fbpx

Dean Wampler

AI Alliance company logo.
IBM's Principal Technical Representative to the AI Alliance
IBM Research

PRESENTATION TITLE:

Can We Make Model Alignment More Like Software Engineering?

PRESENTATION SUMMARY:

Atom icon for The AI Conference 2023, a groundbreaking two-day event on AGI, LLMs, Infrastructure, Alignment, AI Startups, and Neural Architectures.Models are trained by AI experts with significant compute resources, while final model alignment is often done by software engineering teams with less AI expertise, who also want model alignment to fit into conventional software engineering practices. 

While there are established tools and techniques for model alignment, it is still more art than science, more incantation than reproducible engineering process, especially for development teams without strong AI expertise.

Brain icon for The AI Conference 2023, a groundbreaking two-day event on AGI, LLMs, Infrastructure, Alignment, AI Startups, and Neural Architectures.

This talk covers the following:

  1. A quick summary of the objectives for and challenges of alignment.
  2. A summary of what an ideal, repeatable process should be.
  3. An examination of the techniques developed in the last few years for defining and implementing alignment “”capabilities”” in a standardized, programmatic, repeatable, and efficient way.
  4. A walk through of a specific example demonstrating the details.
  5. A summary of the benefits for alignment as an engineering process, as well as additional work required.

2024 PRESENTATION Slides:

About | Dean Wampler

Dean Wampler is co-organizing the AI Alliance (thealliance.ai), helping to define and drive its technical focus areas.

He has led several engineering teams at IBM Research and other companies. Dean is an O'Reilly and Manning author, and a regular conference speaker on data and AI topics.

He has a Ph.D. in Physics from the University of Washington.