My mother wrote a lot of work on health-preserving technologies in education.
And I have now launched a deep research run: can the technology of health-preserving teaching of people be connected in any way to the technology of training artificial intelligence.
Yes, the artificial intelligence writes to me that this is unploughed ground.
ML, machine learning, has already borrowed order and dosage from pedagogy:
curriculum learning
self-paced learning
BabyLM
spaced repetition and so on
But ML has not borrowed protection of the learner during the process:
diagnostics of the learning state
recovery phases by threshold, not by schedule
ergonomics of the form of the material
accountability over a concept graph
Direct work applying what my mother wrote to the training of ML models: zero. Which means artificial intelligence has not worked on this subject yet.
I am simply very interested: is there any money here at all?
Health-preserving machine pedagogy. It sounds silly, of course, but I am just curious whether anyone would give money for it.
A working formula, out of training stress:
loss instability
gradient instability
forgetting
representation damage
calibration loss
plasticity loss
recovery depth
In general it is interesting to study the subject. Maybe there are grants of some kind?
I dug up some startups in EdTech and artificial intelligence:
InBloom, raised $100 million
Knewton, raised $180 million
Woebot, consumer, $124 million
And all of them shut down. I am just curious: were there any companies after them? It is unclear what they were doing.
And here, look, what a turn: there is an OpenAI Development Grant. A Research Grant. Maybe worth trying to apply for it.
I also found Foresight AI Safety Grants. And there is an office in San Francisco. Emerging Ventures exists too.
I need to make a synthesis of my deep research runs
In short, in this subject you can apply for a grant. And that is good. So, shall we apply for a grant?
And who else among you is familiar with technologies like this? Not simply health-preserving technologies, but robot-health-preserving technologies. Sounds trashy
But it is interesting: maybe the code or the model really will learn better if you do not use it so hard?
No, it is interesting.
And it all started with me looking for... My son goes to school from 8:30, roughly, until 4. I started looking for how long it is actually normal for a child to study. And all my deep research runs said that 6 hours is the maximum.
After that I started having active conversations with my wife about how these schools could be changed.
Then I remembered that my mother happened to write several PhDs on this subject, on health-preserving technologies.
And then I thought: why not make money on this?
Something like that.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20261009.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20261009.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.
🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI
Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.
Top comments (0)