Jeremie Harris
@jeremiecharris
Followers
5K
Following
3K
Media
30
Statuses
2K
Co-founder & CEO of Gladstone AI We promote responsible AI R&D and adoption by designing and deploying safeguards against AI-driven national security threats.
Mountain View, CA
Joined November 2017
"We want data scientists.". "Cool, what do you need them for?". "We don't know yet.". 🤔.
7
19
157
A thread about choosing good data science projects, inspired by our method at @SharpestMindsAI. TL;DR: . - build products, not projects.- have a clear "win" condition. 👇.
7
48
159
Things aspiring data scientists think companies care about:. - Jupyter notebooks.- Neural networks.- Hadoop. Things companies actually care about:. - Their bottom line. Show them you can move that needle, and they're all ears. Show them your 6 Kaggle projects and they zone out.
5
32
123
Just published my latest in @TDataScience. It's about all the things that companies want data scientists to know, but that data scientists usually don't know. It probably explains ~70% of #datascience post-interview rejections IMO.
9
32
100
@john_c_palmer This is what everyone tries to do, and therefore attempting to make good decisions is destined to lead to unremarkable outcomes. Successful startups are anomalous. Therefore, a conscious effort to make bad decisions is the only way to produce anomalous outcomes.
4
0
66
A thread on the anatomy of great data science projects. Based on the projects we’ve seen get people hired at @SharpestMindsAI. TL;DR: . -> figure out what the average applicant is doing and don’t do those things. -> use your project to tell a story where you're the hero. 👇.
5
15
63
Data analyst. Requirements:. - Pandas.- Wrangling.- Ability to multitask in a dynamic environment.
What do you call this job?.
1
10
50
Success is the moment that persistence begins to look like luck.
1
11
47
@mckaywrigley - The second law of thermodynamics. - Complex computations will continue to be carried out, whether by biological orgs or machines. - The Ottawa Senators still will not have won a Stanley Cup.
1
0
46
Today we’re announcing the results of the first U.S. government-commissioned assessment of catastrophic risks from AI on the path to AGI. It was a year-long investigation into the national security risks introduced by advanced AI and culminated in an action plan to address them.
4
9
48
The most important differentiator for (most) data scientists in 2020 won't be programming ability, but product sense. Code-free/low-code tools have been adding leverage to the product side of the skill equation for a while now. This year feels like a tipping point.
3
9
46
Higher ed:. We offer a data science program. Taught by people making ~100K/yr. Where experienced data scientists make ~1.5X that. But it's ok, you won't see them much. You'll learn from TAs who are also trying to land their first jobs. It's only $50K upfront. Why are you leaving.
0
10
46
"Do no evil" -> I know what evil looks like, and I can avoid it. "Do the right thing" -> I know what's morally right, and I can make it happen. One of these Google mottos is much more potentially dystopic than the other. And it's not the first.
1
8
38
Learning data science on your own can be hard: problems with motivation, distraction, confusion, etc quickly set in. I wrote a post for @TDataScience to help new data scientists overcome these challenges. Let me know what you think!.
4
15
40
If you're interested in ML and math, and you want to make a disproportionately positive impact, consider working on AI alignment. It's a ridiculously well-funded field that's led by some of our best & brightest. And it's quickly becoming the most important problem in the world.
7
11
42
@antoniogm Ultimately, the algorithms are built by teams that are managed by rooms like this though.
1
1
37
The best people in the world tend to train the future best people in the world. But. The best people in the world tend to be open to training anyone who's hungry enough (if they can possibly make the time). So. Be hungry.Send that DM.Fill out that contact form.Go to that meetup.
1
7
38
I recently published a tweetstorm about why it's so important to build *products* rather than projects in data science. Since people found it useful, I've expanded on it and wrote it up as a blog post 🎉.
0
17
36
90% of data science Master's degrees are just extra-expensive two-year bootcamps with less accountability and almost no public scrutiny over outcomes. Change my mind.
6
5
36
Analytics strategies I rarely regret using:. - Looking at raw data.- Looking at basic historgrams & scatter plots.- Talking to users (yes, it's part of analytics). Strategies I often regret using:. - Building predictive models.- Fancy dimensionality reduction/tSNE etc.
1
5
36
Product = a project with a use case. Before you start your DS project, ask yourself two questions: . --> Who would want what I'm building? <--. --> How good would my model need to be to make them happy? <--. If you can't answer those questions, you have a project, not a product.
To get hired as a data scientist in 2018, you built a project. To get hired as a data scientist in 2020, you build a product. Sure, standards have risen - but today's tools also make it possible to build so much more. New era.
1
10
35
We enjoy @bhutanisanyam1's podcast so much that we thought we should have him on ours 😀. Had a ton of fun recording this one. Sanyam has great insights about transitioning from the academic world to indusry!. Hope you enjoy it as much as we did! 🙌.
2
6
35
There is no "data science interview". There are only the interviews that specific companies give for specific DS positions. Don't prepare for a generic interview. Everyone else is doing that. Research the company. Come with ideas. Ace *their* interview, not someone else's.
1
4
35
My book is out!. Quantum Physics Made Me Do It tells the story of human self-understanding through the lens of physics (and AI). It explores what we can know about reality & how tiny tweaks to quantum theory can reshape our entire picture of the universe. And it’s 96% carb-free!
5
4
31
I'm not the first person to say this, but if the average college student spent their 30K/yr tuition on building a startup rather than taking cookie-cutter courses, they would. 1. Learn way more.2. Be way more employable.3. Maybe have a company at the end.
2
7
34
@powerbottomdad1 I read one analysis that suggested that he had enough money that he could buy every journalist over 10 billion calculators. Not sure why that was the unit chosen, but I was quite alarmed, I assure you.
2
1
31
@jeremyphoward It's interesting how many of these ideas are laundered. - Org 1 says masks don't work, without evidence.- Org 2 says they must not work because Org 1 said so.- Org 1 says they must not work because Org 2 said so. It's turtles all the way down.
2
9
26
One common problem in data science that doesn't get enough attention:. What should you do when you don't have enough data to train a model? . @brodriguezmilla just put together a great post on this topic for @TDataScience. Highly recommend! 👌.
2
8
31
Super excited to partner with @TDataScience on this podcast series. 🙌. Our first interview with @joelgrus is a great listen if you're doing most of your work in Jupyter notebooks!.
The case against the jupyter notebook by @jeremiecharris
0
8
30
Overfitting is seeing patterns that aren't there. --> Examples: conspiracy theories, attribution of malice, excuse-making. Underfitting is not seeing patterns that are there. --> Examples: racism, classism, assuming guilt by association.
1
6
28
Really cool to see @TDataScience launch this awesome concept: short videos explaining data science articles. The first one by @ambervteng, featuring a blog post by @max_pechyonkin is a great watch!.
0
15
29
@EricRWeinstein This type of analysis is what philosophy looks like at its best. The "f(X) isn't true" analysis is what philosophy looks like at its worst. The problem is that we currently call both kinds of thinking "analysis", and both kinds of thinkers "philosophers". They are not.
1
1
24
Had a great time diving into MLOps and data science version control with @lmarsden for the @TDataScience podcast! 🙌.
1
16
23
Had a great time interviewing @solmaz_sh, Head of Data at @Shopify!. We talked about Shopify's full-stack data science philosophy, and what it takes to do high-stakes data science at high-growth companies. This one was a ton of fun! cc @ShopifyData 😃.
0
8
25
We're so bombarded by other people's opinions that the highest leverage thing we can do is step back and take some time to become individuals again. We need to decentralize our cognition.
1
8
24
Uncomfortable reality of regulating #AI:. A government regulator working on AI might get paid ~80K USD/y. An ML engineer in SF makes ~160K USD/y + equity + career advancement. So. If you understood AI well enough to do a good job regulating it, where would you be working?.
5
2
21
Really enjoyed recording this @TDataScience episode with @ykilcher about his take on the "big picture" of current AI research!. Topics. - How to keep up with ML research.- What's overhyped, and what isn't.- When does it make sense to anthropomorphize AI?.
3
6
23
So an easy way to stand out is to design your project as a product, with a clear use case. Show that you can define sensible win conditions. Companies need you to know how to do that. So they’ll be looking for hints that you can. Give yourself the chance to tick that box. fin/.
2
3
22
Just wrote a blog post on the data we've collected about the job search @SharpestMindsAI. Some teasers: . -job boards are awful. -data scientists in Canada make *half* what they do in the US. -don't expect to hear back from companies within a week.
0
8
23
Prediction: in the next decade, technical terminology from machine learning will enter common usage. As ML algorithms become ubiquitous, terms like "overfitting" and "validation" are going to enter the mainstream. And we'll be able to have far better conversations as a result.
2
1
22
@balajis People vastly under-estimate how hard it is to properly allocate capital. In a consumer economy, people have a harder time empathizing with investors and builders. Most people think Musk, Zuck etc "got lucky". Result: a comparative advantage to those who don't believe in luck.
2
3
20
You can either fail many, many times in small ways that don't matter in the scheme of things. Or you can avoid failures at all costs, and turn your life into one giant, unfixable failure that you only notice in retrospect.
0
4
23
No, the right choice is not to delay that scary career decision or life stage by going to grad school.
1
3
21
At the first YC dinner, Sam A started his talk with a quote from General Patton:. "A good plan violently executed now is better than a perfect plan executed next week.". True for startups, but just as true for people.
0
3
21
I just wrote my latest post for @TDataScience on managing impostor syndrome in data science. I hope no one finds out that I don't know what I'm talking about.
1
6
20
We've been asked if we can open-source some of the exchanges that happen on @SharpestMindsAI Slack, between professional data scientists and their mentees, so that others can benefit from them. Excited to publish the next one with @TDataScience! .
0
4
19
With the invention of the physical escape key, Apple re-establishes itself as the world's most innovative company.
0
2
20
Not sure who needs to hear this, but . tweets that begin with "not sure who needs to hear this" would be 31 characters shorter if they did not include the phrase "not sure who needs to hear this".
2
0
22
Really enjoyed talking to former @Shopify Director of Data Science @Cmrn_DP about how data science has evolved, and where it's headed next!. New episode of the @TDataScience podcast is up 😀.
0
6
21
It's never as hard as it seems when you start doing it. It's always more complex than it seems when you're just complaining about it.
2
2
20
Another great post by @SharpestMindsAI mentee @blissfulchar, this time featured on @TDataScience's weekly selection!. 🙌.
0
8
21
Just learned that Why You Shouldn't be a Data Science Generalist was literally *the* most read article on all of @kdnuggets in December. Thanks for reading, y'all!.
0
9
20
We just put out Episode 2 of our podcast series for @TDataScience. 🙌. We talked to Tan Vachiramon, a data scientist @oculus and @SharpestMindsAI mentor, about what factors matter when it comes to model selection. Upshot: simpler is usually better.
0
8
18
@john_c_palmer Related, but also worth mentioning: it's important not to have a scumbag, lazy CEO with no vision whatsoever. Common mistake among first-time founders, and something I've had to advise against often. Would be great to see more people bring attention to this.
1
0
19
Lower short-term expectations. Higher long-term expectations.
1
2
19
I've been working with @TDataScience on a new series of episodes about AI safety and AI ethics, and I'm thrilled to announce that we just released the first one!! 🎉🚀. Really fun conversation with @neutronsNeurons about what it takes to make AI "good".
0
5
20
Another thing I never expected to happen so soon: a @SharpestMindsAI mentee just got hired in a *senior role* (even though it was her first data science job). 🙌.
0
0
19
Tools come and go. In the future, they'll come and go even faster. Don't get married to them; learn what you need and focus on the places where you can really add value. In data science, that means: . - product sense .- user empathy.- data storytelling.
0
4
19
We just launched a podcast about #DataScience jobs and the future of AI! . Awesome conversations with amazing ML engineers and data scientists, at top companies. Let us know what you think 🙌. #MachineLearning @neutronsNeurons @russ_poll @SharpestMindsAI .
2
8
20
@ikennaughanze A big risk in hiring DS people is that they'll treat life like a Kaggle competition, favoring complex models over solving business problems. When you focus on business value instead, you tend to come up with simpler solutions. Visualization & simple models become more important.
1
5
18
Processing time is one of the biggest undiscussed problems with social media. We're *expected* to have Strong Opinions about every piece of news, the moment we see it. By optimizing for speed, we reduce the complexity of our thinking. Nuance dies when thinking becomes reflex.
1
5
19
If you don't have impostor syndrome, maybe you're just not qualified for your job. Look around - everyone else has it. Why don't you have impostor syndrome?. What if they find out you don't have impostor syndrome? . What if they already know??.
1
1
18
I just published my latest piece in @TDataScience on all the problems with data science job postings. but you'll need at least 10 years of experience in deep learning to read it 😉.
0
5
17
New @TDataScience episode is up! @maxjaderberg offered a great introduction to his team's open-ended learning work @DeepMind. Topics. -Procedural game generation.-Learning game theory.-Heuristics vs symbolic logic.-Measuring generalization ability. more!.
0
4
18
I just wrote a post about how building a startup is essentially a machine learning problem. If you're into optimization, and you're into startups, this one's for you.
2
8
18
We should celebrate changing our minds more. It's incredibly hard to do, and unless you think you've *always* had everything figured out, it's a prerequisite for personal growth of any kind.
0
6
17
Don't try to force someone to change their mind *during* an argument. People need time to distance themselves from their previous identities in order to change their minds. "I agree with you now, because time has passed & I'm not the same person I was when I was wrong.".
1
1
19
One of my favourite @TDataScience episodes just dropped. @brianchristian joined me to discuss his new book on the alignment problem!. Topics. - How will transformative AI emerge?.- How might we solve the alignment problem?.- AI's place in cosmic history.
1
7
18
Hey everyone, if you're homeschooling kids because of corona and want someone to speak about. -Kid-level machine learning.-Kid-level quantum mechanics.-Kid-level startup tips. Send me a DM and I'll do my best to set up a ~45 minute slot for it!. Let's get through this together 💪.
2
2
18
Most of my most productive moments don't "look like" they're productive. e.g. - pacing around a room thinking.- casually discussing an important idea with others.- doodling on a notebook page. Performative busywork crowds out true productivity way too often.
0
1
18
Really enjoyed diving into the inner alignment problem with @MIRIBerkeley researcher @EvanHub for the @TDataScience podcast. Topics. -What is inner alignment.-Why it's a hard problem.-Analogies for inner alignment.-Is the alignment effort under-funded?.
0
5
17
Data scientists often ask how they can land roles in consulting companies. Someone did just that on the SharpestMinds Slack recently, and we decided to share that conversation more widely in case others find it useful. Hope it helps! 🙌.
1
5
18
You can get hired with a single, high-quality, end-to-end project. You can’t get hired with 6 jupyter notebooks that use pandas and sklearn to wrangle datasets you found on Kaggle. Quantity makes you blend in. Quality makes you stand out. 1/n.
2
2
18
Most important thing to do when building a data science project:. 👏collect👏. 👏your own👏. 👏data👏. It forces you to show business sense and makes your project unique - both are high on the list of things that get people hired. Here's a great example:.
0
6
17
What questions would you ask someone to help them figure out whether they're best suited for a job in. Data science.Data analytics.Machine learning engineering.Data engineering. or any related area?.
3
7
17
If you're breaking into data science, you might want to see our latest job search webinar featuring @SharpestMindsAI mentee @JohnnyData22!. John managed to land a DS job despite a full-time job, and taking care of 2 kids during COVID. Here's his story:.
2
4
18
No matter what you're trying to pull off, the first step is always to surround yourself with the highest quality people possible.
1
4
17
There's nothing more powerful than a well-networked introvert.
0
3
17
Build data science projects the same way founders build startups:. Be 👏 problem 👏 focused 👏.
1/ It's rare in industry to start with a piece of technology and try to find a way to use it. Yet this is the path many hobby projects take. Especially in data science and machine learning. 👇👇👇.
0
5
17
New @TDataScience episode, ft @divyasiddarth!. A fun 20,000 view of AI research, & whether it's headed in the right direction. Topics. -Why we should prioritize human-cooperative AI.-Taiwan's progress on digital government.-A new framing for AI research.
0
4
16
@AmandaAskell I think the converse is also true: there are a lot of false positives that come from assuming that people with PhDs are capable researchers. In many disciplines, it's possible to earn a PhD through relatively mindless gruntwork.
1
0
16
ML researchers usually avoid discussing concepts like consciousness in the context of AI. Neuroscientists don't feel the same way - so exploring a neuro-phil perspective on AI can be illuminating. My latest @TDataScience podcast with Georg Northoff:.
0
5
16
Books are the only medium I've consistently found can change my long-held views. Podcasts, youtube videos, and TV have the opposite effect. Overwhelmingly short-term, limbic reward.
2
2
15
If you're interested in data visualization/exploration for NLP applications, this is the single best example I've seen in blog form. Bonus: it's from @SharpestMindsAI mentee @blissfulchar!.
1
8
16
When someone asks me for advice on their academic career, it's rare that my advice is anything other than "yeah you need to drop out and do real things".
1
2
15
Really enjoyed recording this episode of the podcast with @rubenharris!. Great listening if you're trying to stay motivated through a career transition in these difficult times. Go get it! 💪.
Learning and looking for jobs in quarantine 🎧 @jeremiecharris and @rubenharris
2
7
14
New @TDataScience episode is out! Really enjoyed speaking with @frossi_t about her AGI thesis. Topics. - What aspects of human cognition should we replicate? .- Are multi-agent models the way to AGI?.- Why AI progress is hard to measure.
0
6
15
Latest @TDataScience podcast is out! This was a fun & offbeat one with the always insightful @anderssandberg. Topics:. -The Fermi paradox and its implications for AI safety.-Why we're probably alone in the universe.-Comparing advanced AIs to human brains.
0
7
15
Today's education system is mostly run by people with reliable salary & benefits, and predictable jobs. Today's job market wants fast-moving, risk-savvy dreamers whose jobs are constantly evolving in totally unpredictable ways. That mismatch might be *the* problem w/ ed today.
0
2
14
In machine learning, an ensemble yields its best predictions when each learner in the ensemble behaves independently. This is true for democracy as well. The more we indulge in other people's narratives, the more correlated and less valuable our opinions become.
We're so bombarded by other people's opinions that the highest leverage thing we can do is step back and take some time to become individuals again. We need to decentralize our cognition.
0
5
16
Really enjoyed this chat with @open_phil's @davidroodman for the @TDataScience podcast. Topics. - Why economic models predict that economic output will become *infinite*.- What AI might do to the human trajectory.- Our history of economic revolutions.
0
5
16