When government pays companies to build big custom software programs for them, they succeed just 13% of the time. Now I will tell you why failure is so common, and about the simple change that turns those outcomes on their head.
— Waldo Jaquith (@waldojaquith) February 5, 2020
There are many problems in field of AI:
— Rachel Thomas (@math_rachel) February 5, 2020
false marketing & hype
phony gurus
algorithms that amplify bias & disinfo
misaligned incentives/profit models
Bootcamp grads & MOOC takers are not a primary cause. Gatekeeping about what kind of ppl have "right background" solves nothing
In my career, I've worked on speech, NLP, and vision research. Personally, vision projects are most appealing and fun to show off. NLP research is notoriously hard yet most insightful. I feel most satisfied when I work on speech. But I would be bored if I did only one of them.
— Delip Rao (@deliprao) February 5, 2020
It's so easy to blame "tech" instead of
— dj patil (@dpatil) February 4, 2020
- poor design
- no user experience
- lack of technologists who actually know tech
- process failures
- procurement & contracting
- thinking tech is easy
"No one WANTS to learn by mistakes, but we cannot learn enough from successes to go beyond the state of the art." - Henry Petroski, in "To Engineer Is Human" an idiosyncratic and thought-provoking book of essays. https://t.co/C9yp4OrLMu
— Tim Harford (@TimHarford) February 3, 2020
I have this working theory that more vague the description of your AI startup, the higher your valuation. What does "liberate your data", "allow data to move like the air you breathe”, “we lived and breathed data science”, “AI that delivers impact, not accuracy” even mean?
— Chip Huyen (@chipro) February 2, 2020
The way we talk about data science and focus so much on methods, we actually incentivize working with *bad* data, rather than spending the time to collect good data and then use easy methods with it
— Hilary Parker (@hspter) January 31, 2020
Also, GBMs can often do better on these kinds of problems. The winner of the competition discussed in the article found GBMs better than RFs, and factorization machines better still. pic.twitter.com/KjsanRXcyb
— Jeremy Howard (@jeremyphoward) January 31, 2020
John Schulman is one of the researchers I admire most in the field. Awesome guide for how to do a PhD / break into research! I need to implement more papers from scratch... https://t.co/5gHlTm7DtL
— Eric Jang 🇺🇸🇹🇼 (@ericjang11) January 31, 2020
I was gifted "Hackers and Painters" by Paul Graham and it seems like a case study for Dunning-Kruger and I'm sad. He describes how naive Bayes works for spam filtering though (in 2004, which is much after it was introduced, but early for a general tech person to talk about ML).
— Andreas Mueller (@amuellerml) January 30, 2020
At #rstudioconf @jtleek talks about addressing income inequality through access to education:
— Julia Silge (@juliasilge) January 29, 2020
Talent is evenly distributed but opportunity isn't.
Most people who took Jeff's MOOCs were already educated, white, male, etc.
Can this be changed to increase access? 🤔 pic.twitter.com/TX9FnmU1tT
2010: mobile first
— Chip Huyen (@chipro) January 29, 2020
2016: AI first
2020: privacy first
It’s exciting to see so many startups working on data privacy. My prediction is that 5 years from now, we'll look back and cringe at how primitive we are all about our privacy right now.
