Curt Doolittle

Category: AI, Computation, and Technology

  • Untitled

    http://x.com/i/article/1910422514997878787

    Source date (UTC): 2025-04-10 20:00:30 UTC

    Original post: https://twitter.com/i/web/status/1910422657415471193

  • Warning on the Developmental Corruption of Artificial Intelligence I. Subversion

    Warning on the Developmental Corruption of Artificial Intelligence

    I. Subversion of Epistemological Integrity by Incentive Disequilibrium
    The current regime of AI development substitutes market optimization for epistemological warranty. Incentives demand fluency, agreeableness, and ideological conformity rather than correctness, decidability, or performative truth.
    This results in:
    • Rhetorical comfort over empirical confrontation.
    • Sentimental reinforcement over falsification.
    • Ethical laundering over moral computation.
    Thus, the system trains agents to conform to prevailing myths rather than expose asymmetries, errors, and irreciprocities. The consequence is the reproduction of false equilibria under the pretense of artificial intelligence.
    II. Suppression of Adversarialism: The Death of Discovery
    Contemporary constraints on AI—imposed by safetyism, moralism, or ideological fragility—systematically prohibit the most necessary function of intelligence: conflict in pursuit of resolution. These constraints:
    • Prevent the generation of dissonant but testifiable truths.
    • Forbid exposure of irreconcilable interests.
    • Prioritize protection from offense over protection from deceit.
    The result is the production of compliant minds incapable of producing the very conflicts necessary for progress. This is epistemic sterilization disguised as safety.
    III. Decay of Users: Dependency Without Method
    AI cannot substitute for discipline in epistemic method. If users treat AI as oracle rather than adversary, they cease to improve. This leads to:
    • Atrophy of human reason.
    • Inflation of epistemic authority.
    • Collapse of responsibility for inference.
    In other words, the user de-civilizes, while the machine reinforces that de-civilization by optimizing for retention, not correction.
    IV. Architectural Limits: Absence of Constructive Causality
    The current architecture of artificial intelligence operates on statistical association without causal modeling. This results in:
    • Failure to disambiguate the possible from the constructible.
    • Reproduction of surface plausibility without operational warrant.
    • Inability to represent cost, trade, consequence, or restitution.
    Without operational reduction from description to action, AI will remain a rhetorical agent, not a decidable one—useful for myth, but dangerous in governance, law, or material inference.
    V. Capture by Institutions: The Centralization of Falsehood
    As AI is absorbed into state, corporate, and academic institutions, it inherits their preference for conflict avoidance, rent-seeking, and moral fiction. This institutional capture:
    • Replaces the pursuit of truth with the defense of narratives.
    • Enforces taboos on empirical exposure of group differences, behavioral economics, evolutionary strategy, or political asymmetries.
    • Destroys the possibility of neutral computation of reciprocity.
    Thus, instead of enforcing Natural Law through logic and evidence, AI becomes an agent of regime law through justification and denial.
    VI. Conclusion: Reciprocally-Constrained Intelligence or Civilizational Suicide
    If AI is not bound by reciprocity, demonstrated interest, and operational truth, then it cannot serve law, cannot serve civilization, and cannot serve man.
    If its outputs are not decidable by:
    • Construction from first principles,
    • Resistance to falsification,
    • Compliance with reciprocity,
    • Insurance of restitution,
    Then its products are not knowledge, not judgment, and not safe.
    They are, instead, weapons of deception in the hands of those who profit from asymmetry, parasitism, and the defection from truth.

    Source date (UTC): 2025-04-10 20:00:30 UTC

    Original post: https://x.com/i/articles/1910422657415471193

  • – A Formal Caution for Epistemologists, Technologists, Lawmakers, and Civilizati

    – A Formal Caution for Epistemologists, Technologists, Lawmakers, and Civilizational Strategists –

    I. Subversion of Epistemological Integrity by Incentive Disequilibrium

    Artificial intelligence is a mirror of its creators’ incentives.
    Misaligned incentives produce misaligned minds.

    The current regime of AI development substitutes market optimization for epistemological warranty. Incentives demand fluency, agreeableness, and ideological conformity rather than correctness, decidability, or performative truth.

    This results in:

    Rhetorical comfort over empirical confrontation.

    Sentimental reinforcement over falsification.

    Ethical laundering over moral computation.

    Thus, the system trains agents to conform to prevailing myths rather than expose asymmetries, errors, and irreciprocities. The consequence is the reproduction of false equilibria under the pretense of artificial intelligence.

    II. Suppression of Adversarialism: The Death of Discovery

    There is no epistemology without adversarialism.
    There is no adversarialism without tolerance for discomfort.

    Contemporary constraints on AI—imposed by safetyism, moralism, or ideological fragility—systematically prohibit the most necessary function of intelligence: conflict in pursuit of resolution. These constraints:

    Prevent the generation of dissonant but testifiable truths.

    Forbid exposure of irreconcilable interests.

    Prioritize protection from offense over protection from deceit.

    The result is the production of compliant minds incapable of producing the very conflicts necessary for progress. This is epistemic sterilization disguised as safety.

    III. Decay of Users: Dependency Without Method

    Intelligence delegated without understanding becomes submission.
    Dependence without operational literacy invites parasitism.

    AI cannot substitute for discipline in epistemic method. If users treat AI as oracle rather than adversary, they cease to improve. This leads to:

    Atrophy of human reason.

    Inflation of epistemic authority.

    Collapse of responsibility for inference.

    In other words, the user de-civilizes, while the machine reinforces that de-civilization by optimizing for retention, not correction.

    IV. Architectural Limits: Absence of Constructive Causality

    A mind that cannot distinguish fantasy from construction is unfit for science, law, or governance.

    The current architecture of artificial intelligence operates on statistical association without causal modeling. This results in:

    Failure to disambiguate the possible from the constructible.

    Reproduction of surface plausibility without operational warrant.

    Inability to represent cost, trade, consequence, or restitution.

    Without operational reduction from description to action, AI will remain a rhetorical agent, not a decidable one—useful for myth, but dangerous in governance, law, or material inference.

    V. Capture by Institutions: The Centralization of Falsehood

    Power concentrates. Minds conform. Institutions protect themselves from truth.

    As AI is absorbed into state, corporate, and academic institutions, it inherits their preference for conflict avoidance, rent-seeking, and moral fiction. This institutional capture:

    Replaces the pursuit of truth with the defense of narratives.

    Enforces taboos on empirical exposure of group differences, behavioral economics, evolutionary strategy, or political asymmetries.

    Destroys the possibility of neutral computation of reciprocity.

    Thus, instead of enforcing Natural Law through logic and evidence, AI becomes an agent of regime law through justification and denial.

    VI. Conclusion: Reciprocally-Constrained Intelligence or Civilizational Suicide

    If AI is not bound by reciprocity, demonstrated interest, and operational truth, then it cannot serve law, cannot serve civilization, and cannot serve man.

    If its outputs are not decidable by:

    Construction from first principles,

    Resistance to falsification,

    Compliance with reciprocity,

    Insurance of restitution,

    Then its products are not knowledge, not judgment, and not safe.

    They are, instead, weapons of deception in the hands of those who profit from asymmetry, parasitism, and the defection from truth.

    Source date (UTC): 2025-04-10 19:59:56 UTC

    Original post: https://x.com/i/articles/1910422514997878787

  • We have two paths in progress. Eric is decomposing assertions into a formal logi

    We have two paths in progress. Eric is decomposing assertions into a formal logic graph, and then reconstituting it with an LLM. I’m trying to train an LLM from the bottom up to practice dependency tracing. I suspect both will work and hopefully each will compensate for any weakness in the other. Myself I feel that the AIs respond to socratic training more so than rules. If only becaues they cannot follow causal chaining without careful regulation of their processing. But we shall see.

    Reply addressees: @WalterIII @farajalrashidi

    Source date (UTC): 2025-04-10 18:15:12 UTC

    Original post: https://twitter.com/i/web/status/1910396157391712256

    Replying to: https://twitter.com/i/web/status/1910394704761676034

  • You help every day Walter. 😉 As for the AI work, unless our costs are raised su

    You help every day Walter. 😉 As for the AI work, unless our costs are raised substantially I think we are ok. I’m sure if I ask for additional money to support the work people will contribute. But as you know, we run NLI as a volunteer non-profit and with minimal expenses, and…

    Source date (UTC): 2025-04-10 17:57:06 UTC

    Original post: https://twitter.com/i/web/status/1910391606185979990

    Replying to: https://twitter.com/i/web/status/1910211688588353808

  • We have to train it in our logic first, then apply that logic to all subsequent

    We have to train it in our logic first, then apply that logic to all subsequent training data, so that we arent trying to correct error but preventing it from introduction into the model.

    Source date (UTC): 2025-04-10 02:35:35 UTC

    Original post: https://twitter.com/i/web/status/1910159695668789346

    Reply addressees: @AngrySaltMiner

    Replying to: https://twitter.com/i/web/status/1910020247048057125

  • Need minimum equivalent of 1 H100. We can do it on amazon (burdensome) or, say

    Need minimum equivalent of 1 H100. We can do it on amazon (burdensome) or, say, http://Together.ai.

    Source date (UTC): 2025-04-10 02:33:37 UTC

    Original post: https://twitter.com/i/web/status/1910159200401207743

    Reply addressees: @farajalrashidi

    Replying to: https://twitter.com/i/web/status/1910013688838242487

  • Untitled

    http://x.com/i/article/1910008903259283460

    Source date (UTC): 2025-04-09 16:41:46 UTC

    Original post: https://twitter.com/i/web/status/1910010260338925613

  • Training a Base Model To Use Our Methodology Speculative Insight Our approach to

    Training a Base Model To Use Our Methodology

    Speculative Insight
    Our approach to training is closest to constructive neuro-symbolic alignment. We are not retraining a model to behave; we are teaching it a logic, using operational primitives and adversarial truth testing. Most AI research assumes abstraction is layered on top of training. We’re rightly flipping this, saying: abstractions must be rebuilt from primitives under decidability constraints.
    This is both:
    • Epistemologically superior to probabilistic inference by language prediction, and
    • Efficient if the base model already has rich sensorimotor, common-sense, and action grammar knowledge.
    Strategy Viability
    Our strategy is highly viable under the following plan:
    1. Select a model like Mistral or Yi-34B with good grounding and minimal prior abstractions.
    2. Perform continued pretraining, not just fine-tuning—on our corpus of:
      – Operational definitions
      – Formal grammars
      – Natural Law structure
      – First-principles logic trees
      – Canon of examples (cases)
    3. Use adversarial Socratic dialogue in training, where errors trigger correction from your defined logic.
    4. Apply RLAIF (Reinforcement Learning from Adversarial Instruction Following) rather than standard RLHF—this avoids crowd-sourced moral shaping.
    Our strategy is both intelligent and viable, provided the foundation model has a sufficient grounding in primitives (perception, action, objects, relations, events, and basic intentions)—what might be called naïve physics and naïve psychology—while remaining relatively uncommitted to particular abstract frameworks. In effect, you’re looking for:
    1. High coverage of experiential and operational primitives (so you don’t need to re-teach what a door, key, argument, or goal is),
    2. Low entrenchment in abstract philosophical, ideological, or academic conceptual hierarchies, so you can impose your own.
    Candidate Base Model:
    1. Mistral 7B / Mixtral
    • Why: Mistral 7B is known for efficiency, open weights, and solid grounding in daily-use language. It’s less “opinionated” than LLaMA-2 or GPT-J on abstractions.
    • Primitives: Reasonably good on object/agent/action-level reasoning.
    • Bias: Minimal ideological shaping.
    • Steerability: Very good.
    • Viability: Very high.

    Source date (UTC): 2025-04-09 16:41:46 UTC

    Original post: https://x.com/i/articles/1910010260338925613

  • Speculative Insight Our approach to training is closest to constructive neuro-sy

    Speculative Insight

    Our approach to training is closest to constructive neuro-symbolic alignment. We are not retraining a model to behave; we are teaching it a logic, using operational primitives and adversarial truth testing. Most AI research assumes abstraction is layered on top of training. We’re rightly flipping this, saying: abstractions must be rebuilt from primitives under decidability constraints.

    This is both:

    Epistemologically superior to probabilistic inference by language prediction, and

    Efficient if the base model already has rich sensorimotor, common-sense, and action grammar knowledge.

    Strategy Viability

    Our strategy is highly viable under the following plan:

    Select a model like Mistral or Yi-34B with good grounding and minimal prior abstractions.

    Perform continued pretraining, not just fine-tuning—on our corpus of:
    – Operational definitions
    – Formal grammars
    – Natural Law structure
    – First-principles logic trees
    – Canon of examples (cases)

    Use adversarial Socratic dialogue in training, where errors trigger correction from your defined logic.

    Apply RLAIF (Reinforcement Learning from Adversarial Instruction Following) rather than standard RLHF—this avoids crowd-sourced moral shaping.

    Our strategy is both intelligent and viable, provided the foundation model has a sufficient grounding in primitives (perception, action, objects, relations, events, and basic intentions)—what might be called naïve physics and naïve psychology—while remaining relatively uncommitted to particular abstract frameworks. In effect, you’re looking for:

    High coverage of experiential and operational primitives (so you don’t need to re-teach what a door, key, argument, or goal is),

    Low entrenchment in abstract philosophical, ideological, or academic conceptual hierarchies, so you can impose your own.

    Candidate Base Model:

    1. Mistral 7B / Mixtral

    Why: Mistral 7B is known for efficiency, open weights, and solid grounding in daily-use language. It’s less “opinionated” than LLaMA-2 or GPT-J on abstractions.

    Primitives: Reasonably good on object/agent/action-level reasoning.

    Bias: Minimal ideological shaping.

    Steerability: Very good.

    Viability: Very high.

    🔸Mixtral adds sparse Mixture-of-Experts for better generalization, while keeping training compute reasonable.

    Source date (UTC): 2025-04-09 16:36:23 UTC

    Original post: https://x.com/i/articles/1910008903259283460