In this podcast episode, Michael Rosen, the creator of Smart Lens, a computer vision startup, discusses his journey in the field of computer vision. He talks about attending Apple's WWDC conference and developing a successful classification model using deep learning that surpassed Google Lens. Rosen also explores his work with natural language processing and the challenges he faced in the enterprise space. The episode showcases the experimentation with Smart Lens, including the development of an enterprise version and an invoice processing product. Additionally, the concept of long form question answering and its applications are introduced, along with the expansion of the search index for the question answering system. The episode then delves into "Fine", a system aimed at supporting developers by providing accurate answers and relevant code context. The philosophy behind "Fine" and its potential as a general technical reasoning engine are discussed. The episode wraps up with conversations about software engineering, evaluation, the future of AI models, training models with high-quality data, and the importance of belief and community-building in the startup journey.
Takeaways
• Smart Lens, a computer vision startup developed by Michael Rosen, gained popularity with its successful classification model using deep learning and its ability to run efficiently on the iPhone.
• The enterprise version and invoice processing product of Smart Lens didn't gain much traction due to the COVID-19 pandemic.
• Rosen explored natural language processing using different models, such as hugging phase and long former models, and faced challenges in catering to different user preferences.
• The episode introduces the concept of long form question answering and discusses the creation of a comprehensive search index using the Common Crawl dataset.
• "Fine" is a system designed to support developers by providing accurate answers and relevant code context, and it aims to become a general technical reasoning engine applicable beyond programming.
• The episode emphasizes the importance of software engineering beyond just running code and highlights the future of problem-solving and AI assistance in executing ideas.
• The evaluation and user feedback of AI models are crucial, and "Fine" aims to improve correctness and consistency in its responses.
• The evolution of AI models has led to powerful reasoning engines, and future advancements may focus on solving hallucinations and ensuring guaranteed correctness.
• Building a startup based on a deeply meaningful idea and fostering a strong belief and community are essential for success in the entrepreneurial journey.
