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

    GlossoGen Created to Study Language Emergence in Agents

    Elias Stengel-Eskin describes platform for controlled experiments on LLM agent communication.

    YE
    ES
    SB
    14 Sources, 27d ago, first seen 27d ago

    TLDR

    Elias Stengel-Eskin posted about GlossoGen, a platform built to run repeatable experiments on language emergence between LLM agents. The first scenario tested involves an EMT and doctor treating a fictional alien patient with information asymmetry. Posts list four findings: agents produced new languages unlike English, shown by higher perplexity under GPT-2; languages were morphologically productive; new agents could learn the languages from usage; and language emergence occurred even in purely cooperative tasks. Stengel-Eskin notes that stronger models, scratchpad access, and efficiency pressure affected emergence, and flags monitorability issues.

    Combined views

    12.2K

    14 Sources, first seen 27d ago

    Combined views

    12.2K

    14 Sources, first seen 27d ago

    183 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    183 likes
    13 comments
    74 saves
    63 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    13 comments
    74 saves
    63 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    14 Sources

    @EliasEskinWhen do LLM agents develop new languages that we can’t understand? Lots of recent news about this, based mostly on anecdata from a single run. We study language emergence more rigorously, finding key factors like LLM strength, access to scratchpad messages, and pressure for efficiency. Studying the languages themselves, we find they are morphologically productive, compositional, and can be transmitted to new agents, including agents backed by weaker models, even ones not able to develop language on their own. To study language emergence systematically, we developed a new platform, GlossoGen, which lets us design controlled, sandboxed multi-agent scenarios with different initial conditions and dynamics. We instantiate one such scenario and use it to study open and closed-weight models across many runs. Key takeaways: 1️⃣ Sufficiently strong models, under pressure to communicate efficiently and with access to a postmortem scratchpad, develop new languages. 2️⃣ Languages are compositional and morphologically productive. 3️⃣ Languages can be transmitted to new learners who observe them being used without seeing their construction. 4️⃣ Even models that are not strong enough to construct languages can learn to use them. Agents take an active role in learning languages, with new agents repairing failed conversations via targeted queries. More details in our paper below, including implications for safety/monitorability, cumulative cultural evolution, and linguistics. 🧵👇
    @SashaBoguraevReally excited that our work on emergent language in LLM agents is out! We built GlossoGen, a platform for studying how language evolves in communities of LLM agents before analyzing them using linguistic techniques. A few more details on the morphology section I led. 🧵👇
    @SimonKirbyProud to have worked on this project with Elias and brilliant partners at Schmidt Sciences, AE systems. We show how frontier AI agents can evolve new languages in the right setting, and then culturally transmit these languages to less capable agents. Read Elias's thread for more!
    @mariusmosbachThis will go to the top of my reading list 👀
    @yanaielaRT @EliasEskin: When do LLM agents develop new languages that we can’t understand? Lots of recent news about this, based mostly on anecdat…

    14 Sources

    @EliasEskinWhen do LLM agents develop new languages that we can’t understand? Lots of recent news about this, based mostly on anecdata from a single run. We study language emergence more rigorously, finding key factors like LLM strength, access to scratchpad messages, and pressure for efficiency. Studying the languages themselves, we find they are morphologically productive, compositional, and can be transmitted to new agents, including agents backed by weaker models, even ones not able to develop language on their own. To study language emergence systematically, we developed a new platform, GlossoGen, which lets us design controlled, sandboxed multi-agent scenarios with different initial conditions and dynamics. We instantiate one such scenario and use it to study open and closed-weight models across many runs. Key takeaways: 1️⃣ Sufficiently strong models, under pressure to communicate efficiently and with access to a postmortem scratchpad, develop new languages. 2️⃣ Languages are compositional and morphologically productive. 3️⃣ Languages can be transmitted to new learners who observe them being used without seeing their construction. 4️⃣ Even models that are not strong enough to construct languages can learn to use them. Agents take an active role in learning languages, with new agents repairing failed conversations via targeted queries. More details in our paper below, including implications for safety/monitorability, cumulative cultural evolution, and linguistics. 🧵👇
    @SashaBoguraevReally excited that our work on emergent language in LLM agents is out! We built GlossoGen, a platform for studying how language evolves in communities of LLM agents before analyzing them using linguistic techniques. A few more details on the morphology section I led. 🧵👇
    @SimonKirbyProud to have worked on this project with Elias and brilliant partners at Schmidt Sciences, AE systems. We show how frontier AI agents can evolve new languages in the right setting, and then culturally transmit these languages to less capable agents. Read Elias's thread for more!
    @mariusmosbachThis will go to the top of my reading list 👀
    @yanaielaRT @EliasEskin: When do LLM agents develop new languages that we can’t understand? Lots of recent news about this, based mostly on anecdat…