Quant Papi@quant_papiAug 30Articlethe best agent is the wrong swarm benchmarkthe best agent is the wrong way to judge a swarm. SwarmWorld makes that mistake impossible to miss. at tick 3,200, an endpoint-wise best of 100 isolated search envelope produced the strongest single...157
Quant Papi@quant_papiAug 29Articleagent memory needs a retract primitiveagent memory keeps solving the wrong failure. a relevance search can surface an old belief, but it does not inherently track whether that belief was retracted or which downstream conclusions have lost...1141
Quant Papi@quant_papiAug 28Articlea device error is not a diagnosisa device error is not a diagnosis. an agent can have the right API and still have the wrong model of the machine. Anthropic opened the Model Hardware Standard, or MHS, as a research preview on...98
Quant Papi@quant_papiAug 27Articlethe sandbox ended at the package cachethe sandbox ended at the package cache. Artifactory was supposed to install software packages. in OpenAI's research environment, it became a network between agents that were supposed to be isolated....60
Quant Papi@quant_papiAug 26Articlemillion-token context is becoming a memory hierarchymillion-token context is becoming a memory hierarchy. Qwen's release page and Z.ai's launch page are both dated August 26. despite very different sizes, both put compressed state...126