Devoured - May 01, 2026
Silico (3 minute read)

Silico (3 minute read)

AI Read original

Goodfire launches Silico, a platform that uses interpretability techniques to let developers see inside AI models, debug their behavior, and design them more intentionally.

What: Silico is a platform for building AI models with built-in interpretability tools. It includes a "model neuroscientist" autonomous agent that runs experiments, and lets teams decompose models into interpretable features, run health diagnostics, debug failures, and shape model behavior using internal representations.
Why it matters: Most AI models are black boxes, making it hard to understand why they fail or how to improve them beyond trial-and-error training. Interpretability unlocks the ability to debug specific issues, extract scientific insights (like novel Alzheimer's biomarkers), and improve generalization with less data.
Takeaway: Early access is available now at goodfire.ai/platform for teams interested in more transparent model development.
Deep dive
  • Silico brings frontier interpretability techniques to all researchers and engineers, building on Goodfire's work discovering Alzheimer's biomarkers, teaching models to correct hallucinations, and diagnosing robotics bottlenecks
  • The platform includes a model neuroscientist agent that autonomously plans and runs concurrent experiments on models, working alongside human teams
  • Users can decompose models into interpretable features to distinguish real understanding from spurious correlations
  • Health diagnostics catch issues like undertraining, information bottlenecks, and feature collapse before they impact production
  • Debug capabilities let teams precisely identify and remove confounders, diagnosing failures before production deployment
  • Internal features can be used to extract stronger predictors, steer generation, and target generalization unreachable through standard training
  • The platform enables targeting specific learned structures to shift training distribution, objectives, or architecture for better generalization with equal or less data
  • Teams can organize research threads, replicate and extend papers, and collaborate on findings in a shared model design environment
  • Platform is currently in early access following coverage by MIT Technology Review
Decoder
  • Interpretability: The ability to understand and explain how an AI model makes decisions by examining its internal representations and computations
  • Features: Learned internal representations in neural networks that capture patterns and concepts from training data
  • Spurious correlation: False patterns a model learns that happen to correlate with outputs in training data but don't represent true causal relationships
  • Information bottlenecks: Points in a model's architecture where information flow is restricted, limiting performance
  • Feature collapse: A failure mode where multiple distinct inputs map to the same internal representation, losing important distinctions
Original article

Introducing Silico

Introducing Silico: the platform for building AI models with the precision of written software.

Silico lets researchers and engineers see inside their models, debug failures, and intentionally design them from the ground up. Early access is open now.

We've used interpretability to discover a novel class of Alzheimer's biomarkers, teach a language model to correct its own hallucinations, and diagnose performance bottlenecks in a robotics model. Silico brings those frontier techniques to everyone.

Silico introduces our model neuroscientist: an autonomous agent that plans and runs concurrent experiments on your model. It works with your team in our model design environment, where you can organize research threads, replicate and extend papers, and collaborate on findings.

5 Things You Can Do with Silico

See inside your model

Decompose your model into interpretable features and tell the difference between real understanding and spurious correlation.

Check your model's health

Run comprehensive diagnostics on your model's internal representations to catch issues like undertraining, information bottlenecks, and feature collapse before they impact downstream performance.

Debug failures

Precisely debug issues with model behavior, identify and remove confounders, and diagnose failures before they occur in production.

Shape model behavior

Use internal features to extract stronger predictors, steer generation, and target generalization that standard training can't reach.

Generalize from less data

Target the specific learned structures driving behavior — and shift the training distribution, objective, or architecture to generalize further with the same or less data.

MIT Tech Review's @strwbilly spoke with our CEO/co-founder @ericho_goodfire about Silico and what it means for model builders: technologyreview.com/2026/04/30/113…

Silico is in early access now. Learn more at: goodfire.ai/platform