There has to be a way to combine these into some form of an armed transport like a Voltron of Merch. Just an idea, @elonmusk

Having become our brother, the Son of God looks at the people; He looks at humanity and sees the oppression that burdens and the violence that causes strength to fade. He sees the wounds of war and the emptiness of consumerism. He sees faces reduced to masks, families torn apart by evil, and young people misled by false ideals. Jesus sees and loves. He loves and suffers for and with us: His compassion expresses not only fraternal closeness, but His desire to redeem. #GospelOfToday (Mt 9:36 – 10:8)

open.substack.com/pub/heyros… The central problem with agentic AI in the workplace is not whether it can be useful. It is clearly useful. The problem is that the commercial positioning has moved faster than the accountability model. AI systems are now being described and sold as agents that can perform work, write code, complete workflows, summarize documents, answer business questions, and take action inside enterprise systems. That is no longer the same claim as “this is a productivity tool that helps a human think faster.” Once a system is marketed as doing work, reliance is no longer an accident. Reliance is part of the product promise. That changes the liability discussion. In a normal employment context, work comes with accountability. A human employee can be trained, corrected, reviewed, demoted, terminated, or held professionally responsible. The organization can assign authority and responsibility through role definitions, compensation, supervision, and consequence. That structure is the backbone of how work functions. Agentic AI breaks that structure because it can produce work-like output without carrying any of the human accountability that normally attaches to work. The result is an accountability gap. The AI provider captures revenue from the system. The employer captures the productivity benefit. The individual employee may absorb the failure when the AI produces a bad answer, bad code, bad analysis, or a bad recommendation. That is not a stable or morally coherent model. It is a way of distributing productivity gains upward while pushing operational risk downward. The issue is made worse by the way these systems communicate. LLMs do not merely make errors. They frequently make errors in the same tone, structure, and confidence as correct answers. The user is not consistently told whether the answer is based on current source material, stale memory, inference, direct observation, or a plausible guess. A fluent answer creates the impression of reliability even when the underlying basis is weak. That is not just a user education problem. It is a product design problem. This is where the product liability question becomes serious. An LLM is not merely a service in the ordinary sense. It is an engineered product delivered through a service interface. The fact that a company chooses to host the model, meter access, and wrap the experience in SaaS does not change the nature of the underlying system. There are downloadable and self-hosted models in the market, so the existence of LLMs as product-like artifacts is not theoretical. Cloud delivery is a business model, not a complete liability shield. The analogy is straightforward. If a taxi is late, that is a service issue. If the taxi explodes because the vehicle was defectively designed, that is a product issue. The passenger did not buy the car, but the harm still arose from a defective product placed into commercial use. The same distinction should apply to AI. If the service is slow or unavailable, that sounds like a service problem. If the engineered model has a known failure mode that produces confident falsehoods in foreseeable work contexts, that looks much more like a product problem... [full article on Substack]

Lower level, non-mythos LLM’s are terrible at math problems because of the way numbers are encoded as tokens in the LLM, this bit is clear to me. But do we presume all humans process the symbols of numbers as math or are people that are simply bad at math encoding the information differently in their brains?

Anthropic Mythos and Fable 5 are now on the same level as nuclear weapons. So… maybe less vibe coding?