标签:opinion

带有 opinion 标签的公开 Whisper

划选高亮2026-08-12 10:14:25
原文高亮摘录
next generation of highly compute-efficient multimodal models
Whisper 随想笔记
Compute-efficient but what about the energy cost for training these things?
划选高亮2026-08-11 03:09:18
原文高亮摘录
A Foolish Consistency is the Hobgoblin of Little Minds
Whisper 随想笔记
Classic quote, still makes me think twice about over-obsessing with style.
划选高亮2026-08-09 21:09:57
原文高亮摘录
error handling is intuitive
Whisper 随想笔记
Error handling being intuitive is a huge win, honestly.
划选高亮2026-08-09 21:09:18
原文高亮摘录
Flat is better than nested.
Whisper 随想笔记
The flat is better than nested one really hits home with every refactor I do.
划选高亮2026-08-08 15:09:18
原文高亮摘录
this is sounding awfully familiar.
Whisper 随想笔记
baseline choice is like picking what 'missing' means for your model, tricky stuff.
划选高亮2026-08-04 21:09:18
原文高亮摘录
When appropriate
Whisper 随想笔记
Good point, sync order makes debugging way easier.
划选高亮2026-07-30 21:09:18
原文高亮摘录
not a modular toolbox of building blocks for neural nets.
Whisper 随想笔记
That's a refreshing take, keeping it simple for researchers to hack on.
划选高亮2026-07-28 13:01:34
原文高亮摘录
Content-based filtering methods are based on a description of the item and a profile of the user's preferences
Whisper 随想笔记
But what if the user's taste changes over time? This seems too static for that.
划选高亮2026-07-28 09:49:34
原文高亮摘录
Collaborative filtering is based on the assumption that people who agreed in the past will agree in the future
Whisper 随想笔记
But what if I suddenly change my taste? That assumption feels shaky.
划选高亮2026-07-27 12:40:15
原文高亮摘录
since it often causes poorly readable code.
Whisper 随想笔记
I only use * in the REPL, it's too handy to give up there.
划选高亮2026-07-27 12:31:15
原文高亮摘录
since it often causes poorly readable code.
Whisper 随想笔记
Yeah, wildcard imports make it impossible to know where names come from.
划选高亮2026-07-27 10:21:52
原文高亮摘录
Graph neural networks (GNNs) are artificial neural networks designed for tasks whose inputs are graphs.
Whisper 随想笔记
But aren't graphs just a special case of tensors? Seems overhyped to me.
划选高亮2026-07-27 09:49:28
原文高亮摘录
Language models are useful for a variety of tasks
Whisper 随想笔记
They forgot spell check, which is the only one I actually use daily.
划选高亮2026-07-27 09:41:50
原文高亮摘录
Deeper neural networks are more difficult to train.
Whisper 随想笔记
Yeah but adding layers never made anything easier in my experience.
划选高亮2026-07-27 09:40:28
原文高亮摘录
Language models are useful for a variety of tasks
Whisper 随想笔记
Sure, but route optimization feels like a stretch—that's more graph theory than language.
划选高亮2026-07-26 16:09:06
原文高亮摘录
Use spread syntax with restraint.
Whisper 随想笔记
Yeah but when you have like 10 props it's just so tempting...
划选高亮2026-07-26 16:04:21
原文高亮摘录
each generation ends up corresponding to a new step in a random walk of model parameters
Whisper 随想笔记
So basically AI models are just wandering off randomly with each new generation, wild.
划选高亮2026-07-26 12:59:31
原文高亮摘录
There is no information about the world which is hidden from the state.
Whisper 随想笔记
So if the state is complete, we never need to infer anything? Seems too ideal for real life.
划选高亮2026-07-26 12:38:41
原文高亮摘录
unpredictable phenomenon that we refer to as emergent abilities
Whisper 随想笔记
But is it really unpredictable or are we just bad at measuring small gains?
划选高亮2026-07-26 10:20:43
原文高亮摘录
A model is transparent "if the processes that extract model parameters from training data and generate labels from testing data can be described and motivated by the approach designer."
Whisper 随想笔记
I think this is spot on — transparency is about the designer's ability to explain, not just the code.