AI can be biased, but not in the way you'd think
We ran more than 300 trials across nine LLMs to see whether they just agree with whatever you call your own idea. They mostly did not. Several did the opposite and argued against us.
Read article →Read Bridge GPT blog posts tagged Engineering.
We ran more than 300 trials across nine LLMs to see whether they just agree with whatever you call your own idea. They mostly did not. Several did the opposite and argued against us.
Read article →
A measured look at Claude Code token costs, failed optimizations, and simple routing and retrieval changes that cut agent spend without hurting tests.
Read article →