SciFi Futures
← Capabilities
Computing & Information/Foundational science

Mechanistic interpretability of neural networks

Reverse-engineering the internal algorithms and representations of trained neural networks into human-understandable features and circuits.

State of the art

Sparse-autoencoder dictionary features and attribution graphs extracted from frontier LLMs (2024–2025).

What’s blocking it
  • Explanations cover only a fraction of model computation

Who’s working on it

  • Chris OlahResearcher
    Anthropic

    Founded the circuits agenda; co-led Scaling Monosemanticity (2024) and the attribution-graph work (2025) that traced multi-step reasoning inside production models

  • Open-sourced circuit-tracing and attribution-graph tooling in May 2025 with Neuronpedia, and has made interpretability an explicit corporate priority

  • Neel NandaResearcher
    Google DeepMind

    Leads DeepMind's mechanistic interpretability team and shipped Gemma Scope (2024), the open sparse-autoencoder suite most external work builds on

  • Raised a $50M Series A in April 2025 (Menlo Ventures, with Anthropic participating) to turn interpretability into commercial model-auditing and steering tools

  • Nonprofit research lab building open tools for eliciting and explaining model behaviour, the main independent counterweight to in-house lab interpretability

  • Coefficient Giving (formerly Open Philanthropy) technical AI safety RFPProgram

    Its 2025 request for proposals funded interpretability benchmarks, white-box applications and more transparent architectures, the largest philanthropic pipeline into the field

What to invest in

KindInvestmentScale
researchAutomated interpretability tooling at frontier scale$$ $10M – $100M

Sci-fi concepts in reach

Artificial Friend companion robotindirectHAL 9000 ship computergatingindirectPositronic robots bound by the Three LawsgatingindirectCulture Minds (benevolent superintelligences)directTechnoCore AI civilizationindirectSelf-aware supercomputer (Mike)indirect

Sources