Skip To Main Content
Craig Smith on Muck Rack

Craig Smith

Verified
New York
As seen in: Eye On A.I., Apple Podcasts, Forbes, MSN UK, The New York Times, Yahoo Canada, Yahoo News Australia, Yahoo News Malaysia, Yahoo Singapore, Harvard University, The Wall Street Journal and
Would-be entrepreneur and longtime writer for The New York Times. Focused on artificial intelligence.

Get in touch with Craig

Contact Craig, search articles and posts on X, monitor coverage, and track replies from one place.

Learn more about Muck Rack

Actions

Share this page

Craig Smith’s Journalist Portfolio

Building a World Where Data Privacy Exists Online

Building a World Where Data Privacy Exists Online

Dawn Song, an expert in computer security and trustworthy artificial intelligence, is working on making that …

View article →
Building a World Where Data Privacy Exists Online

Building a World Where Data Privacy Exists Online

Dawn Song, an expert in computer security and trustworthy artificial intelligence, is working on making that …

View article →

Articles

Week Ending 10.4.2026 — Eye On AI

|
By Craig Smith Verified
| Eye On A.I. Verified
Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis Novel view synthesis should teach models about 3D structure, yet encoder-based approaches produce weak representations. The authors blame two architectural choices: decoders that are too spatially expressive, which dilute the encoder's role, and pixel-space targets, which hinder feature learning.
Open in Who Shared Issue with article?

Week Ending 9.27.2026 — Eye On AI

|
By Craig Smith Verified
| Eye On A.I. Verified
Learning to Stop without Learning to Stop: Self-Supervised Confidence Training Improves Reasoning Efficiency Reasoning models often produce very long traces, which makes inference expensive. Existing fixes either stop generation early at inference time or train with length penalties. This paper takes a different route. It fine-tunes models to predict their own confidence at intermediate points in their reasoning, using only 600 training problems and no length-related objective.
Open in Who Shared Issue with article?

Week Ending 9.20.2026 — Eye On AI

|
By Craig Smith Verified
| Eye On A.I. Verified
Designer-RSI: Evolving Procedural Memory from User Traffic for Agentic Graphic Design Professional graphic design demands many interdependent decisions with no clear right answer, making it hard for AI agents to improve without human labels. Designer-RSI addresses this by pairing a frozen frontier model that operates design software through 230+ tools with an external procedural memory that accumulates natural-language design skills from real usage.
Open in Who Shared Issue with article?
See all 270 licensed articles (282 Total)