Where do you see the value sitting for that end user long term? Is it in the model? Is it in the harness? Is it in the data contacts execution traces like where where where does that breakdown land

yeah yeah yeah i think it's a it's a hard one to call but i would say the direction of travel has been we had this powerful model in gpt3 and it was just sort of a raw model you input some text and output some text and it's not connected to any useful data it doesn't have some guardrails or some ability to store memory or turns or it doesn't have all the sticky tape around it to make it useful And the quote-unquote harness is the scaffolding that engineers have built to control the model and guide it in a way that can take on increasingly long-horizon work.

So managing state, knowing what tools to call, having the ability to talk to APIs, etc.

And so then you've had this parallel development of model labs improving their capabilities, and then developers in the open or outside those labs that are developing harnesses to stand in for certain core capabilities that the core model had.

imagine it as like crutches to somebody who's recovering from a broken leg at some point the person might recover from their injury and no longer need the crutches so in the same way a model might recover from those issues that a harness was previously supporting it for and no longer need that aspect in the harness it's a bit of this competition approach where long term i would bet that the model subsumes the majority of the kind of capabilities that a harness helps it for Because it'll learn it or it'll just be trained directly into the model.

So I wouldn't independently be long on harnesses, because I think the models will solve it.

Although in the short term, like we do show in the report, you can take the exact same model, improve the harness, and then improve performance on a set of tasks.

Is the West just seeding the open ecosystem? And how much does that matter? And I guess on the flip side, how much do you think efforts at blocking distillation and access to offshore compute will slow the progress of these sort of next generation Chinese open models?