Ep 33: CTO and Co-Founder of Sourcegraph on Current Landscape and Future of Software Development, How to Make RAG Better, and Building Towards the Agentic Future
Apr 30, 2024
auto_awesome
CTO and Co-Founder of Sourcegraph, Beyang Liu, discusses AI coding, making RAG better, and building towards an agentic future. They cover AI products in Sourcegraph, the current state of AI coding, customizing RAG, and the future of engineers. Liu shares insights on inference cost, evaluating products, and surprises in their journey.
AI tools like Cody enhance coding experiences with code search and assistance.
AI reshapes daily tasks of developers by streamlining code generation and comprehension.
Source Graph validates AI models through offline benchmarks for efficient code assistance.
Local inference in AI tools ensures privacy, reduces latency, and boosts coding productivity.
Deep dives
AI Transforms Coding with Interactive AI Coding Assistant Cody
Cody, an AI coding assistant developed by Source Graph, leverages AI to enhance coding experiences with features like code search and AI coding assistance. The tool incorporates search techniques and model evaluations to refine and improve code generation. AI benefits both novice and senior developers offering inline code completion and chat features, aiding in software creation.
Impact of AI in Software Engineering
The podcast delves into the impact of AI on software engineering, highlighting the transformative role AI plays in reshaping the daily tasks of developers. By generating code snippets and assisting with understanding complex codebases, AI tools like Cody streamline the development process, enabling faster and more efficient coding practices for programmers. The integration of AI in coding tools signifies a significant shift in how developers approach building software in the modern technological landscape.
Data-driven Model Development
Source Graph's structured approach to model development entails running offline benchmarks and evaluation tests to validate and enhance the performance of AI models. By fine-tuning models and setting up quantitative benchmarks, the team ensures that AI-driven features like code completions and chat responses meet quality standards and effectively assist users in their coding tasks. This data-driven model development methodology underscores the commitment to delivering reliable and efficient AI-powered tools for developers.
Empowering Developers with Local Inference and Latency Optimization
Source Graph acknowledges the importance of local inference for developers, offering options for running models locally to address privacy concerns, reduce latency, and enhance coding experiences. Local inference capabilities in tools like Cody cater to developers' preferences for privacy and lower latency, ultimately improving the overall coding workflow and user experience. The focus on optimizing latency highlights the significance of fast and responsive AI tools in boosting productivity and engagement among developers.
Developing Search Stacks for Context Retrieval
The podcast delves into the development of search stacks for context retrieval, focusing on the integration of code search and retrieving context in the chat UI called 'rag.' The approach involves using parallel threads to address the problem as an end-user search issue and a context optimization challenge. By building pieces from code search that can be repurposed for context retrieval and vice versa, the podcast highlights the synergy between the two domains.
Innovations in AI and Search Strategies
Discussing the evolution of AI in search strategies, the podcast draws comparisons with Google's evolution from relying on technical insights like PageRank to prioritizing user behavior data for search quality. The discussion emphasizes the importance of adapting search strategies based on user intent and behavior, highlighting a shift towards data-driven approaches while also acknowledging the role of human judgment calls in shaping search outcomes.
Future of AI-Enabled Development Tools and Open Source Models
Exploring the future landscape of AI-enabled development tools, the podcast envisions a shift towards democratizing access to AI capabilities for developers. The conversation underscores the value of high-quality context providers across various user experiences, emphasizing the potential for open-source models to drive innovation and empower developers globally. By advocating for building a sustainable ecosystem that fosters choice and flexibility, the podcast anticipates the widespread adoption of open-source models and AI technology in software development.
On this week’s Unsupervised Learning, Pat and I sat down with CTO and Co-Founder of Sourcegraph, Beyang Liu. Sourcegraph is a leader in the AI coding space, and recently launched AI coding assistant, Cody. Beyang shared with us his view on the current landscape of AI coding and the future of coding and software development. He also shared how Sourcegraph has tried to make RAG better, and their model eval approaches.
(0:00) intro (0:47) advice for young coders (3:34) AI products at Sourcegraph (6:17) the current state of AI coding (12:33) what happens when a new GPT model comes out? (20:16) what types of developers benefit from these AI tools? (30:45) how important is inference cost? (35:31) how does Sourcegraph structure AI teams? (41:27) what metrics does Sourcegraph use to evaluate their products? (50:02) customizing RAG (56:55) getting ahead of the agentic future (1:05:05) will there be more or less engineers in the future? (1:13:50) over-hyped/under-hyped (1:16:56) surprises during the Sourcegraph journey (1:18:26) cognition buzz and Devin (1:26:48) Jacob and Pat debrief
With your co-hosts:
@jacobeffron
- Partner at Redpoint, Former PM Flatiron Health
@patrickachase
- Partner at Redpoint, Former ML Engineer LinkedIn
@ericabrescia
- Former COO Github, Founder Bitnami (acq’d by VMWare)
@jordan_segall
- Partner at Redpoint
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
