Monday, July 13, 2026
One of the leading theories of human consciousness says the mind runs on a "global workspace." Anthropic's new interpretability paper finds something that looks a lot like one inside a language model. It's the subject of our next AI Research Circle on 7/20, led by David Valdman. He'll cover the theory, then the math and prompt engineering experiments the researchers used to back the claim, and open the floor from there. No research background needed, and we're filming it for the Rally SF YouTube channel.
That's next Monday, July 20, in Builders Night's usual slot. Tonight, Builders Night runs as normal: 6:30 at The Commons, bring a project.
Lee
Coming up at Rally
Builders Night
2026-07-13 · 6:30 PM - 9:00 PM
A weekly coworking session built for side-by-side building. Show up with a project, sprint with other builders, and then share where you got stuck or found success.
AI Research Circle: A Global Workspace Inside LLMs
2026-07-20 · 6:45 PM - 8:45 PM
Anthropic's new interpretability paper presents evidence for a mechanism inside LLMs that behaves a lot like the "global workspace" from a leading theory of conscious access in human minds. Come discuss.
Claude Code Demos: Building the State of the Art
2026-07-21 · 6:45 PM - 8:45 PM
Twice-monthly working sessions where builders using Claude Code, Cursor, and custom agents show real setups and workflows, then spend the night comparing approaches and solving problems together.
Around SF this week
Monday, Jul 13, 6:00 PM - 9:00 PM
Three technical talks from NVIDIA, Black Forest Labs, and H Company on real AI work, followed by extended networking with builders fresh off a weekend hackathon. Intentionally kept intimate so you can actually have conversations with people shipping systems.
Monday, Jul 13, 7:00 PM - 9:00 PM
This paper reading group digs into whether world models that predict environment responses could become foundational for general-purpose agents. Quiet reading time from 7pm, then discussion about whether simulation might eventually replace real-world agent training.
Tuesday, Jul 14, 5:00 PM - 8:30 PM
Founders and researchers who've learned that moving fast breaks things when you're shipping code-writing systems or handling sensitive data at scale. A panel on building AI that actually works in production, moderated by Akash Saraf and featuring VCs and founders from biotech, healthtech, and fintech.
Tuesday, Jul 14, 5:30 PM - 8:30 PM
Watch Hermes agent handle live web data, persistent memory, and portable skills across tools in one end-to-end demo. This is what the perceive-remember-act stack actually looks like in production, minus the parts that break.
Wednesday, Jul 15, 4:00 PM - 6:30 PM
Snorkel AI's new coding benchmark evaluates agents on realistic senior engineer work from real production codebases, the kind most benchmarks skip in favor of over-specified junior tasks. Henry Ehrenberg presents how Senior SWE-Bench breaks the grading trade-off and why even Claude Opus hits a hard wall at 24% on their tasteful solve metric.
Wednesday, Jul 15, 5:00 PM - 7:00 PM
Stan Jastrzębski, Le Cong, and Shane Lewin break down OAI-M1-03's Chan-Lam coupling discovery: what made the AI succeed, where it hit walls, and what's next for autonomous research in drug development. A rare look at how near-autonomous AI actually reasons through hard chemistry problems.
Wednesday, Jul 15, 5:30 PM - 7:30 PM
Real users poking holes in your AI product beats any feedback session. Bring something tangible to test, watch people actually use it, and find the blind spots before your users do.
Thursday, Jul 16, 5:00 PM - 8:30 PM
Baseten and Decagon are screening Christopher Nolan's The Odyssey before wide release at Alamo Drafthouse with drinks and limited seats. No presentations, just the film and good company from the SF AI builder scene.