Tag: AI

A year ago, most discussions were about which model was smartest. Now it increasingly feels like the bigger differentiators are becoming: latency orchestration context handling reliability inference economics developer workflow deployment flexibility The interesting shift is that model quality is improving across the board fast enough that “best benchmark” doesn’t automatically translate into “best real-world experience” anymore. We’re seeing more teams optimize around: workload routing hybrid local/cloud setups smaller specialized models faster iteration cycles predictable scaling costs In a weird way, AI feels like it’s maturing into a systems/infrastructure problem almost as much as a model problem. Curious if others are seeing the same shift or if frontier model capability still do...

What is the “personality” of an LLM? What actually differentiates models psychometrically? Since LLMs entered public use, researchers have been giving them psychometric questionnaires, with mixed results. Their answers often do not seem to reflect the same psychological constructs these tests measure in humans. So we asked a slightly different question: What do LLM responses to psychometric questionnaires actually reflect? We analyzed responses to 45 validated psychometric questionnaires completed by 50 different LLMs. The strongest source of variation was whether a model endorsed items about inner experience: emotions, sensations, thoughts, imagery, empathy, and other forms of first-person experience. We call this factor the Pinocchio Dimension. Importantly, the Pinocchio Dimension is not...