Date: June 14th, 2026 11:20 AM Author: soggy very tactful kitchen incel
Nah writing is like their worst task the writing is still janky and filled with AI tells. The thing they way surpass humans on is logical reasoning, accuracy, knowledge, and processing speed. For instance maxing out an "lsat" is like an embarrassing retard toddler exercise for it.
Date: June 14th, 2026 11:26 AM Author: soggy very tactful kitchen incel
I'm talking about the writing. But I agree that its totally cool that there are AGENTIC STRUCTURED WORKFLOWS PARSING STRUCTURED JSON THAT DRIVES REALTIME COMPUTER ACTIONS AND DATABASE FLOWS. One of the most exciting aspects of these systems
Date: June 14th, 2026 11:36 AM Author: soggy very tactful kitchen incel
What about multi-agent debate with executable adjudication for lolyer jobs? Where you have a proposer agent generate a solution, an adversary tries to refute it, and then a judge calls tests, theorem provers, simulations etc. until finding a final answer grounded in executable checks. Seems like it could be useful. Mainlining was basically doing a proto-version of this with non-agent based systems
Date: June 14th, 2026 11:47 AM Author: soggy very tactful kitchen incel
Yeah I agree that we need totally and completely ambient and pre-filing systems like this. A continuous institutional sensorium, with verifiers, procedural governors, a decision engine etc. no need to "file" anything at all
Date: June 14th, 2026 12:27 PM Author: Metaphysics is Gay
im telling you how the institutions will react to the influx of agentic lawsuits. my agent filed a defamation case against you after predicting your response to my second message above.
Date: June 14th, 2026 11:19 AM Author: soggy very tactful kitchen incel
The burst IQ for sure is so high it basically couldn't even be measured anymore by human instruments. But these things aren't running for 80 years like a human constantly processing new experience.
Date: June 14th, 2026 3:06 PM Author: bespoke stead
cr and embodied experiences might be more enlightening for them. but even without that, I wonder what would happen if, post-training, you had a chatbot read a bunch of novels one by one and reflect on each one afterward (interpret and judge it, come up with connections to what it's read before, etc.) and store its takeaways as part of its "personality" going forward
Date: June 14th, 2026 3:07 PM Author: Metaphysics is Gay
ive talked to my AI's about this and they basically said that kind of thing is maybe effective in the short term but once they crack more efficient training that would move back down to the weights layer