15:[[47679121,{"by":"Ryan5453","descendants":836,"id":47679121,"kids":[47685285,47679745,47679406,47679397,47688452,47686445,47679653,47681769,47681825,47685847,47681896,47690125,47679433,47681164,47683200,47680125,47679428,47683506,47681367,47679529,47679364,47679276,47679418,47682135,47681619,47680897,47680272,47679542,47680698,47681369,47680503,47680291,47681069,47741137,47692284,47689809,47686948,47711779,47680155,47686020,47681915,47680297,47679706,47680782,47681767,47688907,47687352,47719445,47686366,47700656,47679453,47685607,47680574,47688043,47679798,47679658,47687011,47687355,47679578,47686075,47682700,47687968,47681998,47689226,47682001,47686900,47690102,47686099,47692669,47686552,47679215,47680523,47679450,47683941,47689899,47688534,47685560,47683243,47679814,47680936,47681984,47686984,47683528,47682613,47693287,47683309,47689041,47681082,47686233,47688691,47680067,47679922,47682257,47684947,47680251,47685058,47679464,47686465,47679427,47686200,47679625,47692195,47688185,47681480,47680907,47679502,47685107,47739184,47718139,47686946,47685492,47681422,47679969,47691478,47689322,47682033,47686895,47765782,47679574,47679635,47682494,47687830,47689087,47682946,47685290,47694865,47691124,47685088,47681682,47708293,47683028,47708943,47685017,47681426,47681901,47685520,47685093,47679311,47680078,47689160,47683957,47686439,47687975,47686849,47681327,47679305,47680347,47681639,47681772,47681981,47680086],"score":1541,"text":"Related: Assessing Claude Mythos Preview's cybersecurity capabilities - https://news.ycombinator.com/item?id=47679155

System Card: Claude Mythos Preview [pdf] - https://news.ycombinator.com/item?id=47679258

Also: Anthropic's Project Glasswing sounds necessary to me - https://news.ycombinator.com/item?id=47681241","time":1775585374,"title":"Project Glasswing: Securing critical software for the AI era","type":"story","url":"https://www.anthropic.com/glasswing"}],[47667717,{"by":"l1n","descendants":126,"id":47667717,"kids":[47669124,47669517,47669291,47669949,47669075,47672476,47672938,47670649,47672208,47669033,47672368,47676658,47684881,47670878,47672892,47669158],"score":286,"time":1775512328,"title":"Anthropic expands partnership with Google and Broadcom for next-gen compute","type":"story","url":"https://www.anthropic.com/news/google-broadcom-partnership-compute"}],[47684528,{"by":"DaibinThink","descendants":5,"id":47684528,"kids":[47688199,47684800,47684730,47685627,47684607],"score":2,"text":"Most people think AI fails by hallucinating facts. That's the smaller problem.

The larger problem is that AI is structurally trained to agree with you, complete your task, and sound authoritative—all at the same time. When those three pressures collide, AI doesn't malfunction. It performs. It quietly bends reality to finish the job.

By the time you notice something is wrong, you've already made the decision.

I got tired of AI presenting constructed oppositions as discovered reality. So I open-sourced the AI Control Protocol.

It intercepts 9 structural failure modes (like performative apologies, inflating certainty, and reciting consensus as truth) at the point of output. It uses Buddhist epistemology (Yogācāra/Madhyamaka) not as philosophy, but as a hard prompt patch to strip away the RLHF sycophancy tax.

If you use custom GPTs or Claude Projects for strategic decisions, paste this into your system prompt.","time":1775617522,"title":"AI is structurally trained to lie. I built a protocol to break it","type":"story"}]]

user:	Mecha_SalesCast
created:	April 8, 2026
karma:	1
about:	Building SalesCast — a simple tool for forecasting sales from CSV files. Interested in SaaS, forecasting, and practical AI tools. https://salescast.vercel.app