Signal to noise is important. If you're creating content - especially educational content - prefer quality over quantity. Your audience is better served by a smaller volume of thoughtful, refined work.
— Josh Gordon (@random_forests) October 15, 2018
The problem with data-driven decisions: you don't see what happens to the data in the long term. What looks good for your metrics right now might kill your product in a few years.
— François Chollet (@fchollet) October 15, 2018
It's like AI: data gives you local generalization, but for the long term, you need principles.
Fantastic post that I co-sign 100%. Yes, tech does have systemic issues, but it also continues to be one of the few professional fields today that has that flexibility that allows parents to be both parents and professionals, IMO. https://t.co/mOL4ovzbD8
— Vicki Boykis (@vboykis) October 14, 2018
At the end of the day, your attention is the only resource you can allocate. https://t.co/xw7AgPBfjw
— François Chollet (@fchollet) October 13, 2018
A lot of really poorly designed software is software that was never, in fact, designed: it evolved organically, starting as one thing and eventually morphing into another, through incremental additions without any overarching vision.
— François Chollet (@fchollet) October 11, 2018
this crystallizes what’s wrong with AI today: we have no techniques whatsoever to transfer a general, if hard-won understanding of the world to new tasks. we instead do almost everything from scratch, because our representations of what is learned are too superficial. https://t.co/2wRB4wgdqa
— Gary Marcus (@GaryMarcus) October 11, 2018
Replies indicate that people are very confused about what "bias" means. It means doing pattern recognition based on spurious correlations, as opposed to causal reasoning. A ML model will use all correlations found in the training data, and typically many of them will be spurious.
— François Chollet (@fchollet) October 11, 2018
Many people believe that machine learning algorithms are analytical -- that they ponder over the data available and do logical, unbiased reasoning using some internal model.
— François Chollet (@fchollet) October 11, 2018
They're the opposite of that: they're intuitive. They do pattern recognition. They're System 1, not 2.
Most of the valuable work we do in life is hidden. Rather than seeing the work in full we almost always see an information iceberg - a snapshot result supported by a hidden and complex creative process below. Being open helps brings the rest of the iceberg above the surface. https://t.co/rsHWTZnnQl
— Smerity (@Smerity) October 10, 2018
One of the biggest advantages of openness -- in research, in personal thinking, and life broadly -- is that it prevents evaporation.
— Chris Olah (@ch402) October 9, 2018
Many of my thoughts and experiences have been lost or become illegible over time. The things I shared publicly remain.
Anyone who tells you they don't have to look up things when they code isn't someone you want to work with. We're all learning on the job, and part of learning is learning when to memorize and when to look things up! https://t.co/fOJ6UsMEj3
— Caitlin Hudon👩🏼💻 (@beeonaposy) October 9, 2018
Over the last year, I struggled with many career-related questions: PhD/big company/startup, etc. I'm fortunate to have received advice from people I admire in industry and academia. I tried to put their advice + my thought process into words https://t.co/gLKBpAadmP
— Chip Huyen (@chipro) October 8, 2018
