Infinite Curiosity Pod with Prateek Joshi

Douwe Kiela is the cofounder and CEO of Contextual AI, where they are building the next generation of foundation models that provide fully customizable, trustworthy, and privacy-aware AI. He is also an Adjunct Professor at Stanford. He was previously the Head of Research at Hugging Face and a research scientist at Meta. He has a PhD in Computer Science from the University of Cambridge. In this episode, we cover a range of topics including: - Open source vs closed source AI - White House's Executive Order on AI - Enterprise needs when it comes to LLMs - The founding of Contextual AI - What does RAG 2.0 look like for LLMs - Large foundation models vs smaller specialist models - Do Big tech companies have an unassailable lead in AI? - Size of the data vs size of the model. What is more important and why? - How will the AI compute market shape up? Douwe's favorite books: - The Lord of the Rings trilogy (Author: J.R.R. Tolkien)- The Count of Monte Cristo (Author: Alexandre Dumas)- Amp It Up (Author: Frank Slootman)--------Where to find Prateek Joshi: Newsletter: https://prateekjoshi.substack.com Website: https://prateekj.com LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 Twitter: https://twitter.com/prateekvjoshi 

Ville Tuulos, co-founder and CEO of Outerbounds, has a rich background from Netflix and AdRoll. He discusses the inception of the open-source framework Metaflow, designed to facilitate machine learning collaboration. The conversation highlights the importance of AI compute clusters and the distinction between large foundation models versus smaller specialist ones. Ville also explores trends in industrial and multimodal AI, and emphasizes the need for innovative AI infrastructure solutions. Dive into the future of AI with insights from his extensive experience!

In this episode, the host Prateek Joshi talks about Nvidia's new AI agent called Eureka. It can train robots to do all sorts of complex tasks. Large language models like GPT-4 are really good at making plans and decisions for certain tasks. But when it comes to teaching robots to do complicated physical stuff, like spinning a pen around their fingers, they've been a bit clumsy. That's where Eureka comes in. Eureka takes the best parts of LLMs and uses them to create a reward system for teaching robots. Blog post: https://blogs.nvidia.com/blog/2023/10/20/eureka-robotics-research/ Paper: https://arxiv.org/abs/2310.12931 Code: https://github.com/eureka-research/Eureka --------Where to find Prateek Joshi: Newsletter: https://prateekjoshi.substack.com Website: https://prateekj.com LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 Twitter: https://twitter.com/prateekvjoshi 

Prateek Joshi, AI expert, discusses the impact of Generative AI in 13 areas of Biology including drug discovery, protein design, and personalized medicine. He explores how AI models can speed up protein design, generate synthetic data, analyze gene regulatory networks, and uncover hidden information in protein-protein interactions.

In this episode, the host Prateek Joshi talks about LLMs learning to represent space and time.  Here's the paper where the authors have discussed it in detail:  https://browse.arxiv.org/pdf/2310.02207.pdf--------Where to find Prateek Joshi:  Newsletter: https://prateekjoshi.substack.com Website: https://prateekj.com LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 Twitter: https://twitter.com/prateekvjoshi 

In this episode, the host Prateek Joshi talks about the Reversal Curse phenomenon in LLMs. Here's the paper where the authors have discussed it in detail: https://owainevans.github.io/reversal_curse.pdf--------Where to find Prateek Joshi: Newsletter: https://prateekjoshi.substack.com Website: https://prateekj.com LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 Twitter: https://twitter.com/prateekvjoshi 

Ali Golshan is the cofounder and CEO of Gretel AI, a synthetic data platform for ML developers. They have raised $65M in funding so far from investors such as Greylock and Anthos. He was previously the cofounder of StackRox, which was acquired by Red Hat for about $450M. Prior to that, he was the cofounder of Cyphort, which was acquired by Juniper Networks. In this episode, we cover a range of topics including: - The need for synthetic data - Techniques to generate synthetic data - How can AI enhance the synthetic data generation process - Computational irreducibility - Differential privacy - Measuring the performance of the engine that generates synthetic data Ali's favorite books: - The Order of Time (Author: Carlo Rovelli) - The Coddling of the American Mind (Authors: Greg Lukianoff and Jonathan Haidt) --------Where to find Prateek Joshi: Newsletter: https://prateekjoshi.substack.com Website: https://prateekj.com LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 Twitter: https://twitter.com/prateekvjoshi 

Alex Ratner is the CEO of Snorkel AI, a platform that provides programmatic data labeling and foundation models to enable companies to build AI applications. They've raised $135M so far from amazing investors such as Addition, Greylock, Google Ventures, and Lightspeed. He was previously the cofounder and CEO of SiftPage. He has a bachelors degree from Harvard and a PhD from Stanford. In this episode, we cover a range of topics including: - Making AI data development first-class and programmatic - The data-centric step for every model-centric step - False dichotomy of fine tuning vs RAG - Foundation model dynamics: winner take all vs diverse models - Training compute-optimal LLMs - Designing multimodal datasets (DataComp) - Distilling Step-by-Step - 'GPT-You' for every enterprise Alex's favorite books: Foundation series books (Author: Isaac Asimov)--------Where to find Prateek Joshi: Newsletter: https://prateekjoshi.substack.com Website: https://prateekj.com LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 Twitter: https://twitter.com/prateekvjoshi 

Riddhiman Das is the CEO of TripleBlind, a privacy platform for AI. They have raised $32M in funding, with their most recent round led by General Catalyst. He was previously the Head of International Technology Investments at Ant Financial, which is Alibaba's financial services arm. He was the Product Architect at Zoloz, Chairman at Laplacian, Chief Data Offier at mySideWalk, and CTO of Galleon Labs. He has received the 2013 White House Champions of Change from President Barack Obama. In this episode, we cover a range of topics including: - Attack surface of an AI application - What are the ways in which privacy can be compromised during training and deployment of AI models - Role based access control for Generative AI applications - Data leakage in Generative AI applications - Characteristics of a good privacy product - How is TripleBlind used in healthcare and financial sectors Riddhiman's favorite book: Twenty Thousand Leagues Under the Sea (Author: Jules Verne)--------Where to find Prateek Joshi: Newsletter: https://prateekjoshi.substack.com Website: https://prateekj.com LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 Twitter: https://twitter.com/prateekvjoshi 

In this episode, the host Prateek Joshi talks about AI infrastructure ideas and categories including: Model infrastructure: - Compute hardware - ML frameworks - Foundation model providers - Distributed model training - Model deployment - Building and serving verticalized models - Streaming ML models - Monitoring and logging - Experimentation framework - On-device applications - Orchestration platform - Latency - User feedback loopData infrastructure:  - Storage hardware - Data acquisition - Databases - Data labeling - Cataloging product usage data - Privacy and security - Backup and redundancy systems --------Where to find Prateek Joshi: Newsletter: https://prateekjoshi.substack.com Website: https://prateekj.com LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 Twitter: https://twitter.com/prateekvjoshi