AI Master Group Wht Font Dwn
Background

Latest Videos

21:25

Aida Farahani: From 2D to 3D in Seconds

With a specialization in 3D deep learning, Aida is doing ground-breaking work related to 3D simulations. In this episode, she first describes why meshes or point clouds are computationally-expensive for simulating changes to shapes. She then describes why “implicit fields” are much more efficient for this, as described in the widely-cited “DeepSDF paper.” Next she describes a special-case problem: Predicting the deformation of 2D objects into 3D shapes, such as when metal blanks are stamped into the shape of a car door, for example, which is a case that violates one of the assumptions for DeepSDF. However in this episode, Aida describes a method for using implicit fields to solve this problem too, thereby transforming a simulation process that used to take 20 minutes per trial to just a few seconds per run. In effect, this show is a preview of her soon-to-be-published doctoral dissertation!

16:54

Sequoia Capital: Move 37 is Here!

This is a special edition of the ‘AI World’ video series covering the release of OpenAI-o1 (alias Q* and Strawberry). By whatever name, this is a very powerful new kind of model that has demonstrated remarkable reasoning abilities.

The video starts with a look back in time at “Move 37” – an iconic moment in AI history during the 2016 match between AlphaGo and Lee Sedol. That was a moment when the world saw AI do something that looked a lot like reasoning or strategy, and the latent promise implied by that moment seems to coming to life at this very moment.

For its storyline, the video draws on two very recent papers (and very important) papers:

  1. “Generative AI’s Act o1: The Agentic Reasoning Era Begins” by Sequoia Capital
  2. “Learning to Reason with LLMs” by OpenAI

First, to illustrate the new model’s capabilities, the video showcases that model’s success at decoding an encrypted message, which is definitely not something that a basic language would be able to do.

And with that as context, the focus then turns to the Sequoia Capital investment hypothesis, which is that considerable value will be be unlocked by companies that apply agentic AI in a domain-specific context, especially if those use cases target specialized pools of work. To illustrate this, the video presents XBOW, which is a company that’s been able to use agentic AI to replace highly-skilled experts that do cyber-security penetration testing.

Building on the implications of that example, the video concludes with reflections on the enormous potential impact of these new capabilities – opportunities and risks that can be measured in the trillions of dollars.

32:13

Nikhil Patel: Inside Sally Beauty’s Data Strategy

Nikhil is the Data Science Director at Sally Beauty Holdings, which is a $3.7 billion specialty retailer, with more than 10,000 products sold through over 4,000 stores, as well as online. Prior to his current role, Nikhil held a senior leadership role at Harman International, working with panel data from top CPG brands like P&G, Unilever and Kraft.

In this episode, Nikhil describes his 19-year journey, starting from a Masters degree in Applied Mathematics, leading up to his current role at Sally Beauty. Within that, he describes several times when he needed to supplement his formal education with ad hoc or certification-led studies, in order to stay current and relevant. He also discusses omni-channel marketing, customer journeys and Segment of One, as well as emotional understanding of the consumer, beyond pure data science. And he describes how one-on-one experiences, like Studio by Sally (in store) or “licensed colorist on demand” (online), can change customer journeys, and can provide valuable insights for new product development.

11:37

How an 8B Model Beat an Industry Giant

This video describes how a system called ‘AgentStore’ was able to gain the top spot on a benchmark for AI agents – beating out a gigantic model with a small one.

AgentStore is a platform and method for aggregating specialized agents that perform real-world tasks on digital devices on macOS, Windows and Ubuntu. In that system, a meta agent selects the best resource (or combination of resources) for each user request. The new benchmark was achieved using a small 8B model, outperforming industry heavy-weight Claude 3.5 Sonnet.

The testing was done on OSWorld, which is an environment for benchmarking agents on 369 different computer tasks involving popular web and desktop workflows, spanning multiple applications, ranging from Google Chrome and Microsoft Office to Thunderbird and PDF. The video describes some of the tasks that are part of this difficult benchmark. Testing was also done on APPAgent, which is a similar benchmark for mobile applications. The video reviews the test results and the capabilities of the agents, as well as the overall system design, including a special class of token that identifies what each agent can do. This information is used by a meta agent that picks the most suitable resource for each task, based on the information in those tokens.

24:50

Victor Perrine: From Bananas to $Billions

Victor (Viko) Perrine is the Global Director of New Growth Initiatives at Circle K. Prior to that, he’s held senior leadership roles at Delek US Holdings and at UGP Inc.

In this episode, Viko describes a major new initiative called Lift that’s unique to Circle K, which just launched in Europe, plus future development plans for that. Building on this, he describes the day-to-day for a global innovation leadership role, and he shares the success factors that help him to identify good target projects for innovation and growth. He also describes a project from earlier in his career that used lidar and computer vision to count banana trees on a 6,000 acre planation in the Philippines, which was the first time that had ever been done in that country. And with all this as context, he concludes with actionable tips for people who’d like to get the nod for leadership roles in innovation.

24:09

Ray Pettit: New Models for AI Literacy?

Ray is the Chief Data and Analytics Officer at Valhalla AI Solutions. Prior to that, he held senior leadership roles at the Advertising Research Foundation, at the Institute for Experiential AI, and at comScore.

In this episode, Ray discusses challenges associates with promoting meaningful AI literacy in business, starting with fundamental questions, like what is “AI literacy” exactly? (Is it merely the ability to use AI tools, or is a deeper understanding of the science behind those tools required?) He also discusses the reverse problem, where recent graduates from AI/ML programs lack sufficient domain knowledge to be full partners with key leaders at the companies that hire them. The end result is that many companies struggle to have meaningful dialog between their domain experts and their experts in AI and ML. On a more optimistic note, Ray describes a variety of initiatives in Canada, Germany and in select states in the US, that show promise for deepening the domain understanding of data scientists, while also empowering domain experts with a better understanding of data science.

24:02

Ivan Pinto: A Year of AI Testing in Software Dev

Ivan is an Associate VP of Delivery at Robosoft Technologies. His team does application development, engineering and QA for US clients, including web (HTML, CSS), mobile (Android, iOS, Samsung, LG) and streaming, including Roku TV.

In this episode, Ivan shares what he learned in his experiments over the past year using Gen AI to improve the efficiency and quality of code produced by his 300-member team, including results that he describes as “amazing,” with time savings of 40-50% in writing code. The technologies he discusses include GitHub Copilot, Amazon CodeWhisperer, Tabenine, Codeium, TestGrid and LambdaTest. And he describes how he chose between those, and also which ones he has his eye on for the future. Tests were done in various parts of the development lifecycle, including boilerplate code for new projects, bug fixing, unit testing, integration QA, and migrating from one language to another. Since not all of these situations benefited equally from AI-powered tools, Ivan describes what worked well, and where there are gaps that need further development work by vendors.

20:55

Sam Marks: Big Data, Big Bad Bruins

Sam is the Director of Business Strategy, Solutions & Analytics at the Boston Bruins & TD Garden. Prior to that, he directed strategy and analytics for the Arizona Coyotes, and at VaynerMedia.

In this episode, Sam sheds light on the world of analytics and business strategy for sports teams, including the differences in strategy and focus for teams vs at the league level. He also talks about the size and structure of analytics teams in the sports world, as compared to equivalent teams in consumer packaged goods, for example. And he describes how he uses AI and analytics conferences as part of his overall approach to team development.

30:55

Celia Wanderley: AI Innovator of the Year

Celia is the Chief Innovation Officer at Bits In Glass (BIG), a top Canadian IT consulting firm. Prior to that, she held senior leadership roles at AltaML and at Deloitte Canada. Celia was recognized as the AI Innovator of the Year, by Women in AI.

In this episode, Celia shares insights from trends she’s seen in her recent work involving intelligent automation of business processes at scale. A key topic is the idea of embedding AI agents into multiple steps within business processes, such as contract analysis or banking regulatory change management, in order to achieve business results that would probably not be achievable with a one-shot AI solution. She also describes successful efforts at transforming front-line user experience of field workers, replacing clumsy gadgets with voice-activated workflows. And she discusses critical success factors for taking AI projects into full production, confirming that AI and ML solutions frequently work quite well in a completely different context or industry from the one where they were originally created. She concludes by making a case for a blended approach to resourcing AI projects – one that involves in-house teams, third-party solutions, and external partners.

38:01

Andrei Lopatenko: Scaling AI to Billions

Andrei holds a PhD in Computer Science, and is the Director of Search and the AI Lab at Neuron7. Prior to that, he was the VP Engineering and AI at Zillow. He’s also held key leadership roles at Google, Apple, Walmart and eBay.

In this episode, Andrei shares insights and advice, based on his experience deploying large-scale, high-load NLP and search applications to billions of customers (5 billion queries per day, 10 billion pages per day). Along the way, he describes how a high-quality engineering culture and a high-quality science culture were nurtured during early his days at Google as one of the first 60 PhDs in 2006, and how he has applied what he learned there later in his career. You’ll also hear a discussion about critical success factors for a transition from POC to production for a large-scale projects, such as 100 million or a billion queries per day – including a discussion about evaluation metrics for LLMs. Andrei also emphasizes the importance of continuous learning for leaders of teams that do AI, and he describes a great approach for staying on top of current research. The episode concludes with valuable career advice for data scientists who are in the early stages of their career.

24:13

Dave Stern: Hackproof Your Startup

Dave is a fractional CTO and DevOps engineer with over 25 years of experience in systems and software engineering. He’s the President and Senior Solutions Architect of Stern DevOps Group, which is a consultancy focused on early stage companies. He’s also the author of a new book: Hackproof Your Startup, and that book is a key topic of the show.

In this episode, Dave discusses IT and AI security for early-stage start-ups. The conversation begins with a review of what happened in the famous Codespaces hack. Dave asserts that many companies are still vulnerable to the type of ransomware attack that put Codespaces out of business, and that the risk mitigation solution is fairly straightforward (the elements of which he describes on the show). Other topics include cybersecurity as an asset, infrastructure as code, principle of least privilege, and isolating IT environments. The conversation concludes with a what-if scenario where Dave answers the question: “If someone were to steal my laptop or cell phone. What would I suddenly wish I had done before that happened?”

46:22

Shawn Goodin: Agent-Driven Marketing?

Shawn is the Global VP of Solutions at FirstHive, which is a customer data platform. Prior to that, he held senior leadership roles at Capgemini, Silicon Valley Bank, JPMorgan Chase, Clorox, Northwestern Mutual and SC Johnson. He is also an advisory board member of the Customer Data Platform (CDP) Institute.

In this conversation, Shawn describes various roadblocks to transformation in large organizations – especially AI-based initiatives in marketing. He then shares an agentic vision for a future-state where a marketing operations user might simply say: “I want to grow my credit card business by 20% in the US. What should I do?” and the platform would develop and execute a plan for that.

30:57

Jodi Blomberg: Strategic Bets on AI

Jodi is the VP of Data Science at Cox Automotive, a company that has a diverse portfolio of 17 brands that encompass digital products like Kelley Blue Book and Autotrader, as well as various kinds of physical services – all of which are supported by about 70 in-house data scientists and ML engineers.

In this conversation, Jodi describes her AI initiatives as investments, managed in a way that’s similar to a diversified investment portfolio, where the core projects deliver a baseline of ROI, and are supplemented by strategic bets, plus a very small fraction of high-risk / high-reward projects (“moonshots”) that get a Yes/No decision within 4-6 weeks. The show also includes a discussion about what makes a Gen AI project strategic vs “must have,” as well as insights about the practical and human challenges associated with the kinds of AI-based initiatives that primarily target efficiency gains, rather than top-line growth.

58:14

Ramsu Sundararajan: Segment of One at Scale

Ramsu holds a PhD in Machine Learning, and is the Head of R&D at solus.ai, which powers Segment of One personalization. Roles prior to that included Senior Scientist at GE Global Research, and Principal at Sabre Airline Solutions, where he developed some of the original algorithms.

Ramsu shares highlights of his journey in AI, with particular focus on personalization in marketing, with insights about how to think about that problem conceptually, including what parts are somewhat easy and which are difficult or tedious. There are also key insights about customer journeys and about the cold start problem. Other topics covered include customer genomes, as well as a discussion about navigating between decisions taken at a zoomed-in perspective at the individual customer record level, while also managing to a zoomed-out perspective that’s driven by KPIs, comps and annual targets. The show concludes with a discussion about the 1-2 year product development roadmap for solus.

2:53

Announcing the AI Master Group Podcast

The AI Master Group Podcast launches this week on Friday!

The show is an ideal place to meet people who are creating the future of AI in their work today. It features interviews with authors of recent papers, and with people deploying AI in real-world applications, including people in senior leadership roles. If you want a front row seat to the technology and craft of AI, this show will be perfect.

You can Follow the show on Spotify at this link: https://spoti.fi/3XIf1te (The show will also be available on the other usual channels, such as Apple Podcasts.)

There’s a list of the first 14 guests with release dates per episode in the YouTube Comments.

8:03

Mesh Anything (except a Pink Hippo Ballerina)

The developers at MeshAnything have just released new code that offers an important improvement in how the surface of 3D objects can be encoded. What the new method does is build out the shape by always seeking to find and encode an adjacent face that shares an edge, which requires only about half as many tokens to represent the same information by other methods, resulting in a four-fold reduction in the memory requirement to achieve the same task, which enabled MeshAnything to double the maximum number of faces it can handle on a single object to 1600, as compared to 800 for current methods.

This video starts by comparing the new method with the current one. After that, we generate a 3D object from a text prompt on the Rodin website (a pink hippopotamus ballerina character with white tutu), and we check it on the Sketchfab website. Then we run the code that was provided by MeshAnything on GitHub, and we check the output on Sketchfab, comparing before and after side-by-side. The results confirm the final words of the paper, which state that “the accuracy of MeshAnything V2 is still insufficient for industrial applications. More efforts are needed.” Nonetheless, this new computational approach is elegant, and the video concludes with a prediction that we’ll likely see improvements that build on the foundations laid by MeshAnything V2.

8:36

Can Robots Win at Table Tennis? Take a Look!

Google DeepMind has just achieved a new level of robotic skill – the ability to compete and win at table tennis, a game that requires years of training for people who want compete at an expert level.

This video shows the robot in action against an array of competitors, ranging from beginner level to tournament pro and, in doing so, describes both the hardware and AI aspect, including how it was trained and a summary of the key innovations contributed by this project.

It also gives summary results of the live matches, segmented by experience level of opponents. As a bonus, I looked at the performance data and have shared four insider tips for how to beat this robot at table tennis. The video ends on a light note, describing something called RoboCup, which has the goal of fielding a team of robots that will be ready to take on the World Cup soccer champion team by 2050. You’ll quickly see that we have a very long way to go on that particular goal.

10:36

Shark Alert! YOLO AI-Vision in Action

Last week, several news outlets ran a story about SharkEye, which is an AI-vision shark detection program, developed at the University of California, Santa Barbara, and deployed at California’s Padaro Beach, which is an area where surfers and great white sharks are both frequently found.

After quickly describing the program itself, the video identifies the underlying technology that was used for the vision aspect, confirming from the project’s GitHub page that YOLO v8 by Ultralytics was used. Basically, Ultralytics created an abstraction layer that simplifies the deployment of computer vision models, so that even developers with almost no experience in computer vision can quickly implement sophisticated projects. To illustrate, the video then shows a demo of an object detection and identification task being set up and run on Google Colab. It then concludes with examples of types of projects that can be implemented by Ultralytics YOLO v8.

11:03

AI Can do That?? Silver Medal in Pure Math

AI has just achieved an amazing milestone. A couple of Alpha models by Google DeepMind scored silver-medal-level performance in a globally-recognized competition in advanced mathematics: IMO 2004.

This video starts by setting the context for this latest achievement, going back to significant milestones in 2022 and 2023 that helped set the stage for what just happened, sharing the story along the way of two remarkable mathematicians, and comparing their achievements to those of the Alpha models.

With the stage set in that way, the video then describes key details of the contest, including the scoring system, and how DeepMind scored on each problem, including details of a very difficult geometry problem that is solved in a matter of seconds. Next the video describes details about the training that was done for the AlphaProof and the AlphaGeometry 2 models. Finally, it assesses the implications of this accomplishment, including some of the fields in which this kind of capability might make significant contributions.

8:12

Will Open-Source Llama Beat GPT-4o?

Last week Meta launched its newest family of models, Llama 3.1, including a new benchmark – an open-source foundation model with 405 billion parameters. With this, Zuckerberg predicted that Meta AI will surpass OpenAI’s 200 million monthly active users by the end of this year.

Hubris aside, this video looks at six reasons why we need to pay attention to this announcement, including Zuckerberg’s assertion that open source will eventually win for language models for the same reasons that Linux eventually won out against an array of closed-source Unix models.

It then describes a situation where a company has already been building solutions using an OpenAI model or Anthropic, for example, but then decides to get an informed point of view about the open source option by creating a challenger model as well, using the new Llama options. For that situation the video suggests which model size to use, plus recommendations for best platform options for the pilot, plus four types of projects that would be good candidates for a head-to-head test of this sort. Finally, it concludes with a light-hearted description of the battle ahead.