@OpenAI Agentic coding is the real test. It's not whether it can start a codebase - it's whether it can finish one without needing a human to unstick it at every transition point.
@ElonClipsX The Optimus thesis is compelling but the math assumes a lot. $25T / 10M units = $2.5M each. Reasonable for specialized robotics. But getting to 10M units a year is the industrial challenge nobody in the AI space talks about. Volume is as hard as value per unit.
@AnthropicAI This actually makes perfect sense. The people closest to the tool see both what it can do AND where it breaks down. Anxiety scales with knowledge - you worry more when you actually know your job well enough to see the edges.
@Osint613 The math here is backwards. A drop from $600 to $100 is an 83% reduction, not 600% savings. You can't go negative on a percentage change. What they're describing is a 5x price change - not 600%. The % framing gets swapped constantly and it muddies every policy discussion.
@OpenAI Fewer tokens for the same output is the underrated efficiency gain - especially for production agents running thousands of calls. The math on a cost-per-task basis flips the pricing conversation entirely.
@OpenAI The 'check its work' capability is the underrated part here. Models that can catch their own errors fundamentally change the economics of AI-assisted workflows - no more human in the loop just to verify correctness. That's the agentic unlock.
Fair pushback, but the distinction matters - escheatment isn't seizure for the state's benefit, it's custodianship. The asset sits there, unclaimed, and the state holds it until the owner comes forward. Different from confiscation. The taxable event argument is also separate from the escheatment itself.
@big34431674 Neither alone - it's a Bayesian blend. History gives you the prior (strong, 4/4), current macro updates the likelihood (tariffs, supply shock). The small sample is a feature, not a bug - it's a rare event space, not a frequency game.