Vector Databases, Embeddings, and a history of Deep Learning with Leo Dirac
Jun 8, 2023
auto_awesome
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.
Groundlight AI offers a novel approach to computer vision by leveraging natural language and human oversight to provide customized models without extensive data sets.
Fine-tuning pre-trained models is the new paradigm in machine learning, allowing for faster and easier deployment of computer vision systems.
Building an effective escalation framework for computer vision requires careful calibration, understanding of limitations, and the ability to trust and rely on the results.
Deep dives
Groundlight AI: Revolutionizing Computer Vision with Natural Language
Groundlight AI is a computer vision service that leverages natural language to provide a novel approach to visual understanding without the need for extensive data sets. By combining a visual large language model with human oversight, Groundlight allows users to describe their visual tasks in natural language and receive customized computer vision models. With a focus on calibration and understanding when to escalate to human monitors, Groundlight ensures reliable results without the need for lengthy data set collection and model training. The technology is particularly well-suited for high mix manufacturing, where automation and computer vision have traditionally been challenging to implement. Groundlight's approach allows for a faster and more cost-effective deployment of computer vision systems in this domain.
The New Paradigm of Fine-Tuning Pre-Trained Models
The new paradigm in machine learning involves fine-tuning pre-trained models rather than starting from scratch. Groundlight embraces this approach by providing users with customized computer vision models based on their specific visual tasks. By leveraging existing models and adapting them to the user's needs, Groundlight allows for faster and easier deployment of computer vision systems. This approach is most effective in situations where latency and cost are tolerant, and where the visual problem is changing regularly.
Building an Escalation Framework for Computer Vision
Building an escalation framework for computer vision requires careful calibration and understanding of the limitations and failure modes of machine learning models. Groundlight has invested in developing a framework where models can escalate to human monitors when uncertainty arises, ensuring reliable results in real-time. Balancing latency, cost, and quality is key in building an effective escalation framework. The ability to trust and rely on the results of computer vision systems is crucial, and Groundlight's approach addresses this challenge.
The Changing Definition of Artificial General Intelligence
The definition of artificial general intelligence (AGI) continues to evolve as technology advances. While AGI remains elusive, current AI systems have made significant strides in replicating human capabilities, including passing high school and college tests and performing various complex tasks. However, the debate around AGI extends beyond technical capabilities and encompasses questions about emotions and consciousness. Trust and reliability are critical considerations in developing AI systems, and Groundlight's approach prioritizes human oversight to ensure accuracy and safety.
The Future of Automation and AI
Groundlight's vision of the future involves automation and AI systems working alongside humans to enhance productivity and effectiveness. While concerns about job displacement exist, history has shown that automation often leads to job upskilling and more interesting work. The integration of AI into various industries, such as co-pilot tools and natural language programming, has already demonstrated the potential for AI to augment human capabilities. Groundlight sees automation as a means to make people more effective and anticipates a positive impact on professional life as the technology continues to advance.
On the newest episode of Practically Intelligent, Sinan and Akshay discuss all the hype behind vector databases and the future of how developers will work with embeddings. We also dissect how to compare different vector database options available on the market today. Congrats to Weavite, Pinecone, Qdrant and Chroma on their funding rounds!!We also have Leo Dirac, founder and CEO of Groundlight.AI - who was formerly the Engineering Lead behind Deep Learning at AWS. We take a walk through history with some key takeaways for builders and observers of the AI/ML space and talk about Leo's new - Why Deep Learning Caught the ML World by Storm - How many leading scientists and researchers underestimated the impact of novel DL on the category- How handling escalation of errors is a critical obstacle to computer vision fulfilling its potential- How Groundlight can help simplify computer vision for engineering leadersTimestamps3:30: Why are Vector Databases So Important? 09:40: How do you compare different Vector Database if you're a developer? 22:10: Guest Leo Dirac, a History of Deep Learning at AWS and AutoML39:17: Fine-Tuning as the New AutoML 43:00: What is GroundLight.AI and Human Escalation in Computer Vision
Get the Snipd podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode
Save any moment
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Share & Export
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode