This episode is a follow up to our recent Fully Connected show discussing federated learning. In that previous discussion, we mentioned Flower (a “friendly” federated learning framework). Well, one of the creators of Flower, Daniel Beutel, agreed to join us on the show to discuss the project (and federated learning more broadly)! The result is a really interesting and motivating discussion of ML, privacy, distributed training, and open source AI.
Recently, GitHub released Copilot, which is an amazing AI pair programmer powered by OpenAI’s Codex model. In this episode, Natalie Pistunovich tells us all about Codex and helps us understand where it fits in our development workflow. We also discuss MLOps and how AI is influencing software engineering more generally.
In this Fully-Connected episode, Daniel and Chris ponder whether in-person AI conferences are on the verge of making a post-pandemic comeback. Then on to BigScience from Hugging Face, a year-long research workshop on large multilingual models and datasets. Specifically they dive into the T0, a series of natural language processing (NLP) AI models specifically trained for researching zero-shot multitask learning. Daniel provides a brief tour of the possible with the T0 family. They finish up with a couple of new learning resources.
Each year we discuss the latest insights from the Stanford Institute for Human-Centered Artificial Intelligence (HAI), and this year is no different. Daniel and Chris delve into key findings and discuss in this Fully-Connected episode. They also check out a study called ‘Delphi: Towards Machine Ethics and Norms’, about how to integrate ethics and morals into AI models.
There are a lot of people trying to innovate in the area of specialized AI hardware, but most of them are doing it with traditional transistors. Lightmatter is doing something totally different. They’re building photonic computers that are more power efficient and faster for AI inference. Nick Harris joins us in this episode to bring us up to speed on all the details.
When is the last time you had a eureka moment? Chris had a chat with Nicholas Mohnacky, CEO and Cofounder of bundleIQ, where they use natural language processing algorithms like GPT-3 to connect your Google GSuite with other personal data sources to find deeper connections, go beyond the obvious, and create eureka moments.
It surprises me that when people think of “software that brings about the singularity” they think of text models, or of RL agents. But they sneer at decision tree boosting and the like as boring algorithms for boring problems.
To me, this seems counter-intuitive, and the fact that most people researching ML are interested in subjects like vision and language is flabbergasting. For one, because getting anywhere productive in these fields is really hard, for another, because their usefulness seems relatively minimal.
This is the first episode in a special series we are calling the “Spotlight on AI in Africa”. To kick things off, Joyce and Mutembesa from Makerere University’s AI Lab join us to talk about their amazing work in computer vision, natural language processing, and data collection. Their lab seeks out problems that matter in African communities, pairs those problems with appropriate data/tools, and works with the end users to ensure that solutions create real value.
Federated learning is increasingly practical for machine learning developers because of the challenges we face with model and data privacy. In this fully connected episode, Chris and Daniel dive into the topic and dissect the ideas behind federated learning, practicalities of implementing decentralized training, and current uses of the technique.
Tivadar Danka is an educator and content creator in the machine learning space, and he is writing a book to help practitioners go from high school mathematics to mathematics of neural networks. His explanations are lucid and easy to understand. You have never had such a fun and interesting conversation about calculus, linear algebra, and probability theory before!
Polarity Mapping is a framework to “help problems be solved in a realistic and multidimensional manner” (see here for more info). In this week’s fully connected episode, Chris and Daniel use this framework to help them discuss how an organization can strike a good balance between human intelligence and AI. AI can’t solve everything and humans need to be in-the-loop with many AI solutions.
Eugene Yan, in a post titled The first rule of machine learning: start without machine learning:
Applying machine learning effectively is tricky. You need data. You need a robust pipeline to support your data flows. And most of all, you need high-quality labels. As a result, most of the time, my first iteration doesn’t involve machine learning at all.
Eugene is stating the obvious with this post, but hey sometimes you just gotta state it. What’s even more interesting to me is how nicely the format generalizes! Let’s pattern match this sucker:
The first rule of X: start without X
Now, apply the pattern a few times and see if it holds:
- The first rule of Kubernetes: start without Kubernetes
- The first rule of goroutines: start without goroutines
- The first rule of coding: start without coding
Yeah, that abstraction holds pretty true. Surely there will be cases where it falls flat on its face, though. Can you think of any examples?
As you start developing an AI/ML based solution, you quickly figure out that you need to run workflows. Not only that, you might need to run those workflows across various kinds of infrastructure (including GPUs) at scale. Ville Tuulos developed Metaflow while working at Netflix to help data scientists scale their work. In this episode, Ville tells us a bit more about Metaflow, his new book on data science infrastructure, and his approach to helping scale ML/AI work.
Any AI play that lacks an underlying data strategy is doomed to fail, and a big part of any data strategy is labeling. Michael, from Label Studio, joins us in this episode to discuss how the industry’s perception of data labeling is shifting. We cover open source tooling, validating labels, and integrating ML/AI models in the labeling loop.
Yonatan Geifman of Deci makes Daniel and Chris buckle up, and takes them on a tour of the ideas behind his amazing new inference platform. It enables AI developers to build, optimize, and deploy blazing-fast deep learning models on any hardware. Don’t blink or you’ll miss it!
The news is in the headline on this one, but here’s a bit more meat from the article:
Using rigorous and detailed scientific analysis, the study concluded that upon testing 1,692 programs generated in 89 different code-completion scenarios, 40 percent were found to be vulnerable.
In this episode, Peter Wang from Anaconda joins us again to go over their latest “State of Data Science” survey. The updated results include some insights related to data science work during COVID along with other topics including AutoML and model bias. Peter also tells us a bit about the exciting new partnership between Anaconda and Pyston (a fork of the standard CPython interpreter which has been extensively enhanced to improve the execution performance of most Python programs).
Jina calls itself a “cloud-native neural search framework”. What is neural search, exactly?
The core idea of neural search is to leverage state-of-the-art deep neural networks to build every component of a search system. In short, neural search is deep neural network-powered information retrieval. In academia, it’s often called neural IR.
And what can it do for you?
Thanks to recent advances in deep neural networks, a neural search system can go way beyond simple text search. It enables advanced intelligence on all kinds of unstructured data, such as images, audio, video, PDF, 3D mesh, you name it.
For example, retrieving animation according to some beats; finding the best-fit memes according to some jokes; scanning a table with your iPhone’s LiDAR camera and finding similar furniture at IKEA. Neural search systems enable what traditional search can’t: multi/cross-modal data retrieval.
This project looks quite established and collaborative. 172 contributors and counting…
Codex is a descendant of GPT-3 – its training data contains both natural language and billions of lines of source code from publicly available sources, including code in public GitHub repositories.
“We see this as a tool to multiply programmers,” OpenAI’s CTO and co-founder Greg Brockman told The Verge. “Programming has two parts to it: you have ‘think hard about a problem and try to understand it,’ and ‘map those small pieces to existing code, whether it’s a library, a function, or an API.’” The second part is tedious, he says, but it’s what Codex is best at. “It takes people who are already programmers and removes the drudge work.”
We’re back with another Fully Connected episode – Daniel and Chris dive into a series of articles called ‘A New AI Lexicon’ that collectively explore alternate narratives, positionalities, and understandings to the better known and widely circulated ways of talking about AI. The fun begins early as they discuss and debate ‘An Electric Brain’ with strong opinions, and consider viewpoints that aren’t always popular.
That’s a big addition. Here’s what Hillary Juma (Common Voice’s community mgr) had to say about it:
Internet access is increasingly mediated through speech: Voice assistants and smart speakers give us directions, search for information, connect us to friends, used in assistive technology and much more. Yet this technology doesn’t work for millions of people. For example, neither Amazon’s Alexa, Apple’s Siri, nor Google Home support a single native African language.
By giving individuals the ability to share their speech, we can help ensure all communities have access to voice technology and the opportunity it unlocks.
What a great initiative! (I first heard about Common Voice on Practical AI.)
We’re releasing Triton 1.0, an open-source Python-like programming language which enables researchers with no CUDA experience to write highly efficient GPU code—most of the time on par with what an expert would be able to produce.
OpenAI continues to deliver the goods for the AI community.
Inspired by a recent article from Erik Bernhardsson titled “Building a data team at a mid-stage startup: a short story”, Chris and Daniel discuss all things AI/data team building. They share some stories from their experiences kick starting AI efforts at various organizations and weight the pro and cons of things like centralized data management, prototype development, and a focus on engineering skills.
9 out of 10 AI projects don’t end up creating value in production. Why? At least partly because these projects utilize unstable models and drifting data. In this episode, Roey from BeyondMinds gives us some insights on how to filter garbage input, detect risky output, and generally develop more robust AI systems.