SEA SALTLABS

INDEPENDENT AI RESEARCH

Investigating how neural systems learn, compute, generalize, and fail.

Explore the research
PLATE 00 / STUDIO INDEXCALIFORNIA / EST. 2026 ACTIVE RESEARCH
Enter the research field
OUR POSITION

Neural networks are not only products to evaluate.

They are computational objects to investigate.
RESEARCH REGISTER / 2026

Four questions
define the lab.

01—04

Sea Salt Labs is not organized around one study. These questions determine which experiments, systems, and evidence we build next.

EXPERIMENT REPORT 001 / JULY 2026

Causal convergence in tiny Transformers

Two independently trained populations solving the same arithmetic task showed more similar causal signatures than populations solving different tasks. The gap was real, modest, and not an extracted algorithm.

Read the complete experiment
0.000
0.394SAME TASK
0.281DIFFERENT TASK
+0.113OBSERVED GAP
2,800INDEPENDENT TRANSFORMERS
THE APPARATUS

Every experiment leaves behind machinery.

Working systems are treated as part of the evidence: versioned, inspectable, and kept separate from the story told about the result.

01population_runner2,800 / 2,800VALIDATED
02causal_intervention_suite14 SITESVALIDATED
03complete_domain_sets10 PARTITIONSCURATING
04symbolic_circuit_extractorCAUSAL REPLACEMENTBUILDING
Inspect the research systems
RESEARCH DISPATCH

Follow the work while it is still becoming evidence.

research@seasaltlabs.com