Jenny Sun

2 followers · 245 views

on the atlas — 29

highlights — 31

Watch for the moments you jump to outsourcing or removing the moments of boredom or difficulty. Creativity requires friction
"Instagram Face" is coming for...everything
If the frontier remains crowded, the layer above will be valuable
The Untrainable - Sarah Guo
Illegible value is unfortunately also complicated to sell, for the same reason it’s hard to commoditize: a company can’t tell from the outside whether AI will transform its operations any better than the benchmark can. So the strongest businesses stop trying to prove it externally, get in, and price the outcome instead. Sierra charges when its agent resolves a customer’s issue and nothing when it kicks the problem to a human, so the price becomes the evaluation
The Untrainable - Sarah Guo
trust is built slowly, on relationships, with user’s acquiescence, not gradient descent that erases them
The Untrainable - Sarah Guo
you only get to verify whether AI did something useful inside a system once you’re trusted inside it, after the security review, the integration, the contract with your name on the outcome
The Untrainable - Sarah Guo
Intelligence is not the bottleneck here. Permission is, and so is accountability
The Untrainable - Sarah Guo
engineering has always resisted measurement, and the most measurable parts may not be the only important ones
The Untrainable - Sarah Guo
So when designing this next generation of software, designers / builders should ask: does my user know what they want, or are they figuring it out?
AI UX: The Limits of Declarative Interfaces - Sara Du
The most effective tools aren’t purely declarative or purely procedural; they support execution and discovery, depending on where the user is in their thinking
AI UX: The Limits of Declarative Interfaces - Sara Du
We encourage every startup to measure their progress by weekly growth rate. If you have 100 users, you need to get 10 more next week to grow 10% a week. And while 110 may not seem much better than 100, if you keep growing at 10% a week you'll be surprised how big the numbers get. After a year you'll have 14,000 users, and after 2 years you'll have 2 million
Do Things that Don't Scale
We encourage every startup to measure their progress by weekly growth rate. If you have 100 users, you need to get 10 more next week to grow 10% a week. And while 110 may not seem much better than 100, if you keep growing at 10% a week you'll be surprised how big the numbers get. After a year you'll have 14,000 users, and after 2 years you'll have 2 million.
Do Things that Don't Scale
We encourage every startup to measure their progress by weekly growth rate. If you have 100 users, you need to get 10 more next week to grow 10% a week. And while 110 may not seem much better than 100, if you keep growing at 10% a week you'll be surprised how big the numbers get. After a year you'll have 14,000 users, and after 2 years you'll have 2 million.
Do Things that Don't Scale
The human brain changes constantly as it accumulates new memories through synaptic remodeling and neural pathway evolution
Announcing our investment in Engram, the memory dream team | Amplify Partners
e 4.7 tokenizer produces 32–34% more native tokens than 4.6 for equivalent text
Opus 4.7's New Tokenizer: What It Actually Costs — OpenRouter Blog
everything is a win when the goal is to experience
everything is a win when the goal is to experience
xperiments and hardware design have a certain “latency” and need to be iterated upon a certain “irreducible” number of times in order to learn things that can’t be deduced logically. But massive parallelism may be possible on top of that
Dario Amodei — Machines of Loving Grace
serial dependence” (you need to make discovery A first in order to have the tools or knowledge to make discovery B
Dario Amodei — Machines of Loving Grace
AI can do a better job analyzing your data, but it can’t produce more data or improve the quality of the data. Garbage in, garbage out
Dario Amodei — Machines of Loving Grace
Pudding does not unstir. Chips can only have so many transistors per square centimeter before they become unreliable
Dario Amodei — Machines of Loving Grace
Things that are hard constraints in the short run may become more malleable to intelligence in the long run
Dario Amodei — Machines of Loving Grace
how much does being smarter help with this task, and on what timescale
Dario Amodei — Machines of Loving Grace
(“the Singularity”), as superior intelligence builds on itself and solves every possible scientific, engineering, and operational task almost immediately
Dario Amodei — Machines of Loving Grace
train the model can be repurposed to run millions of instances of it
Dario Amodei — Machines of Loving Grace
control existing physical tools, robots, or laboratory equipment through a computer
Dario Amodei — Machines of Loving Grace
“interfaces” available to a human working virtually, including text, audio, video, mouse and keyboard control, and internet access
Dario Amodei — Machines of Loving Grace
powerful AI (I dislike the term AGI)
Dario Amodei — Machines of Loving Grace
named as "Naive" because it assumes the presence of one feature does not affect other feature
Naive Bayes Classifiers - GeeksforGeeks
outputs the input directly if it is positive; otherwise, it outputs zero
ReLU Activation Function in Deep Learning - GeeksforGeeks
One reason for polysemanticity is thought to be the phenomenon of superposition , in which models must represent more concepts than they have neurons, and thus must “smear” their representation of concepts across many neurons
Circuit Tracing: Revealing Computational Graphs in Language Models
features, interpretable building blocks that the model uses in its computations. Second, we describe the processes, or circuits, by which these features interact to produce model outputs
Circuit Tracing: Revealing Computational Graphs in Language Models
neural network layer of dimension N may linearly represent many more than N features
Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet