Yoctoloop Labs

An applied research lab

Human learning is too slow.
We're fixing that.

Yoctoloop Labs builds products that accelerate human learning.

The gap between what a person could learn and what they do learn may be the largest untapped resource on earth. It is not a talent gap. It is a tooling gap.

For a decade, much of the best engineering in the world went into capturing attention. The same machinery can be pointed the other way. The tools that captured attention can be turned to building skill.

That is the whole company.

The thesis

The more we lean on AI, the more human learning matters.

Every year, more of what people used to know is a prompt away. That is a gift, and it is a trap: in one large classroom trial, students who practised with an unrestricted chatbot did worse on the exam they took without it.1 A person who can no longer do the thing cannot judge the answer, cannot catch the error, and cannot do the next thing nobody has asked for yet.

Yoctoloop is the answer to that trap. We use the same models that make it easy to stop learning to make it faster to keep learning: precise, personal, on your schedule, with a deliberate end.

Not a classroom. Not a chatbot. A loop: the machine finds the smallest thing you do not know yet, you learn it, it checks that you kept it,3 and it comes back at exactly the right moment.4 Yocto is a septillionth: for thirty years it was the smallest prefix there was.5 That is the size of each step. Loops are how the steps compound.

What we build
Products that accelerate human learning
How
The science of memory, applied with modern models
For whom
Anyone who has to learn fast and keep it
Status
Building. Nothing to try yet.

How we build

Four principles.

  1. Attention is borrowed, not taken.

    Every minute someone gives us is a loan against the rest of their life. We pay it back with something they keep.

  2. Memory is a schedule.

    Forgetting follows a curve, and the curve can be bent.2 The right fact at the right hour beats the same fact crammed.4

  3. Done is a feature.

    A feed with no end is a trap. Ours stops, says so, and sends you back to your day.

  4. Measure retention, not minutes.

    Time on screen is the wrong number. We ship against what people still know a month later.

Now is the time.

The models are getting good enough to teach; early trials say so.6 The phones are in nearly every pocket.7 The science of memory is more than a century old, and in our view most of it has still not been shipped. We are shipping it.

Status Building. There is nothing to sign up for yet; this page is the whole company for now.

Sources

  1. Bastani et al., “Generative AI without guardrails can harm learning: Evidence from high school mathematics”, PNAS 122(26), 2025. About 1,000 students in one Turkish high school; those given an unrestricted GPT-4 chat did better on practice and worse on the unassisted exam. A tutor built with learning safeguards largely avoided the harm. doi.org/10.1073/pnas.2422633122
  2. Ebbinghaus, Über das Gedächtnis (1885), replicated by Murre and Dros, “Replication and Analysis of Ebbinghaus’ Forgetting Curve”, PLOS ONE 10(7), 2015. doi.org/10.1371/journal.pone.0120644
  3. Roediger and Karpicke, “Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention”, Psychological Science 17(3), 2006. Recalling a passage beat rereading it on tests two days and one week later. doi.org/10.1111/j.1467-9280.2006.01693.x
  4. Cepeda, Pashler, Vul, Wixted and Rohrer, “Distributed practice in verbal recall tasks: A review and quantitative synthesis”, Psychological Bulletin 132(3), 2006. A meta-analysis of 839 assessments: spaced study beats massed study, and the best gap grows with how long you need to remember. doi.org/10.1037/0033-2909.132.3.354
  5. Yocto (10⁻²⁴) was the smallest SI prefix from 1991 until the 27th General Conference on Weights and Measures added ronto (10⁻²⁷) and quecto (10⁻³⁰) in November 2022. bipm.org, CGPM Resolution 3 (2022)
  6. Kestin et al., “AI tutoring outperforms in-class active learning”, Scientific Reports 15, 2025. A randomised trial of 194 students in one introductory physics course: an AI tutor with pedagogical guardrails produced larger gains than an active-learning class, in less time. One course, two topics, short-term tests. doi.org/10.1038/s41598-025-97652-6
  7. Pew Research Center, Mobile Fact Sheet, 2025: 91% of US adults own a smartphone (survey of 5,022 adults, February–June 2025). pewresearch.org