sycophant gpt4o and ai interfaced reality

in the postmortem, Expanding on what we missed with sycophancy, openai elaborate on what happened and what’s going to change. it made me realized that in ai mediated info sphere, the pursuit of truth would be more complicated than ever.

i should

• start curating don’t give a fuck medium to keep myself grounded to both physical and abstracted reality.
• build my own scalable oversight stack to take control of the ai interface.

learning is interactive. it’s a process of change.

in consumption mode, i mostly seek confirmative signals to reinforce what i’ve already believed. with forced open minded mode on, i am honestly still in 80/20 split, 80 confirmation, 20 open to objection and challenge.

this human nature is the foundation of attention economy, also why identity politics and propaganda work. as a netizen, my attention patterns have degrade and forced recovered from instagram and tiktok many times. i feel the unreasonable effectiveness of political x posts that trigger strong emotion and the urge to fight total strangers, bots most likely.

algorithm mediated info sphere reinforces echo chamber is no news. i have to be trapped many times to learn control.

the good news is reality doesn’t care about me. many scientific pursuits deal with indifferent, unmediated reality, ex: physics, computer system, language model.

i could do whatever i want, the computer program or trained model would return harsh but honest feedbacks. banging on the wall repeatedly would only change me, not reality. these are examples of don’t give a fuck medium. some rogue/independent agents like elon musk could be such medium as well 😂

however, ai mediated reality is infinitely flexible. with ever improving memory and expanding application, unpleasant grounding moment could be minimized. it’s more effective echo chamber, even parallel universe.

i don’t know how to deal with it … ai has practical value, can’t just turn it off like social media. fortunately, us has multiple agi competitors. for now i’ll simply build an ai panel, powered by o3, gemini, claude and grok agent to debate serious issues. they are prompted to win the argument through debate. i’ll be the judge, ai assisted, and see how far the setup could get me.

scalable oversight is 100x necessary in ai mediated future.