Suggestions
Matthew Honnibal
Founder at Explosion
Matthew Honnibal is a prominent figure in the field of artificial intelligence, particularly known for his expertise in natural language processing (NLP). He is the co-founder of Explosion AI, a company that specializes in AI technologies and is the creator of spaCy, one of the leading open-source libraries for NLP.
Educational Background and Career
- PhD: Honnibal completed his PhD in 2009, focusing on syntactic parsing and structured prediction problems. Following this, he spent five years conducting research in natural language understanding systems.
- Academic Positions: He held post-doctoral fellowships at Macquarie University and worked as a research scientist at the University of Sydney before transitioning to industry.
Founding Explosion AI
In 2014, Honnibal left academia to develop spaCy, recognizing a gap in the Python ecosystem for robust NLP tools. By 2016, he co-founded Explosion AI with Ines Montani. The company aims to create custom algorithms and applications for AI, emphasizing developer experience and open-source solutions. Under his leadership, Explosion has also released Prodigy, an annotation tool designed to streamline data labeling processes for machine learning projects.1245
Recent Developments
As of 2024, Explosion AI has returned to operating as a self-sufficient company after a brief period of venture capital involvement. Honnibal continues to focus on maintaining and developing spaCy and Prodigy, ensuring they remain aligned with user needs and technological advancements.34
Honnibal's contributions to the field are significant, as he has been instrumental in making advanced NLP accessible to developers and organizations worldwide.
Highlights
claude -p lets you execute Claude as a single command. This is nice for workflows that need intelligence for one step but need privileges for a different step.
For instance you can get Claude to generate release notes or PR text, then deterministically make the PR or release, without giving Claude write permissions to your Github.
Something I've always wondered: How come LLMs can't be architected better than just concatenating the prompt and the context and embedding it all? Like, why can't you concatenate a representation of just the prompt higher up in the model, to prevent prompt injection?

