Connecting 15 open-source research systems to GPT-6 Astra to make them fight
The author had multiple research agents cross-check the same question, left conflicts unresolved, and kept only conclusions traceable to original evidence.
last night i connected 15 open-source research systems to GPT-6 Astra and made them fight over the same answer i expected more sources instead, i started finding contradictions one research run would return a claim with 3 citations another would investigate the exact same claim and come back with different numbers a third would find the original source and make both of them look wrong that gave me an idea instead of feeding all the research into Astra and asking it to make sense of the mess, i stopped letting information move forward without evidence attached every useful claim had to carry its source with it every citation had to point back to the exact evidence that earned it and conflicting findings stayed unresolved instead of being quietly averaged into something that sounded reasonable the whole system started looking like the thing in this video dozens of research paths branching out, finding evidence, colliding with each other and slowly collapsing into one report the weird part was that Astra ended up seeing less information weak findings disappeared duplicate research disappeared claims nobody could properly support disappeared what survived was a much smaller evidence map where i could trace almost any sentence back through the system report → citation → evidence → source that completely changed how i think about AI research i used to spend most of my effort making GPT-6 Astra better at finding information now i spend it making bad information prove it deserves to survive
This post quotes another post: x.com/lummox_eth/status/2100288151897981255