Explorez tous les épisodes du podcast Towards Data Science

Progress in AI has been accelerating dramatically in recent years, and even months. It seems like every other day, there’s a new, previously-believed-to-be-impossible feat of AI that’s achieved by a world-leading lab. And increasingly, these breakthroughs have been driven by the same, simple idea: AI scaling.

For those who haven’t been following the AI scaling sage, scaling means training AI systems with larger models, using increasingly absurd quantities of data and processing power. So far, empirical studies by the world’s top AI labs seem to suggest that scaling is an open-ended process that can lead to more and more capable and intelligent systems, with no clear limit.

And that’s led many people to speculate that scaling might usher in a new era of broadly human-level or even superhuman AI — the holy grail AI researchers have been after for decades.

And while that might sound cool, an AI that can solve general reasoning problems as well as or better than a human might actually be an intrinsically dangerous thing to build.

At least, that’s the conclusion that many AI safety researchers have come to following the publication of a new line of research that explores how modern AI systems tend to solve problems, and whether we should expect more advanced versions of them to perform dangerous behaviours like seeking power.

This line of research in AI safety is called “power-seeking”, and although it’s currently not well understood outside the frontier of AI safety and AI alignment research, it’s starting to draw a lot of attention. The first major theoretical study of power seeking was led by Alex Turner, who’s appeared on the podcast before, and was published in NeurIPS (the world’s top AI conference), for example.

And today, we’ll be hearing from Edouard Harris, an AI alignment researcher and one of my co-founders in the AI safety company (Gladstone AI). Ed’s just completed a significant piece of AI safety research that extends Alex Turner’s original power-seeking work, and that shows what seems to be the first experimental evidence suggesting that we should expect highly advanced AI systems to seek power by default.

What does power seeking really mean though? And does all this imply for the safety of future, general-purpose reasoning systems? That’s what this episode will be all about.

***

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

*** 

Chapters:

- 0:00 Intro

- 4:00 Alex Turner's research

- 7:45 What technology wants

- 11:30 Universal goals

- 17:30 Connecting observations

- 24:00 Micro power seeking behaviour

- 28:15 Ed's research

- 38:00 The human as the environment

- 42:30 What leads to power seeking

- 48:00 Competition as a default outcome

- 52:45 General concern

- 57:30 Wrap-up

It’s no secret that a new generation of powerful and highly scaled language models is taking the world by storm. Companies like OpenAI, AI21Labs, and Cohere have built models so versatile that they’re powering hundreds of new applications, and unlocking entire new markets for AI-generated text.

In light of that, I thought it would be worth exploring the applied side of language modelling — to dive deep into one specific language model-powered tool, to understand what it means to build apps on top of scaled AI systems. How easily can these models be used in the wild? What bottlenecks and challenges do people run into when they try to build apps powered by large language models? That’s what I wanted to find out.

My guest today is Amber Teng, and she’s a data scientist who recently published a blog that got quite a bit of attention, about a resume cover letter generator that she created using GPT-3, OpenAI’s powerful and now-famous language model. I thought her project would be make for a great episode, because it exposes so many of the challenges and opportunities that come with the new era of powerful language models that we’ve just entered.

So today we’ll be exploring exactly that: looking at the applied side of language modelling and prompt engineering, understanding how large language models have made new apps not only possible but also much easier to build, and the likely future of AI-powered products.

***

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

***

Chapters:

- 0:00 Intro

- 2:30 Amber’s background

- 5:30 Using GPT-3

- 14:45 Building prompts up

- 18:15 Prompting best practices

- 21:45 GPT-3 mistakes

- 25:30 Context windows

- 30:00 End-to-end time

- 34:45 The cost of one cover letter

- 37:00 The analytics

- 41:45 Dynamics around company-building

- 46:00 Commoditization of language modelling

- 51:00 Wrap-up

AI scaling has really taken off. Ever since GPT-3 came out, it’s become clear that one of the things we’ll need to do to move beyond narrow AI and towards more generally intelligent systems is going to be to massively scale up the size of our models, the amount of processing power they consume and the amount of data they’re trained on, all at the same time.

That’s led to a huge wave of highly scaled models that are incredibly expensive to train, largely because of their enormous compute budgets. But what if there was a more flexible way to scale AI — one that allowed us to decouple model size from compute budgets, so that we can track a more compute-efficient course to scale?

That’s the promise of so-called mixture of experts models, or MoEs. Unlike more traditional transformers, MoEs don’t update all of their parameters on every training pass. Instead, they route inputs intelligently to sub-models called experts, which can each specialize in different tasks. On a given training pass, only those experts have their parameters updated. The result is a sparse model, a more compute-efficient training process, and a new potential path to scale.

Google has been pushing the frontier of research on MoEs, and my two guests today in particular have been involved in pioneering work on that strategy (among many others!). Liam Fedus and Barrett Zoph are research scientists at Google Brain, and they joined me to talk about AI scaling, sparsity and the present and future of MoE models on this episode of the TDS podcast.

***

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

***

• 2:15 Guests’ backgrounds
• 8:00 Understanding specialization
• 13:45 Speculations for the future
• 21:45 Switch transformer versus dense net
• 27:30 More interpretable models
• 33:30 Assumptions and biology
• 39:15 Wrap-up

Revenues drop unexpectedly, and management pulls aside the data science team into a room. The team is given its marching orders: “your job,” they’re told, “is to find out what the hell is going on with our purchase orders.”

That’s a very open-ended question, of course, because revenues and signups could drop for any number of reasons. Prices may have increased. A new user interface might be confusing potential customers. Seasonality effects might have to be considered. The source of the problem could be, well, anything.

That’s often the position data scientists find themselves in: rather than having a clear A/B test to analyze, they frequently are in the business of combing through user funnels to ensure that each stage is working as expected.

It takes a very detail-oriented and business-savvy team to pull off an investigation with that broad a scope, but that’s exactly what Medium has: a group of product-minded data scientists dedicated to investigating anomalies and identifying growth opportunities hidden in heaps of user data. They were kind enough to chat with me and talk about how Medium does data science for this episode of the Towards Data Science podcast.

If you want to know where data science is heading, it helps to know where it’s been. Very few people have that kind of historical perspective, and even fewer combine it with an understanding of cutting-edge tooling that hints at the direction the field might be taking in the future.

Luckily for us, one of them is Cameron Davidson-Pillon, the former Director of Data Science at Shopify. Cameron has been knee-deep in data science and estimation theory since 2012, when the space was still coming into its own. He’s got a great high-level perspective not only on technical issues but also on hiring and team-building, and he was kind enough to join us for today’s episode of the Towards Data Science podcast.

It’s easy to think of data science as a purely technical discipline: after all, it exists at the intersection of a number of genuinely technical topics, from statistics to programming to machine learning.

But there’s much more to data science and analytics than solving technical problems — and there’s much more to the data science job search than coding challenges and Kaggle competitions as well. Landing a job or a promotion as a data scientist calls on a ton of career skills and soft skills that many people don’t spend nearly enough time honing.

On this episode of the podcast, I spoke with Emily Robinson, an experienced data scientist and blogger with a pedigree that includes Etsy and DataCamp, about career-building strategies. Emily’s got a lot to say about the topic, particularly since she just finished authoring a book entitled “Build a Career in Data Science” with her co-author Jacqueline Nolis. The book explores a lot of great, practical strategies for moving data science careers forward, many of which we discussed during our conversation.

Most of us believe that decisions that affect us should be made rationally: they should be reached by following a reasoning process that combines data we trust with a logic that we find acceptable.

As long as human beings are making these decisions, we can probe at that reasoning to find out whether we agree with it. We can ask why we were denied that bank loan, or why a judge handed down a particular sentence, for example.

Today however, machine learning is automating away more and more of these important decisions, and as a result, our lives are increasingly governed by decision-making processes that we can’t interrogate or understand. Worse, machine learning algorithms can exhibit bias or make serious mistakes, so a black-box-ocracy risks becoming more like a dystopia than even the most imperfect human-designed systems we have today.

That’s why AI ethics and AI safety have drawn so much attention in recent years, and why I was so excited to talk to Alayna Kennedy, a data scientist at IBM whose work is focused on the ethics of machine learning, and the risks associated with ML-based decision-making. Alayna has consulted with key players in the US government’s AI effort, and has expertise applying machine learning in industry as well, through previous work on neural network modelling and fraud detection.

In mid-January, China launched an official investigation into a string of unusual pneumonia cases in Hubei province. Within two months, that cluster of cases would snowball into a full-blown pandemic, with hundreds of thousands — perhaps even millions — of infections worldwide, with the potential to unleash a wave of economic damage not seen since the 1918 Spanish influenza or the Great Depression.

The exponential growth that led us from a few isolated infections to where we are today is profoundly counterintuitive. And it poses many challenges for the epidemiologists who need to pin down the transmission characteristics of the coronavirus, and for the policy makers who must act on their recommendations, and convince a generally complacent public to implement life-saving social distancing measures.

With the coronas in full bloom, I thought now would be a great time to reach out to Jeremy Howard, co-founder of the incredibly popular Fast.ai machine learning education site. Along with his co-founder Rachel Thomas, Jeremy authored a now-viral report outlining a data-driven case for concern regarding the coronavirus.

It’s easy to think of data scientists as “people who explore and model data”. Bur in reality, the job description is much more flexible: your job as a data scientist is to solve problems that people actually have with data.

You’ll notice that I wrote “problems that people actually have” rather than “build models”. It’s relatively rare that the problems people have actually need to be solved using a predictive model. Instead, a good visualization or interactive chart is almost always the first step of the problem-solving process, and can often be the last as well.

And you know who understands visualization strategy really, really well? Plotly, that’s who. Plotly is a company that builds a ton of great open-source visualization, exploration and data infrastructure tools (and some proprietary commercial ones, too). Today, their tooling is being used by over 50 million people worldwide, and they’ve developed a number of tools and libraries that are now industry standard. So you can imagine how excited I was to speak with Plotly co-founder and Chief Product Officer Chris Parmer.

Chris had some great insights to share about data science and analytics tooling, including the future direction he sees the space moving in. But as his job title suggests, he’s also focused on another key characteristic that all great data scientists develop early on: product instinct (AKA: “knowing what to build next”).

Most machine learning models are used in roughly the same way: they take a complex, high-dimensional input (like a data table, an image, or a body of text) and return something very simple (a classification or regression output, or a set of cluster centroids). That makes machine learning ideal for automating repetitive tasks that might historically have been carried out only by humans.

But this strategy may not be the most exciting application of machine learning in the future: increasingly, researchers and even industry players are experimenting with generative models, that produce much more complex outputs like images and text from scratch. These models are effectively carrying out a creative process — and mastering that process hugely widens the scope of what can be accomplished by machines.

My guest today is Xander Steenbrugge, and his focus is on the creative side of machine learning. In addition to consulting with large companies to help them put state-of-the-art machine learning models into production, he’s focused a lot of his work on more philosophical and interdisciplinary questions — including the interaction between art and machine learning. For that reason, our conversation went in an unusually philosophical direction, covering everything from the structure of language, to what makes natural language comprehension more challenging than computer vision, to the emergence of artificial general intelligence, and how all these things connect to the current state of the art in machine learning.

I can’t remember how many times I’ve forgotten something important.

I’m sure it’s a regular occurrence though: I constantly forget valuable life lessons, technical concepts and useful bits of statistical theory. What’s worse, I often forget these things after working bloody hard to learn them, so my forgetfulness is just a giant waste of time and energy.

That’s why I jumped at the chance to chat with Iain Harlow, VP of Science at Cerego — a company that helps businesses build training courses for their employees by optimizing the way information is served to maximize retention and learning outcomes.

Iain knows a lot about learning and has some great insights to share about how you can optimize your own learning, but he’s also got a lot of expertise solving data science problems and hiring data scientists — two things that he focuses on in his work at Cerego. He’s also a veteran of the academic world, and has some interesting observations to share about the difference between research in academia and research in industry.

You train your model. You check its performance with a validation set. You tweak its hyperparameters, engineer some features and repeat. Finally, you try it out on a test set, and it works great!

Problem solved? Well, probably not.

Five years ago, your job as a data scientist might have ended here, but increasingly, the data science life cycle is expanding to include the steps after basic testing. This shouldn’t come as a surprise: now that machine learning models are being used for life-or-death and mission-critical applications, there’s growing pressure on data scientists and machine learning engineers to ensure that effects like feature drift are addressed reliably, that data science experiments are replicable, and that data infrastructure is reliable.

This episode’s guest is Luke Marsden, and he’s made these problems the focus of this work. Luke is the founder and CEO of Dotscience, a data infrastructure startup that’s creating a git-like tool for data science version control. Luke has spent most of his professional life working on infrastructure problems at scale, and has a lot to say about the direction data science and MLOps are heading in.

When I think of the trends I’ve seen in data science over the last few years, perhaps the most significant and hardest to ignore has been the increased focus on deployment and productionization of models. Not all companies need models deployed to production, of course but at those that do, there’s increasing pressure on data science teams to deliver software engineering along with machine learning solutions.

That’s why I wanted to sit down with Adam Waksman, Head of Core Technology at Foursquare. Foursquare is a company built on data and machine learning: they were one of the first fully scaled social media-powered recommendation services that gained real traction, and now help over 50 million people find restaurants and services in countries around the world.

Our conversation covered a lot of ground, from the interaction between software engineering and data science, to what he looks for in new hires, to the future of the field as a whole.

There’s an idea in machine learning that most of the progress we see in AI doesn’t come from new algorithms of model architectures. instead, some argue, progress almost entirely comes from scaling up compute power, datasets and model sizes — and besides those three ingredients, nothing else really matters.

Through that lens the history of AI becomes the history f processing power and compute budgets. And if that turns out to be true, then we might be able to do a decent job of predicting AI progress by studying trends in compute power and their impact on AI development.

And that’s why I wanted to talk to Jaime Sevilla, an independent researcher and AI forecaster, and affiliate researcher at Cambridge University’s Centre for the Study of Existential Risk, where he works on technological forecasting and understanding trends in AI in particular. His work’s been cited in a lot of cool places, including Our World In Data, who used his team’s data to put together an exposé on trends in compute. Jaime joined me to talk about compute trends and AI forecasting on this episode of the TDS podcast.

***

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

*** 

Chapters:

• 2:00 Trends in compute
• 4:30 Transformative AI
• 13:00 Industrial applications
• 19:00 GPT-3 and scaling
• 25:00 The two papers
• 33:00 Biological anchors
• 39:00 Timing of projects
• 43:00 The trade-off
• 47:45 Wrap-up

In this podcast interview, YK (CS Dojo) interviews Chanchal Chatterjee, who’s an AI leader at Google. 

Podcast interview with one of our top data science writers, Will Koehrsen.

Let’s go!  Here’s Will’s article about what he learned from writing a data science article every week for a year: https://towardsdatascience.com/what-i-learned-from-writing-a-data-science-article-every-week-for-a-year-201c0357e0ce

This episode was hosted by YK from CS Dojo: https://www.instagram.com/ykdojo/

Getting hired as a data scientist, machine learning engineer or data analyst is hard. And if there’s one person who’s spent a *lot* of time thinking about why that is, and what you can do about it if you’re trying to break into the field, it’s Edouard Harris.

Ed is the co-founder of SharpestMinds, a data science mentorship program that’s free until you get a job. He also happens to be my brother, which makes this our most nepostistic episode yet.

If there’s one trend that not nearly enough data scientists seem to be paying attention to heading into 2020, it’s this: data scientists are becoming product people.

Five years ago, that wasn’t the case at all: data science and machine learning were all the rage, and managers were impressed by fancy analytics and build over-engineered predictive models. Today, a healthy dose of reality has set in, and most companies see data science as a means to an end: it’s way of improving the experience of real users and real, paying customers, and not a magical tool whose coolness is self-justifying.

At the same time, as more and more tools continue to make it easier and easier for people who aren’t data scientists to build and use predictive models, data scientists are going to have to get good at new things. And that means two things: product instinct, and data storytelling.

That’s why we wanted to chat with Nate Nichols, a data scientist turned VP of Product Architecture at Narrative Science — a company that’s focused on addressing data communication. Nate is also the co-author of Let Your People Be People, a (free) book on data storytelling.

In this podcast episode, Helen Ngo and YK (aka CS Dojo) discuss deep fake, NLP, and women in data science.

In this podcast interview, YK (aka CS Dojo) asks Ian Xiao about why he thinks machine learning is more boring than you may think. 

Original article: https://towardsdatascience.com/data-science-is-boring-1d43473e353e 

The other day, I interviewed Jeremie Harris, a SharpestMinds cofounder, for the Towards Data Science podcast and YouTube channel. SharpestMinds is a startup that helps people who are looking for data science jobs by finding mentors for them.

In my opinion, their system is interesting in a way that a mentor only gets paid when their mentee lands a data science job. I wanted to interview Jeremie because I had previously spoken to him on a different occasion, and I wanted to personally learn more about his story, as well as his thoughts on today’s data science job market.

Hi! It's YK here from CS Dojo. In this episode, I interviewed Jessica Li from Kaggle about how she worked with NASA to predict snowmelt patterns using deep learning. Hope you enjoy!

One question I’ve been getting a lot lately is whether graduate degrees — especially PhDs — are necessary in order to land a job in data science. Of course, education requirements vary widely from company to company, which is why I think the most informative answers to this question tend to come not from recruiters or hiring managers, but from data scientists with those fancy degrees, who can speak to whether they were actually useful.

That’s far from the only reason I wanted to sit down with Rachael Tatman for this episode of the podcast though. In addition to holding a PhD in computational sociolinguistics, Rachael is a data scientist at Kaggle, and a popular livestreaming coder (check out her Twitch stream here). She’s has a lot of great insights about breaking into data science, how to get the most out of Kaggle, the future of NLP, and yes, the value of graduate degrees for data science roles.

One thing that you might not realize if you haven’t worked as a data scientist in very large companies is that the problems that arise at enterprise scale (and well as the skills that are needed to solve them) are completely different from those you’re likely to run into at a startup.

Scale is a great thing for many reasons: it means access to more data sources, and usually more resources for compute and storage. But big companies can take advantage of these things only by fostering successful collaboration between and among large teams (which is really, really hard), and have to contend with unique data sanitation challenges that can’t be addressed without reinventing practically the entire data science life cycle.

So I’d say it’s a good thing we booked Sanjeev Sharma, Vice President of Data Modernization and Strategy at Delphix, for today’s episode. Sanjeev’s specialty is helping huge companies with significant technical debt modernize and upgrade their data pipelines, and he’s seen the ins and outs of data science at enterprise scale for longer than almost anyone.

Generating well-referenced and accurate Wikipedia articles has always been an important problem: Wikipedia has essentially become the Internet's encyclopedia of record, and hundreds of millions of people use it do understand the world.

But over the last decade Wikipedia has also become a critical source of training data for data-hungry text generation models. As a result, any shortcomings in Wikipedia’s content are at risk of being amplified by the text generation tools of the future. If one type of topic or person is chronically under-represented in Wikipedia’s corpus, we can expect generative text models to mirror — or even amplify — that under-representation in their outputs.

Through that lens, the project of Wikipedia article generation is about much more than it seems — it’s quite literally about setting the scene for the language generation systems of the future, and empowering humans to guide those systems in more robust ways.

That’s why I wanted to talk to Meta AI researcher Angela Fan, whose latest project is focused on generating reliable, accurate, and structured Wikipedia articles. She joined me to talk about her work, the implications of high-quality long-form text generation, and the future of human/AI collaboration on this episode of the TDS podcast.

--- 

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

---

• 1:45 Journey into Meta AI
• 5:45 Transition to Wikipedia
• 11:30 How articles are generated
• 18:00 Quality of text
• 21:30 Accuracy metrics
• 25:30 Risk of hallucinated facts
• 30:45 Keeping up with changes
• 36:15 UI/UX problems
• 45:00 Technical cause of gender imbalance
• 51:00 Wrap-up

A few years ago, there really wasn’t much of a difference between data science in theory and in practice: a jupyter notebook and a couple of imports were all you really needed to do meaningful data science work. Today, as the classroom overlaps less and less with the realities of industry, it’s becoming more and more important for data scientists to develop the ability to learn independently and go off the beaten path.

Few people have done so as effectively as Sanyam Bhutani, who among other things is an incoming ML engineer at H2O.ai, a top-1% Kaggler, popular blogger and host of the Chai Time Data Science Podcast. Sanyam has a unique perspective on the mismatch between what’s taught in the classroom and what’s required in industry: he started doing ML contract work while still in undergrad, and has interviewed some of the world’s top-ranked Kagglers to better understand where the rubber meets the data science road. 

The trend towards model deployment, engineering and just generally building “stuff that works” is just the latest step in the evolution of the now-maturing world of data science. It’s almost guaranteed not to be the last one though, and staying ahead of the data science curve means keeping an eye on what trends might be just around the corner. That’s why we asked Ben Lorica, O’Reilly Media’s Chief Data Scientist, to join us on the podcast.

Not only does Ben have a mile-high view of the data science world (he advises about a dozen startups and organizes multiple world-class conferences), but he also has a perspective that spans two decades of data science evolution.

Each week, I have dozens of conversations with people who are trying to break into data science. The main topic of the conversations varies, but it’s rare that I walk away without getting a question like, “Do you think I have a shot in data science given my unusual background in [finance/physics/stats/economics/etc]?”.

From now on, my answer to that question will be to point them to today’s guest, George John Jordan Thomas Aquinas Hayward.

George [names omitted] Hayward’s data science career is a testament to the power of branding and storytelling. After completing a JD/MBA at Stanford and reaching top-ranked status in Hackerrank’s SQL challenges, he went on to work on contract for a startup at Google, and subsequently for a number of other companies. Now, you might be tempted to ask how comedy and law could possibly lead to a data science career.

For today’s podcast, we spoke with someone who is laser-focused on considering this second possibility: the idea that data science is becoming an engineer’s game. Serkan Piantino served as the Director of Engineering for Facebook AI Research, and now runs machine learning infrastructure startup Spell. Their goal is to make dev tools for data scientists that make it as easy to train models on the cloud as it is to train them locally. That experience, combined with his time at Facebook, have given him a unique perspective on the engineering best practices that data scientists should use, and the future of the field as a whole.

I’ve said it before and I’ll say it again: “data science” is an ambiguous job title. People use the term to refer to data science, data engineering, machine learning engineering and analytics roles, and that’s bad enough. But worse still, being a “data scientist” means completely different things depending on the scale and stage of the company you’re working at. A data scientist at a small startup might have almost nothing in common with a data scientist at a massive enterprise company, for example.

So today, we decided to talk to someone who’s seen data science at both scales. Jay Feng started his career working in analytics and data science at Jobr, which was acquired by Monster.com (which was itself acquired by an even bigger company). Among many other things, his story sheds light on a question that you might not have thought about before: what happens to data scientists when their company gets acquired?

Most software development roles are pretty straightforward: someone tells you what to build (usually a product manager), and you build it. What’s interesting about data science is that although it’s a software role, it doesn’t quite follow this rule.

That’s because data scientists are often the only people who can understand the practical business consequences of their work. There’s only one person on the team who can answer questions like, “What does the variance in our cluster analysis tell us about user preferences?” and “ What are the business consequences of our model’s ROC score?”, and that person is the data scientist. In that sense, data scientists have a very important responsibility not to leave any insights on the table, and to bring business instincts to bare even when they’re dealing with deeply technical problems.

For today’s episode, we spoke with Rocio Ng, a data scientist at LinkedIn, about the need for strong partnerships between data scientists and product managers, and the day-to-day dynamic between those roles at LinkedIn. Along the way, we also talked about one of the most common mistakes that early career data scientists make: focusing too much on that first role.

If you’ve been following developments in data science over the last few years, you’ll know that the field has evolved a lot since its Wild West phase in the early/mid 2010s. Back then, a couple of Jupyter notebooks with half-baked modeling projects could land you a job at a respectable company, but things have since changed in a big way.

Today, as companies have finally come to understand the value that data science can bring, more and more emphasis is being placed on the implementation of data science in production systems. And as these implementations have required models that can perform on larger and larger datasets in real-time, an awful lot of data science problems have become engineering problems.

That’s why we sat down with Akshay Singh, who among other things has worked in and managed data science teams at Amazon, League and the Chan-Zuckerberg Initiative (formerly Meta.com).

It’s easy to think of data science as a technical discipline, but in practice, things don’t really work out that way. If you’re going to be a successful data scientist, people will need to believe that you can add value in order to hire you, people will need to believe in your pet project in order to endorse it within your company, and people will need to make decisions based on the insights you pull out of your data.

Although it’s easy to forget about the human element, managing it is one of the most useful skills you can develop if you want to climb the data science ladder, and land that first job, or that promotion you’re after. And that’s exactly why we sat down with Susan Holcomb, the former Head of Data at Pebble, the world’s first smartwatch company.

When Pebble first hired her, Susan was fresh out of grad school in physics, and had never led a team, or interacted with startup executives. As the company grew, she had to figure out how to get Pebble’s leadership to support her effort to push the company in a more data-driven direction, at the same time as she managed a team of data scientists for the first time. 

You import your data. You clean your data. You make your baseline model. 

Then, you tune your hyperparameters. You go back and forth from random forests to XGBoost, add feature selection, and tune some more. Your model’s performance goes up, and up, and up.

And eventually, the thought occurs to you: when do I stop?

Most data scientists struggle with this question on a regular basis, and from what I’ve seen working with SharpestMinds, the vast majority of aspiring data scientists get the answer wrong. That’s why we sat down with Tan Vachiramon, a member of the Spatial AI team Oculus, and former data scientist at Airbnb. 

Tan has seen data science applied in two very different industry settings: once, as part of a team whose job it was to figure out how to understand their customer base in the middle of a the whirlwind of out-of-control user growth (at Airbnb); and again in a context where he’s had the luxury of conducting far more rigorous data science experiments under controlled circumstances (at Oculus). 

My biggest take-home from our conversation was this: if you’re interested in working at a company, it’s worth taking some time to think about their business context, because that’s the single most important factor driving the kind of data science you’ll be doing there. Specifically:

• Data science at rapidly growing companies comes with a special kind of challenge that’s not immediately obvious: because they’re growing so fast, no matter where you look, everything looks like it’s correlated with growth! New referral campaign? “That definitely made the numbers go up!” New user onboarding strategy? “Wow, that worked so well!”. Because the product is taking off, you need special strategies to ensure that you don’t confuse the effectiveness of a company initiative you’re interested in with the inherent viral growth that the product was already experiencing. 
• The amount of time you spend tuning or selecting your model, or doing feature selection, entirely depends on the business context. In some companies (like Airbnb in the early days), super-accurate algorithms aren’t as valuable as algorithms that allow you to understand what the heck is going on in your dataset. As long as business decisions don’t depend on getting second-digit-after-the-decimal levels of accuracy, it’s okay (and even critical) to build a quick model and move on. In these cases, even logistic regression often does the trick!
• In other contexts, where tens of millions of dollars depend on every decimal point of accuracy you can squeeze out of your model (investment banking, ad optimization), expect to spend more time on tuning/modeling. At the end of the day, it’s a question of opportunity costs: keep asking yourself if you could be creating more value for the business if you wrapped up your model tuning now, to work on something else. If you think the answer could be yes, then consider calling model.save() and walking away.

To most data scientists, the jupyter notebook is a staple tool: it’s where they learned the ropes, it’s where they go to prototype models or explore their data — basically, it’s the default arena for their all their data science work. 

But Joel Grus isn’t like most data scientists: he’s a former hedge fund manager and former Googler, and author of Data Science From Scratch. He currently works as a research engineer at the Allen Institute for Artificial Intelligence, and maintains a very active Twitter account. 

Oh, and he thinks you should stop using Jupyter noteoboks. Now. 

When you ask him why, he’ll provide many reasons, but a handful really stand out:

• Hidden state: let’s say you define a variable like a = 1 in the first cell of your notebook. In a later cell, you assign it a new value, say a = 3 . This results is fairly predictable behavior as long as you run your notebook in order, from top to bottom. But if you don’t—or worse still, if you run the a = 3 cell and delete it later — it can be hard, or impossible to know from a simple inspection of the notebook what the true state of your variables is. 
• Replicability: one of the most important things to do to ensure that you’re running repeatable data science experiments is to write robust, modular code. Jupyter notebooks implicitly discourage this, because they’re not designed to be modularized (awkward hacks do allow you to import one notebook into another, but they’re, well, awkward). What’s more, to reproduce another person’s results, you need to first reproduce the environment in which their code was run. Vanilla notebooks don’t give you a good way to do that. 
• Bad for teaching: Jupyter notebooks make it very easy to write terrible tutorials — you know, the kind where you mindlessly hit “shift-enter” a whole bunch of times, and make your computer do a bunch of stuff that you don’t actually understand? It leads to a lot of frustrated learners, or even worse, a lot of beginners who think they understand how to code, but actually don’t.

Overall, Joel’s objections to Jupyter notebooks seem to come in large part from his somewhat philosophical view that data scientists should follow the same set of best practices that any good software engineers would. For instance, Joel stresses the importance of writing unit tests (even for data science code), and is a strong proponent of using type annotation (if you aren’t familiar with that, you should definitely learn about it here). 

But even Joel thinks Jupyter notebooks have a place in data science: if you’re poking around at a pandas dataframe to do some basic exploratory data analysis, it’s hard to think of a better way to produce helpful plots on the fly than the trusty ol’ Jupyter notebook. 

Whatever side of the Jupyter debate you’re on, it’s hard to deny that Joel makes some compelling points. I’m not personally shutting down my Jupyter kernel just yet, but I’m guessing I’ll be firing up my favorite IDE a bit more often in the future.

Trustworthy AI is one of today’s most popular buzzwords. But although everyone seems to agree that we want AI to be trustworthy, definitions of trustworthiness are often fuzzy or inadequate. Maybe that shouldn’t be surprising: it’s hard to come up with a single set of standards that add up to “trustworthiness”, and that apply just as well to a Netflix movie recommendation as a self-driving car.

So maybe trustworthy AI needs to be thought of in a more nuanced way — one that reflects the intricacies of individual AI use cases. If that’s true, then new questions come up: who gets to define trustworthiness, and who bears responsibility when a lack of trustworthiness leads to harms like AI accidents, or undesired biases?

Through that lens, trustworthiness becomes a problem not just for algorithms, but for organizations. And that’s exactly the case that Beena Ammanath makes in her upcoming book, Trustworthy AI, which explores AI trustworthiness from a practical perspective, looking at what concrete steps companies can take to make their in-house AI work safer, better and more reliable. Beena joined me to talk about defining trustworthiness, explainability and robustness in AI, as well as the future of AI regulation and self-regulation on this episode of the TDS podcast.

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

• 1:55 Background and trustworthy AI
• 7:30 Incentives to work on capabilities
• 13:40 Regulation at the level of application domain
• 16:45 Bridging the gap
• 23:30 Level of cognition offloaded to the AI
• 25:45 What is trustworthy AI?
• 34:00 Examples of robustness failures
• 36:45 Team diversity
• 40:15 Smaller companies
• 43:00 Application of best practices
• 46:30 Wrap-up

Until recently, very few people were paying attention to the potential malicious applications of AI. And that made some sense: in an era where AIs were narrow and had to be purpose-built for every application, you’d need an entire research team to develop AI tools for malicious applications. Since it’s more profitable (and safer) for that kind of talent to work in the legal economy, AI didn’t offer much low-hanging fruit for malicious actors.

But today, that’s all changing. As AI becomes more flexible and general, the link between the purpose for which an AI was built and its potential downstream applications has all but disappeared. Large language models can be trained to perform valuable tasks, like supporting writers, translating between languages, or write better code. But a system that can write an essay can also write a fake news article, or power an army of humanlike text-generating bots.

More than any other moment in the history of AI, the move to scaled, general-purpose foundation models has shown how AI can be a double-edged sword. And now that these models exist, we have to come to terms with them, and figure out how to build societies that remain stable in the face of compelling AI-generated content, and increasingly accessible AI-powered tools with malicious use potential.

That’s why I wanted to speak with Katya Sedova, a former Congressional Fellow and Microsoft alumna who now works at Georgetown University’s Center for Security and Emerging Technology, where she recently co-authored some fascinating work exploring current and likely future malicious uses of AI. If you like this conversation I’d really recommend checking out her team’s latest report — it’s called “AI and the future of disinformation campaigns”.

Katya joined me to talk about malicious AI-powered chatbots, fake news generation and the future of AI-augmented influence campaigns on this episode of the TDS podcast.

***

Intro music:

➞ Artist: Ron Gelinas

➞ Track Title: Daybreak Chill Blend (original mix)

➞ Link to Track: https://youtu.be/d8Y2sKIgFWc

*** 

• 2:40 Malicious uses of AI
• 4:30 Last 10 years in the field
• 7:50 Low handing fruit of automation
• 14:30 Other analytics functions
• 25:30 Authentic bots
• 30:00 Influences of service businesses
• 36:00 Race to the bottom
• 42:30 Automation of systems
• 50:00 Manufacturing norms
• 52:30 Interdisciplinary conversations
• 54:00 Wrap-up

Imagine, for example, an AI that’s trained to identify cows in images. Ideally, we’d want it to learn to detect cows based on their shape and colour. But what if the cow pictures we put in the training dataset always show cows standing on grass?

In that case, we have a spurious correlation between grass and cows, and if we’re not careful, our AI might learn to become a grass detector rather than a cow detector. Even worse, we could only realize that’s happened once we’ve deployed it in the real world and it runs into a cow that isn’t standing on grass for the first time.

So how do you build AI systems that can learn robust, general concepts that remain valid outside the context of their training data?

That’s the problem of out-of-distribution generalization, and it’s a central part of the research agenda of Irina Rish, a core member of the Mila— Quebec AI Research institute, and the Canadian Excellence Research Chair in Autonomous AI. Irina’s research explores many different strategies that aim to overcome the out-of-distribution problem, from empirical AI scaling efforts to more theoretical work, and she joined me to talk about just that on this episode of the podcast.

***

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

***

• 2:00 Research, safety, and generalization
• 8:20 Invariant risk minimization
• 15:00 Importance of scaling
• 21:35 Role of language
• 27:40 AGI and scaling
• 32:30 GPT versus ResNet 50
• 37:00 Potential revolutions in architecture
• 42:30 Inductive bias aspect
• 46:00 New risks
• 49:30 Wrap-up

Google the phrase “AI over-hyped”, and you’ll find literally dozens of articles from the likes of Forbes, Wired, and Scientific American, all arguing that “AI isn’t really as impressive at it seems from the outside,” and “we still have a long way to go before we come up with *true* AI, don’t you know.”

Amusingly, despite the universality of the “AI is over-hyped” narrative, the statement that “We haven’t made as much progress in AI as you might think™️” is often framed as somehow being an edgy, contrarian thing to believe.

All that pressure not to over-hype AI research really gets to people — researchers included. And they adjust their behaviour accordingly: they over-hedge their claims, cite outdated and since-resolved failure modes of AI systems, and generally avoid drawing straight lines between points that clearly show AI progress exploding across the board. All, presumably, to avoid being perceived as AI over-hypers.

Why does this matter? Well for one, under-hyping AI allows us to stay asleep — to delay answering many of the fundamental societal questions that come up when widespread automation of labour is on the table. But perhaps more importantly, it reduces the perceived urgency of addressing critical problems in AI safety and AI alignment.

Yes, we need to be careful that we’re not over-hyping AI. “AI startups” that don’t use AI are a problem. Predictions that artificial general intelligence is almost certainly a year away are a problem. Confidently prophesying major breakthroughs over short timescales absolutely does harm the credibility of the field.

But at the same time, we can’t let ourselves be so cautious that we’re not accurately communicating the true extent of AI’s progress and potential. So what’s the right balance?

That’s where Sam Bowman comes in. Sam is a professor at NYU, where he does research on AI and language modeling. But most important for today’s purposes, he’s the author of a paper titled, “When combating AI hype, proceed with caution,” in which he explores a trend he calls under-claiming — a common practice among researchers that consists of under-stating the extent of current AI capabilities, and over-emphasizing failure modes in ways that can be (unintentionally) deceptive.

Sam joined me to talk about under-claiming and what it means for AI progress on this episode of the Towards Data Science podcast.

***

Intro music: 

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc 

***

• 2:15 Overview of the paper
• 8:50 Disappointing systems
• 13:05 Potential double standard
• 19:00 Moving away from multi-modality
• 23:50 Overall implications
• 28:15 Pressure to publish or perish
• 32:00 Announcement discrepancies
• 36:15 Policy angle
• 41:00 Recommendations
• 47:20 Wrap-up

It’s no secret that AI systems are being used in more and more high-stakes applications. As AI eats the world, it’s becoming critical to ensure that AI systems behave robustly — that they don’t get thrown off by unusual inputs, and start spitting out harmful predictions or recommending dangerous courses of action. If we’re going to have AI drive us to work, or decide who gets bank loans and who doesn’t, we’d better be confident that our AI systems aren’t going to fail because of a freak blizzard, or because some intern missed a minus sign.

We’re now past the point where companies can afford to treat AI development like a glorified Kaggle competition, in which the only thing that matters is how well models perform on a testing set. AI-powered screw-ups aren’t always life-or-death issues, but they can harm real users, and cause brand damage to companies that don’t anticipate them.

Fortunately, AI risk is starting to get more attention these days, and new companies — like Robust Intelligence — are stepping up to develop strategies that anticipate AI failures, and mitigate their effects. Joining me for this episode of the podcast was Yaron Singer, a former Googler, professor of computer science and applied math at Harvard, and now CEO and co-founder of Robust Intelligence. Yaron has the rare combination of theoretical and engineering expertise required to understand what AI risk is, and the product intuition to know how to integrate that understanding into solutions that can help developers and companies deal with AI risk.

--- 

Intro music:

➞ Artist: Ron Gelinas

➞ Track Title: Daybreak Chill Blend (original mix)

➞ Link to Track: https://youtu.be/d8Y2sKIgFWc

--- 

• 0:00 Intro
• 2:30 Journey into AI risk
• 5:20 Guarantees of AI systems
• 11:00 Testing as a solution
• 15:20 Generality and software versus custom work
• 18:55 Consistency across model types
• 24:40 Different model failures
• 30:25 Levels of responsibility
• 35:00 Wrap-up

Until very recently, the study of human disease involved looking at big things — like organs or macroscopic systems — and figuring out when and how they can stop working properly. But that’s all started to change: in recent decades, new techniques have allowed us to look at disease in a much more detailed way, by examining the behaviour and characteristics of single cells.

One class of those techniques now known as single-cell genomics — the study of gene expression and function at the level of single cells. Single-cell genomics is creating new, high-dimensional datasets consisting of tens of millions of cells whose gene expression profiles and other characteristics have been painstakingly measured. And these datasets are opening up exciting new opportunities for AI-powered drug discovery — opportunities that startups are now starting to tackle head-on.

Joining me for today’s episode is Tali Raveh, Senior Director of Computational Biology at Immunai, a startup that’s using single-cell level data to perform high resolution profiling of the immune system at industrial scale. Tali joined me to talk about what makes the immune system such an exciting frontier for modern medicine, and how single-cell data and AI might be poised to generate unprecedented breakthroughs in disease treatment on this episode of the TDS podcast.

---

Intro music:

➞ Artist: Ron Gelinas

➞ Track Title: Daybreak Chill Blend (original mix)

➞ Link to Track: https://youtu.be/d8Y2sKIgFWc

--- 

Chapters:

0:00 Intro

2:00 Tali’s background

4:00 Immune systems and modern medicine

14:40 Data collection technology

19:00 Exposing cells to different drugs

24:00 Labeled and unlabelled data

27:30 Dataset status

31:30 Recent algorithmic advances

36:00 Cancer and immunology

40:00 The next few years

41:30 Wrap-up

If you were scrolling through your newsfeed in late September 2021, you may have caught this splashy headline from The Times of London that read, “Can this man save the world from artificial intelligence?”. The man in question was Mo Gawdat, an entrepreneur and senior tech executive who spent several years as the Chief Business Officer at GoogleX (now called X Development), Google’s semi-secret research facility, that experiments with moonshot projects like self-driving cars, flying vehicles, and geothermal energy. At X, Mo was exposed to the absolute cutting edge of many fields — one of which was AI. His experience seeing AI systems learn and interact with the world raised red flags for him — hints of the potentially disastrous failure modes of the AI systems we might just end up with if we don’t get our act together now.

Mo writes about his experience as an insider at one of the world’s most secretive research labs and how it led him to worry about AI risk, but also about AI’s promise and potential in his new book, Scary Smart: The Future of Artificial Intelligence and How You Can Save Our World. He joined me to talk about just that on this episode of the TDS podcast.

Imagine you’re a big hedge fund, and you want to go out and buy yourself some data. Data is really valuable for you — it’s literally going to shape your investment decisions and determine your outcomes.

But the moment you receive your data, a cold chill runs down your spine: how do you know your data supplier gave you the data they said they would? From your perspective, you’re staring down 100,000 rows in a spreadsheet, with no way to tell if half of them were made up — or maybe more for that matter.

This might seem like an obvious problem in hindsight, but it’s one most of us haven’t even thought of. We tend to assume that data is data, and that 100,000 rows in a spreadsheet is 100,000 legitimate samples.

The challenge of making sure you’re dealing with high-quality data, or at least that you have the data you think you do, is called data observability, and it’s surprisingly difficult to solve for at scale. In fact, there are now entire companies that specialize in exactly that — one of which is Zectonal, whose co-founder Dave Hirko will be joining us for today’s episode of the podcast.

Dave has spent his career understanding how to evaluate and monitor data at massive scale. He did that first at AWS in the early days of cloud computing, and now through Zectonal, where he’s working on strategies that allow companies to detect issues with their data — whether they’re caused by intentional data poisoning, or unintentional data quality problems. Dave joined me to talk about data observability, data as a new vector for cyberattacks, and the future of enterprise data management on this episode of the TDS podcast.

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

• 0:00 Intro
• 3:00 What is data observability?
• 10:45 “Funny business” with data providers
• 12:50 Data supply chains
• 16:50 Various cybersecurity implications
• 20:30 Deep data inspection
• 27:20 Observed direction of change
• 34:00 Steps the average person can take
• 41:15 Challenges with GDPR transitions
• 48:45 Wrap-up

Today’s episode is somewhat special, because we’re going to be talking about what might be the first solid quantitative study of the power-seeking tendencies that we can expect advanced AI systems to have in the future.

For a long time, there’s kind of been this debate in the AI safety world, between:

• People who worry that powerful AIs could eventually displace, or even eliminate humanity altogether as they find more clever, creative and dangerous ways to optimize their reward metrics on the one hand, and
• People who say that’s Terminator-bating Hollywood nonsense that anthropomorphizes machines in a way that’s unhelpful and misleading.

Unfortunately, recent work in AI alignment — and in particular, a spotlighted 2021 NeurIPS paper — suggests that the AI takeover argument might be stronger than many had realized. In fact, it’s starting to look like we ought to expect to see power-seeking behaviours from highly capable AI systems by default. These behaviours include things like AI systems preventing us from shutting them down, repurposing resources in pathological ways to serve their objectives, and even in the limit, generating catastrophes that would put humanity at risk.

As concerning as these possibilities might be, it’s exciting that we’re starting to develop a more robust and quantitative language to describe AI failures and power-seeking. That’s why I was so excited to sit down with AI researcher Alex Turner, the author of the spotlighted NeurIPS paper on power-seeking, and discuss his path into AI safety, his research agenda and his perspective on the future of AI on this episode of the TDS podcast.

***

Intro music:

➞ Artist: Ron Gelinas

➞ Track Title: Daybreak Chill Blend (original mix)

➞ Link to Track: https://youtu.be/d8Y2sKIgFWc

***

Chapters: 

- 2:05 Interest in alignment research

- 8:00 Two camps of alignment research

- 13:10 The NeurIPS paper

- 17:10 Optimal policies

- 25:00 Two-piece argument

- 28:30 Relaxing certain assumptions

- 32:45 Objections to the paper

- 39:00 Broader sense of optimization

- 46:35 Wrap-up

Until recently, AI systems have been narrow — they’ve only been able to perform the specific tasks that they were explicitly trained for. And while narrow systems are clearly useful, the holy grain of AI is to build more flexible, general systems.

But that can’t be done without good performance metrics that we can optimize for — or that we can at least use to measure generalization ability. Somehow, we need to figure out what number needs to go up in order to bring us closer to generally-capable agents. That’s the question we’ll be exploring on this episode of the podcast, with Danijar Hafner. Danijar is a PhD student in artificial intelligence at the University of Toronto with Jimmy Ba and Geoffrey Hinton and researcher at Google Brain and the Vector Institute.

Danijar has been studying the problem of performance measurement and benchmarking for RL agents with generalization abilities. As part of that work, he recently released Crafter, a tool that can procedurally generate complex environments that are a lot like Minecraft, featuring resources that need to be collected, tools that can be developed, and enemies who need to be avoided or defeated. In order to succeed in a Crafter environment, agents need to robustly plan, explore and test different strategies, which allow them to unlock certain in-game achievements.

Crafter is part of a growing set of strategies that researchers are exploring to figure out how we can benchmark and measure the performance of general-purpose AIs, and it also tells us something interesting about the state of AI: increasingly, our ability to define tasks that require the right kind of generalization abilities is becoming just as important as innovating on AI model architectures. Danijar joined me to talk about Crafter, reinforcement learning, and the big challenges facing AI researchers as they work towards general intelligence on this episode of the TDS podcast.

***

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

***

• 0:00 Intro
• 2:25 Measuring generalization
• 5:40 What is Crafter?
• 11:10 Differences between Crafter and Minecraft
• 20:10 Agent behavior
• 25:30 Merging scaled models and reinforcement learning
• 29:30 Data efficiency
• 38:00 Hierarchical learning
• 43:20 Human-level systems
• 48:40 Cultural overlap
• 49:50 Wrap-up

2021 has been a wild ride in many ways, but its wildest features might actually be AI-related. We’ve seen major advances in everything from language modeling to multi-modal learning, open-ended learning and even AI alignment.

So, we thought, what better way to take stock of the big AI-related milestones we’ve reached in 2021 than a cross-over episode with our friends over at the Last Week In AI podcast.

***

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

***

Chapters:

• 0:00 Intro
• 2:15 Rise of multi-modal models
• 7:40 Growth of hardware and compute
• 13:20 Reinforcement learning
• 20:45 Open-ended learning
• 26:15 Power seeking paper
• 32:30 Safety and assumptions
• 35:20 Intrinsic vs. extrinsic motivation
• 42:00 Mapping natural language
• 46:20 Timnit Gebru’s research institute
• 49:20 Wrap-up

Imagine for a minute that you’re running a profitable business, and that part of your sales strategy is to send the occasional mass email to people who’ve signed up to be on your mailing list. For a while, this approach leads to a reliable flow of new sales, but then one day, that abruptly stops. What happened?

You pour over logs, looking for an explanation, but it turns out that the problem wasn’t with your software; it was with your data. Maybe the new intern accidentally added a character to every email address in your dataset, or shuffled the names on your mailing list so that Christina got a message addressed to “John”, or vice-versa. Versions of this story happen surprisingly often, and when they happen, the cost can be significant: lost revenue, disappointed customers, or worse — an irreversible loss of trust.

Today, entire products are being built on top of datasets that aren’t monitored properly for critical failures — and an increasing number of those products are operating in high-stakes situations. That’s why data observability is so important: the ability to  track the origin, transformations and characteristics of mission-critical data to detect problems before they lead to downstream harm.

And it’s also why we’ll be talking to Kevin Hu, the co-founder and CEO of Metaplane, one of the world’s first data observability startups. Kevin has a deep understanding of data pipelines, and the problems that cap pop up if you they aren’t properly monitored. He joined me to talk about data observability, why it matters, and how it might be connected to responsible AI on this episode of the TDS podcast.

Intro music:

➞ Artist: Ron Gelinas

➞ Track Title: Daybreak Chill Blend (original mix)

➞ Link to Track: https://youtu.be/d8Y2sKIgFWc 0:00

Chapters: 

• 0:00 Intro
• 2:00 What is data observability?
• 8:20 Difference between a dataset’s internal and external characteristics
• 12:20 Why is data so difficult to log?
• 17:15 Tracing back models
• 22:00 Algorithmic analyzation of a date
• 26:30 Data ops in five years
• 33:20 Relation to cutting-edge AI work
• 39:25 Software engineering and startup funding
• 42:05 Problems on a smaller scale
• 46:40 Future data ops problems to solve
• 48:45 Wrap-up

Historically, AI systems have been slow learners. For example, a computer vision model often needs to see tens of thousands of hand-written digits before it can tell a 1 apart from a 3. Even game-playing AIs like DeepMind’s AlphaGo, or its more recent descendant MuZero, need far more experience than humans do to master a given game.

So when someone develops an algorithm that can reach human-level performance at anything as fast as a human can, it’s a big deal. And that’s exactly why I asked Yang Gao to join me on this episode of the podcast. Yang is an AI researcher with affiliations at Berkeley and Tsinghua University, who recently co-authored a paper introducing EfficientZero: a reinforcement learning system that learned to play Atari games at the human-level after just two hours of in-game experience. It’s a tremendous breakthrough in sample-efficiency, and a major milestone in the development of more general and flexible AI systems.

--- 

Intro music:

➞ Artist: Ron Gelinas

➞ Track Title: Daybreak Chill Blend (original mix)

➞ Link to Track: https://youtu.be/d8Y2sKIgFWc

---

Chapters: 

- 0:00 Intro

- 1:50 Yang’s background

- 6:00 MuZero’s activity

- 13:25 MuZero to EfficiantZero

- 19:00 Sample efficiency comparison

- 23:40 Leveraging algorithmic tweaks

- 27:10 Importance of evolution to human brains and AI systems

- 35:10 Human-level sample efficiency

- 38:28 Existential risk from AI in China

- 47:30 Evolution and language

- 49:40 Wrap-up

There once was a time when AI researchers could expect to read every new paper published in the field on the arXiv, but today, that’s no longer the case. The recent explosion of research activity in AI has turned keeping up to date with new developments into a full-time job.

Fortunately, people like YouTuber, ML PhD and sunglasses enthusiast Yannic Kilcher make it their business to distill ML news and papers into a digestible form for mortals like you and me to consume. I highly recommend his channel to any TDS podcast listeners who are interested in ML research — it’s a fantastic resource, and literally the way I finally managed to understand the Attention is All You Need paper back in the day.

Yannic is joined me to talk about what he’s learned from years of following, reporting and doing AI research, including the trends, the challenges and the opportunities that he expects are going to shape the course of AI history in coming years.

--- 

Intro music:

➞ Artist: Ron Gelinas

➞ Track Title: Daybreak Chill Blend (original mix)

➞ Link to Track: https://youtu.be/d8Y2sKIgFWc

--- 

Chapters:

- 0:00 Intro

- 1:20 Yannic’s path into ML

- 7:25 Selecting ML news

- 11:45 AI ethics → political discourse

- 17:30 AI alignment

- 24:15 Malicious uses

- 32:10 Impacts on persona

- 39:50 Bringing in human thought

- 46:45 Math with big numbers

- 51:05 Metrics for generalization

- 58:05 The future of AI

- 1:02:58 Wrap-up

Today, most machine learning algorithms use the same paradigm: set an objective, and train an agent, a neural net, or a classical model to perform well against that objective. That approach has given good results: these types of AI can hear, speak, write, read, draw, drive and more.

But they’re also inherently limited: because they optimize for objectives that seem interesting to humans, they often avoid regions of parameter space that are valuable, but that don’t immediately seem interesting to human beings, or the objective functions we set. That poses a challenge for researchers like Ken Stanley, whose goal is to build broadly superintelligent AIs — intelligent systems that outperform humans at a wide range of tasks. Among other things, Ken is a former startup founder and AI researcher, whose career has included work in academia, at UberAI labs, and most recently at OpenAI, where he leads the open-ended learning team.

Ken joined me to talk about his 2015 book Greatness Cannot Be Planned: The Myth of the Objective, what open-endedness could mean for humanity, the future of intelligence, and even AI safety on this episode of the TDS podcast.