Self-correction and instruction-following remain inconsistent across ALL models. It's not that progress isn't real, but pretending these systems are reliable is how bubbles form. The honest path is is pushing for better foundations (better data, reasoning loops, evaluation, etc.)
Do you actually trust AI with anything important? This is @claudeai admitting — in its own words — that its "verification system" is mostly smoke and mirrors.
I personally don't think anyone should trust current frontier models "wholeheartedly" on high-stakes tasks without heavy verification. This is a reminder that many systems still rely heavily on prompting theater rather than robust, verifiable architecture.
@billybinion Yeah... I just got hit with a $750 bill in tariffs and fees from @FedEx for a small 15-item order from Japan. So how exactly is Japan the one paying for this? Spoiler: They're not—US buyers like me foot the bill. #TradeRealityCheck