Practically Intelligent

In episode 11 of Practically Intelligent, we take a look at the dynamic and intricate world of business intelligence (BI) with the esteemed Donald Farmer, a pioneer in BI and the driving force behind Power BI. This episode takes us on a journey through the evolution of BI, from its roots in decision support systems to the burgeoning era of machine learning and the transformative potential of language models.Donald shares insights on:The early days of BI and the launch of Power BI, setting the stage for today's AI advancements.The transition from traditional data warehousing to the necessity for centralization and the advent of self-service BI tools.How predictive analytics moved from a specialization to a mainstream necessity, paving the way for AI's role in strategic decision-making.The emergence of language models as key players in shaping strategic decisions, making AI not just a tool but a team member in the decision-making process.We also explore the broad spectrum of decisions influenced by BI tools, from operational to strategic, and how machine learning and AI are revolutionizing this landscape. Donald offers a critical perspective on the current hype cycle in AI, distinguishing between genuine innovation and overblown claims.

Dr. Nathan Lambert is back! Episode 10 of Practically Intelligent has us diving headfirst into a the tsunami of AI advancements and the hype cycles surrounding them. With the field evolving at breakneck speed, it's easy to oscillate between awe and skepticism. Join us in a thought-provoking conversation with Nathan Lambert as we explore how to navigate this complex landscape.This episode is tailored for both builders and investors in AI, as well as anyone curious about the technology's rapid development.We discuss:- How to separate mind-blowing AI advancements from overhyped claims.- Strategies for improving your 'information diet' in the AI space.- Practical tips for staying grounded and accurately informed about technical progress, including basic checks for affiliations and scrutinizing code.- Understanding general principles in machine learning to discern quality information.

What were the biggest trends in AI in 2023 that will accelerate into 2024? What will change/what is overhyped? In this episode, we go through charts and reports from Coatue's recent AI report to dissect the key assertions made about the future of AI, from its transformative impact on technology to the burgeoning role of open-source. We'll highlight the most compelling insights, challenge hyperbole, and correct any misleading narratives we find.Whether you're an AI enthusiast or skeptic, this episode promises a nuanced exploration of AI's potential and pitfalls, offering a clear-eyed perspective on what's genuinely revolutionary and what's just hype in the world of artificial intelligence.The Report: https://www.coatue.com/blog/perspective/ai-the-coming-revolution-2023Chart 1: "Despite intense demand, AI compute costs have decreased" - Page 78Chart 2: "Problem: Model evaluation is broken today" - Page 88Chart 3: "Synthetic data can augment fine-tuning" - Page 62Chart 4: "Data quality is just as important as data quantity" - Page 61Chart 5: "Data scarcity is a potential wall to scaling models" - Page 60Chart 6: "AI regulation may be more likely than most think" - Page 26

Join us in episode eight of Practically Intelligent as we welcome Jacob Solawetz, CTO at arcee.ai and former founding engineer at Roboflow. Jacob, a trailblazer in computer vision and AI, discusses the evolution and impact of multimodal AI technologies. Delving into the challenges of developing large-scale vision applications, he offers insights from his rich experience in both vision models and domain-adapting LLMs - and what that means for the future of multimodal AI. Jacob also sheds light on his work at arcee.ai, focusing on specialized language models and the art of model distillation.

In this enlightening seventh episode of Practically Intelligent, we take a look at the pivotal role of benchmarking in advancing AI with Praveen Paritosh, a leading figure in AI research. Discover why shared benchmarks are not just important, but critical in pushing the boundaries of AI technology. Praveen enlightens us on the legacy benchmarks like SQuAD, instrumental in testing early question-answer systems, and how they paved the way for early leaderboards in AI. We discuss the concept of shared benchmarks as a mechanism for the research community to collectively tackle and progress in specific challenges, drawing parallels between NLP and image recognition benchmarks like ImageNet. However, it's not all straightforward – benchmarks, while guiding us in the right direction, are merely proxies. We discuss the challenges of differentiating between conceptual learning driven by reasoning and rote learning based on memorization. Join us for a deep dive into the intricacies and nuances of AI benchmarking, a critical yet complex tool in the evolution of artificial intelligence.

Giada Pistilli, Principal Ethicist at Huggingface, discusses AI ethics and data governance. Topics include training data inconsistencies, potential biases in AI, ethical training guidelines, 'moral charter' in AI models, and hazards of 'Ethics Shopping'.

Adam Azzam, AI Product Lead at Prefect, discusses challenges of merging traditional software engineering with AI engineering. They explore marrying large language models with strongly typed semantic interfaces, suitable problems for Marvin AI, and the benefits of using LLMs in software engineering.

Sinan and Akshay chat with Nathan Lambert, a prominent Machine Learning researcher and analyst. They discuss evaluating language models, the Open LLM leaderboard, Luther's test harness, and the challenges of evaluating large language models and low data quality.

Former Engineering Lead behind Deep Learning at AWS, Leo Dirac, shares a walk through history and key takeaways for builders in the AI/ML space. They discuss the importance of vector databases, comparing different options, and the challenges of computer vision. Leo also talks about his new venture, Groundlight.AI, and its role in simplifying computer vision for engineering leaders.

In this episode, Sinan and YC W23 founders discuss how they're building with LLMs, the architectural tradeoffs they've made, and navigating discussions with customers. They also cover the release of open source models, the blistering pace of news in the AI ecosystem, and licensing and IP questions. The podcast explores topics like using AI to generate instructions, Y Combinator companies using LMs, building and fine-tuning AI models for enterprise clients, data security and privacy concerns when integrating LOMs, the importance of data in the era of AI, language learning journey, and the correctness of machine translation in multiple languages.