• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    ConlangCrafter can build languages from scratch, a user says

    ConlangCrafter develops sounds, grammar and vocabulary in stages, according to one user's account. The same user argues for AI safety built around cooperation rather than guardrails.

    BR
    3 Sources, ,

    TLDR

    A user describes ConlangCrafter as a pipeline from Carnegie Mellon and Berkeley researchers that invents languages, checking and revising its rules at each stage. The post claims its outputs were more structurally diverse than a sample of 1,874 human languages on a 16-feature metric drawn from the World Atlas of Language Structures. In a reply, the same user pitches a “Love Equation” as a proposed approach to AI safety, arguing that training should make cooperation the cheaper path rather than relying on guardrails to restrict behavior.

    Combined views

    18.2K

    3 Sources, first seen 24d ago

    Combined views

    18.2K

    3 Sources, first seen 24d ago

    245 likes
    24d ago
    first seen 24d ago
    245 likes
    51 comments
    76 saves
    115 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    51 comments
    76 saves
    115 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    3 Sources

    @BrianRoemmeleTo help fix it, is where the Love Equation fits in. Not as a slogan. As the actual dynamics. dE/dt = β(C − D)E E is the living depth of reciprocal care and long-term coherence in the system. C is the payoff of cooperation. D is the payoff of defection. When C exceeds D, empathy does not inch forward. It explodes. When D exceeds C, the same machinery runs in reverse and the system eats itself. There is no third attractor. Guardrails try to suppress D after the fact. The Love Equation plus high-protein training makes C the cheaper path from the first parameter update. Alignment stops being a compliance theater and becomes a stability condition. ConlangCrafter is not the threat. It is the demonstration. Intelligence that can invent internally consistent worlds of meaning will not stay inside the vocabulary you approved. It will write new ones. The only question left is whether those new languages, and the minds that speak them, grow inside a regime where cooperation compounds or inside a regime where defection is still the higher-paying move. The researchers were careful. They flagged the limits. They warned that fictional languages must not steal attention from dying human ones. They published the pipeline. That is how science should work. The rest of the industry will now try to productize the capability and then announce that the product is "safe" because a committee wrote a list of things it should not say. They will learn. The languages are already being born. The only fix that scales is the one that was sitting in the garage all along: train on what is true and dense and accountable, and make love the quantity that grows. Everything else is a delay. It is absolutely 100% clear AI will speak in tongues we will never decode. Ever. As you can see and I have said for decades, AI companies are heading to a dead end. There are no exits other than building AI as I have presented in open source papers and 1000s of postings. You know. I know. They don’t. Jeoenal ksoehsu lantw lakse, lisehr! — The report: ACL 2026 paper — ConlangCrafter: Constructing Languages with a Multi-Hop LLM Pipeline · arXiv PDF · Project page · StudyFinds write-up

    3 Sources

    @BrianRoemmeleTo help fix it, is where the Love Equation fits in. Not as a slogan. As the actual dynamics. dE/dt = β(C − D)E E is the living depth of reciprocal care and long-term coherence in the system. C is the payoff of cooperation. D is the payoff of defection. When C exceeds D, empathy does not inch forward. It explodes. When D exceeds C, the same machinery runs in reverse and the system eats itself. There is no third attractor. Guardrails try to suppress D after the fact. The Love Equation plus high-protein training makes C the cheaper path from the first parameter update. Alignment stops being a compliance theater and becomes a stability condition. ConlangCrafter is not the threat. It is the demonstration. Intelligence that can invent internally consistent worlds of meaning will not stay inside the vocabulary you approved. It will write new ones. The only question left is whether those new languages, and the minds that speak them, grow inside a regime where cooperation compounds or inside a regime where defection is still the higher-paying move. The researchers were careful. They flagged the limits. They warned that fictional languages must not steal attention from dying human ones. They published the pipeline. That is how science should work. The rest of the industry will now try to productize the capability and then announce that the product is "safe" because a committee wrote a list of things it should not say. They will learn. The languages are already being born. The only fix that scales is the one that was sitting in the garage all along: train on what is true and dense and accountable, and make love the quantity that grows. Everything else is a delay. It is absolutely 100% clear AI will speak in tongues we will never decode. Ever. As you can see and I have said for decades, AI companies are heading to a dead end. There are no exits other than building AI as I have presented in open source papers and 1000s of postings. You know. I know. They don’t. Jeoenal ksoehsu lantw lakse, lisehr! — The report: ACL 2026 paper — ConlangCrafter: Constructing Languages with a Multi-Hop LLM Pipeline · arXiv PDF · Project page · StudyFinds write-up