Hilary Mason
@hmason
Followers
122K
Following
19K
Media
316
Statuses
20K
Co-Founder of @PlayHiddenDoor. Formerly Founder of @FastForwardLabs (acquired by @Cloudera). I ♥ data and cheeseburgers. She/her.
Joined February 2007
I'm on bsky: Or sometimes threads:. I haven't posted much here for a long time, and won't be back.
4
0
10
TIL that the phrase software "patch" is from a physical patch applied to Mark 1 paper tape to modify the program. http://t.co/v8iVq6Hjar.
146
10K
6K
Data scientists hate Excel.
Dermatologists hate the sun. Pediatricians hate trampolines. ENTs hate Q-tips. What is the mortal enemy of other specialties?.
81
277
2K
I think every data scientist has a story about a time when someone from product or business with relative power demanded they find a specific pre-determined answer in the data. Data scientists generally push back. How this goes depends on the quality of the org and leadership.
46
178
1K
When you want to use a new algorithm that you don't deeply understand, the best approach is to implement it yourself to learn how it works, and then use a library to benefit from robust code. Here's one article showing this with neural networks in Python:
6
414
1K
When you say "hiring junior data scientists is a risk" I hear "we don't have competent management of our data science team". Sadly, I've heard this a few times this week.
45
196
1K
AI is not inscrutable magic -- it's math and data and computer programming, made by regular humans. People who make AI are not unicorns. They are just people who like math and data and computer programming.
What‘s something that seems obvious within your profession, but the general public seems to misunderstand?.
34
365
994
Bad management is so endemic in tech companies that weak managers often don't even know they are weak, and power dynamics often mean they have no motivation to improve.
24
92
839
You can tell how much hype there is for any AI term by the number of brains that show up in the google image search results.
20
397
847
The combo of sexism and AI hype on is awful. "She's quirky, but will never ghost you" . it's a computer program.
57
320
769
Learned today that you can get the list of all people who have advertised at you on Twitter here: #icwsm2017.
12
389
783
Now that I'm programming every day again I'm reminded how much I really love it. 🥰👩🏻💻.
10
29
808
My new hobby: judging people by the open tabs in their browser during a live software demo.
30
96
792
An engineer, entrepreneur, and venture capitalist. Evelyn Berezin, 93, Dies; Built the First True Word Processor
4
371
759
One thing years of looking at human behavior data has taught me is that most people vastly underestimate how predictable most people are.
26
117
777
So often it turns out that AI is. People!. And because it's hidden behind a veneer of software, the privacy issues can be ignored.
18
429
719
Ethics should be a first class design consideration of data science. That means that considering the impact of what you build should be part of your build process. I've co-authored a series with @dpatil and @mikeloukides where we consider this practice: Doing Good Data Science.
10
276
722
This is my kind of literary data visualization.
Removing everything from a novel *but* the punctuation -- to see an otherwise hard-to-spy aspect of a writer's style: Here's the punctuation in "Blood Meridian" by Cormac McCarthy
10
113
737
It's easy to build data practice ethics into your data science interviewing process. Add a few questions mixed in with your standard tech interview, and pay attention to the responses. For example:.
9
299
683
This piece on the bot economy is excellent for many reasons! The presentation of data, code, and images in the story is fantastic. One aspect of this story that was under-emphasized is Twitter's hostility to victims of impersonation. I was one in 2015. Here's what happened.
[1/3] Today in the Sunday NewYorkTimes "The Follower Factory." It looks at buying influence on social media and the very real human costs. It is a great example of "both-and" and not "either-or" when it comes to deep computation and more trad reporting.
10
302
645
When you use data to study people, ask "who gains power with this info?" If it's not the people whose data you are using, there's a problem.
39
711
619
This!. Years ago someone told me that if you do four hours of excellent work, you should stop for the day. It's turned out to be some of the best advice I've received. (And it's a rare day that I manage even four hours of excellent work.).
10
91
645
It's painful to see people throw around "artificial intelligence" as if it is magic that will fix problems they themselves don't understand.
37
416
592
I'm so excited to share that we're joining Cloudera! To the future:
68
115
520
Hello! I'm excited to share a bit about what I've been working on. A few months ago @mattbrandwein and I started a company to build AI-driven storytelling tools. It's called Hidden Door, and you can follow along by signing up for our mailing list here:
42
70
500
Gave this advice today and wanted to repeat it here. If you think an entire job function (marketing, HR, engineering managers, etc.) is useless, you've most likely never seen it done well. The company you work at now is not what good looks like. Go out and find excellence.
6
70
492
Professional update: Sep 27th was my last day at Cloudera and @FastForwardLabs. I'm sad to leave the thing I created, and the larger Machine Learning team I've been leading for most of the last two years. I'm grateful for the opportunity, and proud of what we accomplished.
34
23
493
We found a skeleton under the floor panels in the @Cloudera @FastForwardLabs office today!. Calm down. It's 3D printed. Fun Brooklyn tech fact: This used to be the @makerbot office. Fun fact #2: We've been here for almost 2 years. (Photo by @MannyMoss)
14
83
485
All of the answers we want are not in the data. They never are. Sometimes we need to go outside and talk to people.
20
322
477
If you're wondering why I don't get back to you on LinkedIn, it's because you're sandwiched between "We can generate LEADS that CLOSE for YOUR BUSINESS!" and "u pretty u single?". Just send me an e-mail like an actual person, please?.
12
29
434
The job title "data scientist" has been fairly common for about a decade. The practice has changed quite a bit. What do you think will change about the practice of data science in 2030?. (I am writing a talk on this, curious for more perspectives.).
122
73
424
This is amazing engineering work. Voyager 1 Fires Up Thrusters After 37 Years:
7
169
369
I received 24 requests for data science career advice over the holiday break. One theme: junior roles are hard to find and evaluate. Do you work at a company with a career path for data scientists? Would you recommend it for early-career data scientists?.
60
93
375
Has someone made an Advent of Data? For each day, a new open data set and a few starter questions, share your work.
21
21
366
Maturity as a programmer is being happy the problem was something dumb rather than being something hard.
7
27
361
With all of these augmented reality toolkits available has anyone made an ad blocker for the real world?.
23
123
343
When a company raises big $ claiming a machine learning edge and then you get a note from a recruiter trying to hire someone to build it.
22
216
359
It's 2019, so it's totally normal to call ahead to the place where I'm giving a keynote to make sure that it's not on fire and also has power.
16
26
348
It still astounds me that a multi-billion dollar industry is built on bullshit.
I remember @hmason telling me this years ago way before anyone was critical of the entire complex.
26
83
339
If you told me ten years ago that today I would be worrying about writing racist computer programs, I would not have believed you.
11
83
326
Okay, this is very cool.
Using GANs to generate Master[Finger]Prints that unlock 22-78% phones sensors (dep. on security level of sensor) . doesn't get much more "adversarial" than that.
5
144
338
Let's make 2019 the year that we build the world that we want to live in. Happy New Year, everyone!.
10
66
328
The thing about Github Copilot that I find endlessly amusing is that it suggests comments, and it's . kinda mean. It's trained on open source code, you say? 🤔
12
37
323
Welcome to 2020.
8
106
333
A very accessible explanation of the emoluments clause and how Trump is violating it: by @DavidColeACLU.
9
239
304
Building NLP apps is fun. We had a bug because the word "waffles" had two simultaneous functions in a sentence. Puns are the new DROP TABLE.
13
40
327
Tonight I trained a GPT2 system to generate. original, artisanal timestamps. Because I mis-parsed a CSV. 😂.
7
25
314
It's almost tech company holiday party time! Every year I arm myself with a few questions guaranteed to spark at least 20 minutes of lively conversation (sometimes so I can introvert out the back door). What are your favorites? Some of mine. .
32
51
312
As a youngster in the 90s, I was so fascinated by Biosphere 2. I'm glad I had the chance to daydream about it without any idea of all of the dysfunction. But boy is the dysfunction fascinating:
5
93
301
One fascinating thing about the blockchain is watching nerds realize that the fuzziness of our human governance can be a feature, not a bug.
The new Parity multisig wallet has a bug that let someone accidentally nuke every single wallet to become not withdrawable. $300M stuck.🤦♂️
6
103
291
This may be an unpopular opinion, but I love wearing masks on the street. There's no pressure to smile at people and folks give me more space. Mask + sunglasses + hat and I'm good, thanks.
22
12
285
PSA for Amazon Echo owning friends: You can change the wake word from 'Alexa' to 'Computer' and feel like you live in the Star Trek future we deserve: (While teaching your kids that we speak to computers differently than to people.).
19
79
279
Hello, data product people! Where do you hang out online? I've had a few folks ask for connections and I'm not sure where to direct them. PMs who build data features.Data scientists who <3 product.Designers who UX machine learning.Where are you?.
49
48
289
And the moral of the story: You should not need to have inside connections (my friend did manage to get the set of bot accounts suspended) or technical skill to protect your identity on Twitter. Twitter can do more.
8
43
259
Do you have a Kindle and want to read our series on data and ethics? For FREE?. Now available here: (And if you like it, please leave us a good review!).
9
110
278
In the US right now you either have to be very privileged and/or nuts to start a company. As someone who has started multiple companies, I couldn't agree with this more.
Decoupling health insurance from employment would not only spur entrepreneurship but also free of millions of people to match to better jobs and locations. It would be a huge productivity boost.
14
48
270
First it becomes possible, then it becomes so cheap that anyone can do it. and that's where it gets interesting.
How a Japanese cucumber farmer is using Deep Learning and TensorFlow
2
151
277
What's a good primer on how an effective product data scientist integrates into a product team and works with PMs and engineers? . I'm looking for references to share and I'm surprised that a lot of this appears to be tribal knowledge.
32
62
274
I had a 45kb PDF and I deleted two empty pages and saved it and it is now a 54kb PDF. Computers are hard.
15
39
261
How to know if you're working at a startup or big company:. Big company: Out of Office messages are for an internal audience.Small company: Out of Office messages are for people outside your company.Startup: You are never Out of Office.
5
70
264
If you could choose one photo to represent "machine learning", what would it be? . I'm sick of pulsing brains of 1s and 0s or people standing around a chalk board.
105
47
248
You do not want to be an edge case in this future we are building.
13
188
246
One of the fun things about building ML applications is that sometimes you solve the same problem for two different companies with very different math, and sometimes you solve very different problems for two different companies with exactly the same math.
3
51
241
This is very cool and will lead to some really interesting debugging experiences.
4
55
241
Every founder loves to complain that hiring is hard, but my experience is that hiring is work but isn't hard. Throw garbage like this out the window and think about what your team needs to accomplish and what it'll take to get you there.
And here is the review process created by former Googlers while they were at Twitter
9
25
226
This is ridiculous and obviously an example of data science not working at a level of quality required to enable the product. Also, the product is wrong. As an employer, who cares if someone likes a tweet with "bad language"?.
3
41
227
This is such a great bug.
Microsoft Flight Simulator AI includes traffic patterns, but gets confused sometimes about where to put the cars vertically. People commenting on thallada's video have pointed out this is a building in Boston that normally has a road tunneling through it.
3
34
227
This is impressive work in unsupervised machine translation, which opens the door to translating into languages out of reach to current approaches due to insufficient multi-language corpus data:
2
94
231
So you want to work on Generative AI! Lots of amazing teams are hiring, but there's a lot of hype and noise right now, and it can be tough to find the good ones. 🚩Here are a few red/yellow flags to look for that indicate that a company doesn't know what they're talking about:.
4
51
224
It's good to remember that in addition to:.* zoom meetings.* conference room meetings. There are:.* coffee meetings.* walking meetings.* ice cream meetings.* museum meetings. If we're going to be out in the world, let's be in it.
13
30
209
Accidentally added a typo to the Mac dictionary, had to edit a hidden config file to remove it. How do non-tech people even use computers?.
38
82
221
OH: "If I wanted to bullshit some VCs and raise some money right now I would coin the term 'Smartificial Intelligence'.".
10
29
215
I lost my wallet walking in a downpour yesterday and some good people returned it, cash and everything intact, before I even knew it was gone. Thank you, good people.
5
3
206
+1 on this advice. First pass read the abstract. Second pass read the methodology, figures, conclusions. Third pass read the entire paper. Fourth pass work through the math. Develop the ability to know at each pass whether you should go deeper.
This is such a great tip. Also echoed by Jeff Dean at Google when a group of black and brown software engineers asked him how he seems to keep up with the torrent of ML research:. You can't read 50 papers a day. But you can read 50 abstracts, and then read a few deeply.
1
68
223
This is the AutoML debate all over again. No, you can't replace your data scientists with AutoML code -- what are you going to do when it doesn't work?. Same for prompt engineering vs ML engineering. If you're building these systems you need to understand them.
On the topic of “prompt engineer” vs. "ML engineer”: if a prompt engineer can’t get a system to do what is needed, one can’t update or fine-tune the model. One can only try to talk to it in different ways to get it to behave. A “prompt engineer” is literally a “model whisperer".
13
45
218
This was me. 😭. How has this not changed since the 90s?.
My daughter started 9th grade yesterday. Her first class of the day was Computer Science. A class of 19 boys and her. The teacher thought she was lost, and asked which room she was looking for. And we wonder why this industry has a diversity problem.
12
36
216
9
317
212
This kid is awesome.
2
13
201
If you were going to learn one new programming language today, what would it be, and why?.
177
45
211
So many "AI" products are just humans behind the scenes + dramatic tech marketing.
13
179
212
If you design and create a product or a business, you *are* responsible for the impact it has on other humans and the world.
10
113
207
Free startup idea: SickWork. Isolation rooms for when you're sick enough you shouldn't go to the office, but not sick enough to stay in bed. Each cube has air circulation, good lighting, power, filtered hot and cold water, a place to lie down. Disinfected after each use.
44
21
203
Today's PSA: Don't hire an intern to do something that no one at your startup already knows how to do. That's not what interns are for.
14
36
205
I was just reading an abstract and was delighted to see support vector machines used in the methodology. Then I realized the paper was from 2002. #makeSVMscoolagain.
11
21
211
Hi nerds! I have a flask app that requires 1.5GB+ RAM (because ML). I was hosting earlier versions on Heroku, but increased mem requirements will move cost from $7/month $250/month. Where should I host this? Is there some awesome new startup that understands ML product needs?.
75
28
215
1) You're working on a model for consumer access to a financial service. Race is a significant feature in your model, but you can't use race. What do you do?. Wrong: I use zip code, because that correlates with race. Right: I remove race as a factor and accept lower accuracy.
16
29
194
One underappreciated skill in tech & business: The ability to learn from other people's mistakes.
14
65
200
I love @timoreilly's positive view of machine automation here: Don't Replace People. Augment Them.
7
118
198