Teaching in the rubble: the education thread at 8am AI

Most of the 8am AI corpus is builders talking shop. The education thread is different — it's the group's anxiety, and its conscience. It runs from casual 2024 experiments to a genuinely hard 2026 question: if the tools can do the work, what is school for?

The thread is anchored by Ying, an educator who joins the conversation and keeps pulling it back to a problem the builders can wave away but she can't: her students.

2024: AI as a thing to learn with

Early on, education shows up as personal exploration. James works through a machine-learning challenge on his own. David talks about AI-assisted learning around his algorithmic-trading research. It's optimistic and individual — AI as a faster way to learn whatever you're curious about. Nobody's worried yet.

2025: the channel problem and the franchise idea

By 2025 two concrete ideas surface. First, where do you even keep up? The group maps the real channels — Reddit (Local Llama), specific YouTubers, newsletters — and notes the uncomfortable truth that academia is the worst channel for timely information, because anything in a curriculum is by definition already accepted dogma, not the edge.

Second, 8am-ai.com as a pattern. David floats the idea of the peer group itself as an open-source, franchisable model — a way for other people to run their own version of this learning loop, outside any institution. The meeting becomes the method.

By autumn, the funding-and-policy layer enters: David reports back from a conference on government, research, and industry in Canadian AI. The thread widens from "how do I keep up" to "how does a whole system keep up."

2026: the foundational-knowledge problem

In 2026 the thread gets hard. The question is no longer logistics; it's whether the next generation will be able to think at all.

The worry, stated repeatedly: students use AI to produce acceptable-looking output without the foundational knowledge to evaluate it — a false confidence that they know things they don't. David's sharp image: his daughter using Google Lens to clear an exam, circumventing the very struggle that builds the skill. Juan's counter-requirement: you have to be able to combat an AI suggestion, to critique an approach — which you can only do if you actually understand the domain.

The group's working answers, none of them tidy:

• Raise the complexity. Make the problems hard enough that AI alone can't clear them, forcing students to use it as a learning aid rather than a substitute.
• Go direct. Bypass the slow institution — direct to student, direct to teacher — because the five-year curriculum cycle can't track a field that moves weekly.
• Teach the meta-skill. Critical thinking, design thinking, first principles — "airdropped into" the curriculum as the durable thing, since the specific tools won't survive the semester.
• Use the students as compute. Assign them to become domain-AI experts, to do "bad research" and then validate it — learning evaluation by doing evaluation.

What the thread is really about

Strip away the specifics and the education thread is asking the same question as the rest of the corpus, just where it hurts most: when the machine can produce the answer, the scarce human skill is judging whether the answer is true. The builders call it "how do you know." Ying calls it the thing her students can't yet do. They're the same problem.

The optimistic reading, which the group keeps returning to: this is a chance to teach the thing schools always claimed to teach and rarely did — genuine critical thinking — because now there's no point teaching anything else.

Open question

If foundational knowledge is what lets you evaluate AI, and AI is how students now acquire knowledge, the loop eats its own tail. The corpus names the trap clearly and hasn't escaped it. Whether anyone can is, maybe, the most important open question in the whole two years.