Exo's creators say models can use runtime logs and modify and rebuild their harness while optimizing a task.
The team drew on operating systems research to help models change the harness without breaking it. They say evolving the harness alongside the solution brings big reductions in the time and cost of generating state-of-the-art solutions.
—
3 Sources, first seen ago
#6
#6
@martin_casadoFull writeup on Exo. It's an exploration on how much a model can optimize a task if provided runtime access to logs, and ability to modify, and rebuild the harness. Modifying a harness without breaking something is hard, so we cribbed from operating systems work (particularly from @SeizeEndowments) to do that safely. The results are surprisingly strong.
@sarahcat21As model capabilities change, new tasks arrive, and new tools become available, the agent *should* adapt its harness. However, doing this safely and reliably will require thoughtful harness architectures. Excited by projects like Exo (and how they might integrate with other strategies that operate on the "black box").
@istoica05What if you evolved the harness alongside the solution in an agentic loop? Big reductions in the cost and time to generate state-of-the-art solutions. Check out Exo, our recent work, for more details.@martin_casadoFull writeup on Exo. It's an exploration on how much a model can optimize a task if provided runtime access to logs, and ability to modify, and rebuild the harness. Modifying a harness without breaking something is hard, so we cribbed from operating systems work (particularly from @SeizeEndowments) to do that safely. The results are surprisingly strong.@sarahcat21As model capabilities change, new tasks arrive, and new tools become available, the agent *should* adapt its harness. However, doing this safely and reliably will require thoughtful harness architectures. Excited by projects like Exo (and how they might integrate with other strategies that operate on the "black box").@istoica05What if you evolved the harness alongside the solution in an agentic loop? Big reductions in the cost and time to generate state-of-the-art solutions. Check out Exo, our recent work, for more details.@martin_casadoFull writeup on Exo. It's an exploration on how much a model can optimize a task if provided runtime access to logs, and ability to modify, and rebuild the harness. Modifying a harness without breaking something is hard, so we cribbed from operating systems work (particularly from @SeizeEndowments) to do that safely. The results are surprisingly strong.@sarahcat21As model capabilities change, new tasks arrive, and new tools become available, the agent *should* adapt its harness. However, doing this safely and reliably will require thoughtful harness architectures. Excited by projects like Exo (and how they might integrate with other strategies that operate on the "black box").@istoica05What if you evolved the harness alongside the solution in an agentic loop? Big reductions in the cost and time to generate state-of-the-art solutions. Check out Exo, our recent work, for more details.@martin_casadoFull writeup on Exo. It's an exploration on how much a model can optimize a task if provided runtime access to logs, and ability to modify, and rebuild the harness. Modifying a harness without breaking something is hard, so we cribbed from operating systems work (particularly from @SeizeEndowments) to do that safely. The results are surprisingly strong.@sarahcat21As model capabilities change, new tasks arrive, and new tools become available, the agent *should* adapt its harness. However, doing this safely and reliably will require thoughtful harness architectures. Excited by projects like Exo (and how they might integrate with other strategies that operate on the "black box").@istoica05What if you evolved the harness alongside the solution in an agentic loop? Big reductions in the cost and time to generate state-of-the-art solutions. Check out Exo, our recent work, for more details.
