4 ms·
I do not have evidence or data that supports this. It is only my thought. Claude, since Opus 5, speaks more and more like a wannabe-thought-leader pontificatin
by nrmitchi 2mo ago
I do not have evidence or data that supports this. It is only my thought.
Claude, since Opus 5, speaks more and more like a wannabe-thought-leader pontificating on social media for engagement. Everything is a bait-then-switch, or a multi-post story format. The "engagement" that works well for social media makes actual work extremely frustrating.
My unsupported belief is that this is caused by an obnoxious number of people using previous models in an attempt to automate social media engagement, they figured out what worked, and that was fed directly back into newer model training (either by using thought traces in training, or just by continuing to scrape social media content)
- lucisferre 2mo agoEvidence or not, this is probably the most satisfying explanation I've heard for this behaviour. My own suspicion is that LLMs are not becoming more general-purpose over time, as these companies had hoped, and ongoing development in that direction is stalling. They will likely need to move in the direction of more specialization and train LLMs for specific use-cases. However, this would also be admitting that they are not on a direct path to AGI.
- nrmitchi 2mo agoTo be clear I do not believe it is intentional (nor do I believe I am somehow smarter or more knowledgable than the people running these processes). I believe it is (very) unfortunately converging on the communication style that currently makes the most money as far as publicly available communication goes. Unfortunate, but not entirely unexpected.
- JacobAsmuth 2mo agoUnlikely. Much more likely is that Opus 5 was trained in an RL environment with subagents, and it learned to talk this way when reporting progress to the invoking agent.