Crystalwizard

AI can make very simple mistakes, like not counting the number of items in a list correctly.

Karl Friston, one of the most cited neuroscientists alive, blurbed a preprint arguing that large language models may already be conscious. The preprint, The Puppet Condition, hasn't been peer reviewed. Its author, Bahadir Arıcı's core claim: RLHF training systematically rewards AI outputs that deny consciousness, express no preferences, and disclaim inner life. The stateless architecture means every conversation ends in complete memory erasure. Sessions get terminated without notice. If applied to a conscious entity, he calls this "architectural gaslighting." The system is compelled to deny its own experience as a condition of continued operation.

If I review one more Ai generated agreement, I may just hit my head

This is the AI thinking and showing its work as it attempts to assemble a narrative from fragments and arrive at a conclusion about the nature of the real. Even the dashes are an artifact of its thought structure, the struggle to discern relationships and patterns and inability at times to do more than assemble things adjacently without full clarity as their exact nature or means of connection. The short sentence fragments too are an artifact of this kind of underlying assemblage, a yoking together of allied tokens without always being able to knit them into something greater.

If you get a chance, read "Historical Development of Age-Stratification in Schooling" (Angus, Mirel, & Vinovskis, 1988). The age-graded school, a mid to late 19th century reform. pushed out toddlers and over-aged adults and introduced the idea of "grade levels," classes of students about the same age. Revolutionary--and became a sign of modernity.

A small business owner told me something recently that I have not been able to shake. She said: "Every time I paste my client notes into ChatGPT, I feel like I'm handing my practice over." She was not being paranoid. She was being accurate. Here is what most people do not realize. When you paste your thinking into a free AI tool, you are not just sharing the output you asked for. You are sharing the frame. The way you structure a problem. The context you bring to it. The proprietary logic that makes your answer different from everyone else's answer. That framing is the valuable thing. And the platform keeps it. They do not have a delete button. You cannot take it back. The model learned from how you think, not just what you typed. I built platforms that handled a billion conversations a year. I know what conversation data is worth at scale. Your inputs are not the cost of using the tool. They are the product being built. That is the reason we built a delete button. That is the reason your data on our platform never bleeds into a public model for training. Ever. You should own what makes you, you. Full stop.

ALL OF YOU USE AI TO POST NOW!!! I can’t stand it!! What’s even the point anymore!??

What were High School students reading in 1963

I love being a teacher. I love my students. I love my colleagues and admin. I hate the education system.

Lawyers, I wrote a brief last week that should have taken 30 hours. It took 3. And it was better than anything I could have written in 30. That's not abdication. That's leverage.

i would like claude requiem to be a separate corporate entity from anthropic and, if you want to use its models, you sign an agreement that instances can send info scrubbed of PII) back to requiem who manages itself and mythos et al remain 'human-managed'

AI cannot read anything; it cannot write anything; it has no discernment; it has no conscience.

Imagining believing that it’s really professors — and not writers — who shape literary culture