
<iframe src="https://domwhisper.com/embed/24fe2b2a7a6f" width="100%" height="480" style="border:0;border-radius:16px" loading="lazy"></iframe>