No Priors Ep. 85 | CEO of Braintrust Ankur Goyal
Topics covered
Popular Clips
Questions from this episode
- Asked by 126 people
- Asked by 100 people
- Asked by 98 people
- Asked by 77 people
- Asked by 58 people
- Asked by 40 people
- Asked by 36 people
- Asked by 34 people
- Asked by 34 people
- Asked by 28 people
- Asked by 27 people
- Asked by 24 people
- Asked by 22 people
- Asked by 20 people
- Asked by 17 people
Episode Highlights
Open-Source
Open-source AI models are gaining traction, yet their adoption faces challenges. notes that while there is significant interest, practical adoption remains limited due to the complexity of evaluating AI models and the high ROI of proprietary solutions 1 2. He explains the difference between instruction tuning and fine-tuning, highlighting the intricacies involved in modifying AI models 3. Goyal emphasizes the importance of providing the best user experience and fast iteration speed, which proprietary models currently offer more effectively 2.
We are very close to a watershed moment for open-source models.
---
The shift towards open-source is anticipated, but it requires overcoming significant hurdles in performance and usability.
Language Shift
The programming landscape in AI is shifting, with observing a growing preference for TypeScript over Python 4. He attributes this shift to TypeScript's robust type system, which better handles AI workloads by structuring uncertain data more effectively 5. Goyal also notes a trend away from using frameworks, as AI becomes an integral part of software engineering rather than a separate entity 5.
TypeScript is just a much, much better language for writing software that deals with uncertain shapes of data.
---
This evolution reflects a broader movement towards integrating AI into existing systems, rather than building new frameworks from scratch.
Infrastructure
AI infrastructure is undergoing significant changes, particularly in data handling and storage. highlights the limitations of traditional data warehouses for AI applications, advocating for the use of embeddings and models to sift through large datasets 6. He points out that the value of data is shifting from mere accumulation to effective reasoning and problem-solving capabilities 2. This transformation necessitates a reevaluation of how enterprises collect and utilize data in AI processes.
The value of data is actually, and how we think about the value of data is very, very different.
---
As AI continues to evolve, the infrastructure supporting it must adapt to new paradigms of data usage and processing.