Evals are essentially automated tests where a larger, more powerful LLM judges the output of a smaller, faster one to overcome the non-deterministic nature of AI responses.
Developers are now facing the reality of token-based pricing for AI, making it crucial to find value and consider cheaper open-source models to avoid high costs.
The current AI trend is compared to the iPhone's launch, suggesting that while there's a bubble, a core of real value will remain and integrate into every company's tech stack.
The misuse of AI to generate low-quality content, known as "slop," is overwhelming open source projects with bad pull requests and creating a significant moderation problem.
AI-powered security research is creating an imbalance where attackers find vulnerabilities in the NPM ecosystem faster than defenders can patch them, increasing supply chain risks.
While frameworks and AI can accelerate development, there's a growing debate about whether to slow down releases to improve quality and security in the face of AI-generated code.
AI is eliminating entry-level coding tasks, forcing new developers to start with higher-level skills like architecture and detailed specification writing instead of just writing code.
In-person events in tech hubs like Silicon Valley remain vital for fostering the unpredictable human connections that lead to cross-domain discoveries and new ventures.
Jobs that call for the skills explored in this talk.