🎯 What bothers me so much about this rhetoric is that the purported benefits to society are amorphous yet assumed, whereas the tangible losses to actual people are clear and concrete but collateral.
— Ryan Calo (@rcalo) February 4, 2022
🎯 What bothers me so much about this rhetoric is that the purported benefits to society are amorphous yet assumed, whereas the tangible losses to actual people are clear and concrete but collateral.
— Ryan Calo (@rcalo) February 4, 2022
This! A common question people ask is whether they should work with .py vs .ipynb files. It doesn't have to be exclusive. E.g. want a nb with plots but have a loss function that you keep reusing? Put it into a .py file (doesn't have to be a pgk) and import into your notebooks. https://t.co/Eib4iJUyFs
— Sebastian Raschka (@rasbt) February 2, 2022
Every week I hear about some folks that look at GradCAM (apparently most of medical imaging research) or SHAP as though anyone knows what, if anything, these “explanations” mean and it’s terrifying. Overall, I believe it’s already in “actively harmful” territory.
— Zachary Lipton (@zacharylipton) February 2, 2022
A lot of machine learning research has detached itself from solving real problems, and created their own "benchmark-islands".
— Christoph Molnar (@ChristophMolnar) January 24, 2022
How does this happen? And why are researchers not escaping this pattern?
A thread 🧵 pic.twitter.com/uggKd7RsJf
The more we understand the theory of deep learning, all the cushy ML engineering jobs will become mundane data processing jobs (some might argue it already is), and it will lower the competence/training needed to fill those jobs.
— Delip Rao (@deliprao) January 23, 2022
Good discussion about ConvNets vs Transformers https://t.co/MmPInthvFH
— hardmaru (@hardmaru) January 13, 2022
For the same reason therapy is hard. https://t.co/YisEFvmORc
— francesc (@francesc) January 13, 2022
After my post on real-time machine learning last year, many people asked me how to do it.
— Chip Huyen (@chipro) January 3, 2022
This post discusses the challenges + solutions for online prediction, online evaluation, and continual learning, with use cases and examples.
Feedback appreciated!https://t.co/lywDVpYnUD
This is a terrific comparison of the main free GPU @ProjectJupyter providers available. I agree with the conclusion - the new @awscloud SageMaker Studio Lab is a fantastic option.
— Jeremy Howard (@jeremyphoward) December 9, 2021
It's my 1st choice now for training models where I don't need much disk space. https://t.co/M1XsEoMBH6
The ongoing consolidation in AI is incredible. Thread: ➡️ When I started ~decade ago vision, speech, natural language, reinforcement learning, etc. were completely separate; You couldn't read papers across areas - the approaches were completely different, often not even ML based.
— Andrej Karpathy (@karpathy) December 8, 2021
Hugely important settlement of a lawsuit against the terrible Value-Added Modeling for teachers. https://t.co/lFZ9KpfZpq
— Cathy O'Neil (@mathbabedotorg) December 6, 2021
Brilliant op-ed by @timnitGebru
— ❄️Emily M. Bender❄️ (@emilymbender) December 6, 2021
"So what is the way forward? In order to truly have checks and balances, we should not have the same people setting the agendas of big tech, research, government and the non-profit sector. We need alternatives."https://t.co/N5dEntD7E8
>>