d_model GTM brain

d_model partners with frontier AI labs to build reinforcement learning environments that train models to conduct interpretability and alignment research.

category

AI interpretability research and RL environments

market profile

  • text: Frontier model labs need ways to understand and steer increasingly capable models; alignment remains an open research problem.
  • text: Specialist vendors are emerging to supply reinforcement learning environments for model developers, with different task domains.

personas

  • title: Head of interpretability research at a frontier lab
    pains:
    • Mechanistic analysis requires scarce specialist researchers
    • Research agents need to discover meaningful insights beyond fixed benchmarks
  • title: AI alignment and safety research lead
    pains:
    • Model steering and reliable understanding remain unsolved
    • Must evaluate whether automated research advances safety rather than produces superficial results
  • title: Post-training research lead
    pains:
    • Building open-ended research environments requires domain expertise and compute
    • Poor rewards can encourage shortcuts rather than real research progress

icp

  • text: Frontier AI labs training their own models and investing in alignment and interpretability.
  • text: Research organizations with post-training infrastructure and capacity to evaluate specialized agent environments; size is not publicly specified.

message pillars

  • text: Train models to become interpretability and alignment researchers.
  • text: Build environments for open-ended research.
  • text: Make model safety and steerability a research objective.

objections

  • objection: Can agents produce substantive research instead of gaming a benchmark?
    response: Review the published concept-editing work and evaluate on held-out research tasks and expert review.
  • objection: How will environments fit our model training and evaluation pipeline?
    response: Scope a research partnership around model access, infrastructure, evaluation and confidentiality requirements.
  • objection: Is the work relevant to our specific safety priorities?
    response: Compare the lab's interpretability research and proposed tasks with the team's concrete alignment agenda.

competitors

  • name: Goodfire
    why: Interpretability research and tools also help labs understand model internals; its product differs from d_model's RL environment focus.
  • name: Abundant
    why: Supplies RL environments and evaluation benchmarks to AI model developers, with broader task domains.

influencers

  • name: Anish Tondwalkar
    url: https://www.alignmentforum.org/users/anish-tondwalkar
    why: Co-founder and interpretability researcher publishing on model behavior.

keywords

  • mechanistic interpretability
  • AI alignment research
  • RL environments for AI research
  • model steering
  • automated interpretability
  • frontier model safety
  • concept erasure

voice

  • Research-led and technical
  • Direct about unsolved safety problems
  • Concise

signals

  • text: A frontier lab is hiring interpretability researchers and investing in model post-training.
  • text: A lab is exploring agents that can conduct open-ended model analysis.

pricing

model: unknown

motion

sales_led

channels

  • Research publications
  • research partnerships
  • LinkedIn
  • events

confidence

medium

Founders

  • name: Daniel Moon
    title: Founder/COO
    role: COO
    bio: COO @ d_model. 1st Gen MIT, EE BS, EECS MEng.
  • name: Anish Tondwalkar
    title: Founder
    role: Founder
    bio: CEO @ dmodel I study core problems in artificial intelligence and ontology. I spend most of my time outside work trying to find excuses to move tiny marks around on pieces of paper.

Back to YC companies