Emma Guo
113 followers · 74 following · 5990 views
on the atlas · 245
- Building a C compiler with parallel Claudes \ Anthropic1 savers
- Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights18 savers
- The Sane Asylum - Colossus21 savers
- turbopuffer v31 savers
- Six Gutenberg Revolutions in a Row—Asterisk3 savers
- Making Eg-walker even faster4 savers
- Safe AI Forum2 savers
- Claude reveals ECDSA curve origins \ Anthropic4 savers
- Slim Lim9 savers
- The AI-native interview | Sierra4 savers
- The importance of a (or many) Switzerland(s) in AI · ai minutes1 savers
- Hyper-𝜏-bench: Evaluating agents that build agents | Sierra1 savers
- Science as Collective Sensemaking - by Charles Yang2 savers
- How to be more ambitious: A tutorial for humans with models37 savers
- Research Lessons I Learned at MATS & Redwood — LessWrong13 savers
- Redesigning source control at Anthropic \ Anthropic10 savers
- Better Living Through Algorithms11 savers
- AI Writing Is Fine — Matthew Phillips1 savers
- RIP, vector database2 savers
- Find The Fast Moving Water2 savers
- Fishing Log — Andrew1 savers
- Anthropic announces its molecular research group and laboratory \ Anthropic4 savers
- How Notion handles concurrent editing with CRDTs7 savers
- the senior engineer death spiral • Solving the decision problem7 savers
- Optimism Shapes Reality - by Alexandr Wang9 savers
- A Farewell to Jef Raskin | Toolness1 savers
- The Two Cultures of Mathematics | The n-Category Café1 savers
- Home - TypeSafe AI5 savers
- mimo-v2.6 RL7 savers
- library — matthew lee2 savers
- Introducing System One Models and Jev - TypeSafe AI Blog21 savers
- beijing travel notes · matthew lee1 savers
- concept-essay.pdf1 savers
- The lifecycle of a sharded Postgres query — PlanetScale1 savers
- How We Built Safety Into Muse | Meta AI Research1 savers
- Project Cambria: Translate your data with lenses6 savers
- [2412.06269] Schema Evolution in Interactive Programming Systems1 savers
- On Self-Respect – Alex Burgin16 savers
- WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution1 savers
- on doing Real work - by jessica dai7 savers
- eve – The Agent Framework - Vercel1 savers
- Small Models Have Arrived8 savers
- Speculative Programmatic Tool Calling | Alex L. Zhang2 savers
- Code was our medium for thought8 savers
- Agent Plugins1 savers
- Zawinski's Law - Laws of Software1 savers
- Git at any scale · Cursor18 savers
- The Structure of Scientific Revolutions5 savers
- A Programmer's Introduction to Mathematics3 savers
- What's Easy Now? What's Hard Now? - Marc's Blog2 savers
- Darklang1 savers
- RLHF Book by Nathan Lambert15 savers
- POND1 savers
- Blog Posts | Tom Cunningham – Tom Cunningham2 savers
- Everything I know about good API design7 savers
- Nicotine · Gwern.net6 savers
- Harbor3 savers
- No Data Centers In My Backyard - by Jasmine Sun14 savers
- DeltaDB — Early Access1 savers
- Discovering cryptographic weaknesses with Claude \ Anthropic6 savers
- The Training Imperative5 savers
- Superlogical5 savers
- If you let AI do your writing, I will come to your house and kill you29 savers
- tuicr — terminal UI for code review1 savers
- Wordgard Release 0.11 savers
- Mount Tamalpais: Pantoll to East Peak · Hiking by Transit1 savers
- The Mother of All Demos6 savers
- Sealed boxes | Libsodium documentation1 savers
- Stables and Volatiles – Rands in Repose4 savers
- Herdr — one terminal for the whole herd3 savers
- Roughdraft1 savers
- Ben Kamens / Claw, reflect on our work and express your feelings as animated ASCII art2 savers
- Wink1 savers
- Designing agentic loops3 savers
- Introducing Mesa: a versioned filesystem for agents | Mesa1 savers
- List of posts - <antirez>2 savers
- Claude Code Opus 4.7 Performance Tracker | Marginlab1 savers
- Mesa1 savers
- Table of Contents (Stanford Encyclopedia of Philosophy)2 savers
- Keyhive: Local-first access control3 savers
- Organic web1 savers
- Large Language Diffusion Models1 savers
- Screen Slate | Cinematic Excellence Est. 2011 | Upcoming Film Showtimes NYC3 savers
- Letter from Utopia9 savers
- Consensus Compare: Casper vs. Tendermint | by Interchain | Cosmos Blog2 savers
- Why I think AI take-off is relatively slow - Marginal REVOLUTION14 savers
- Understanding Variational Autoencoders (VAEs) | by Joseph Rocca | Towards Data Science8 savers
- How Mozilla’s Rust dramatically improved our server-side performance | Figma Blog3 savers
- Honey I SNARKED the GPT - EZKL Blog2 savers
- eecs127_reader.pdf2 savers
- Looking for Alice - by Henrik Karlsson - Escaping Flatland94 savers
- Staring into the abyss as a core life skill88 savers
- the friendship theory of everything69 savers
- Thirty Observations at Thirty62 savers
- i wish we’d grown up on the same advice - by vincent huang61 savers
- Why I do what I do | Jocelyne Murphy58 savers
- Conviction is the Scarcest Resource: Essays About Gaining Real Conviction45 savers
- crushes are often just misplaced ambition - by Isabel42 savers
- Understanding41 savers
- Relationships are coevolutionary loops - by Henrik Karlsson40 savers
highlights · 705
Distribution is what (always) matters.
The Training ImperativeNo language models were used in the creation of this software. I design and build my systems by hand.
Wordgard Release 0.1I wish a pox upon them and their slop machines.
Wordgard Release 0.1So in a document of length 10, inserting an L at 4 is represented [keep 4] [replace 0 with "L"] [keep 6]
Wordgard Release 0.1it sparked interest in a small vigorous research community but it didn't have impact on the computer field as a whole
The Mother of All DemosData either lives forever or only lives for the duration of the event loop
How Mozilla’s Rust dramatically improved our server-side performance | Figma Blogtemporary solution was to isolate the crazy documents to a completely separate pool of “heavy” workers
How Mozilla’s Rust dramatically improved our server-side performance | Figma Bloga single slow operation would lock up the entire worker until it was complete
How Mozilla’s Rust dramatically improved our server-side performance | Figma BlogEven if an agent is compromised, there's nothing to exfiltrate
Safely inject credentials in HTTP headers with Vercel Sandbox - Verceloutside the sandbox VM boundary, so apps running inside the sandbox can call authenticated services without ever accessing the credentials
Safely inject credentials in HTTP headers with Vercel Sandbox - VercelEvery emerging AI use case has a big elicitation gap
I Solved My Mystery Fatigue with AI - by Amy Dengour effective technology for manipulating the physical world seems to have drastically deteriorated
Questions · Patrick CollisonWhy can't I connect my editor to a running program and hover over values to see what they last were
Questions · Patrick Collisontypes will be redundantly created on different threads
Why does tsgo use so much memory?When a Checker is created, it is given the entire Typescript program AST and all its files:
Why does tsgo use so much memory?The typechecker takes up only ~50MB instead of ~400MB! This strongly suggests to me that there is some overhead with multi-threading here.
Why does tsgo use so much memory?I gain a better approximation of the north star I am drawn to
there is no *true north*life is a random walk
there is no *true north*There is nothing inherent to myself which motivated what I have worked on
there is no *true north*many successful Stables used to be Volatiles who are recovering from the last war
Stables and Volatiles – Rands in Reposeand I find myself rooting for traffic
How to Pick Your Life Partner - Part 2 — Wait But Whyalso adding tighter controls on system prompt changes. We will run a broad suite of per-model evals for every system prompt change to Claude Code
An update on recent Claude Code quality reports \ AnthropicI consume for inspiration as I edit each file myself
Ben Kamens / Code with AI the Hard Waywe use the chat history as files to improve the quality of summarization
Dynamic context discovery · Cursorwrite the output to a file and give the agent the ability to read it
Dynamic context discovery · Cursorcurrently has ~20 tools
Lessons from Building Claude Code: Seeing like an Agentwe send lightweight stubs — just the tool name, with defer_loading: true — that the model can "discover" via a ToolSearch tool when needed
Lessons from Building Claude Code: Prompt Caching Is Everythingso we run alerts on our prompt cache hit rate and declare SEVs if they're too low
Lessons from Building Claude Code: Prompt Caching Is Everythingre-use our intuition for physical objects, movement, and space to see and express ideas in more abstract domains.
What makes a good human interface? | thesephist.comSometimes, these hidden metaphors are worth reifying into concrete visuals
What makes a good human interface? | thesephist.comPrimitives like charts, canvases, (reverse-)chronological timelines, calendars, are all based on taking some meaningful dimension of information, like time or importance, and mapping it onto some space.
What makes a good human interface? | thesephist.comYou need to keep a certain rate of improvement for things not to break. If you are too slow in adapting to each other, you grow apart.
Relationships are coevolutionary loops - by Henrik KarlssonYour friends will change you, even in ways you initially reject.
the friendship theory of everythingOur logCount function gets re-created because the App component re-renders when the count state is updated
Hooks, Dependencies and Stale Closures | TkDodo's blogclosures are created every time a function is created, at function creation time.
Hooks, Dependencies and Stale Closures | TkDodo's bloga closure gives you access to an outer function's scope from an inner function
Hooks, Dependencies and Stale Closures | TkDodo's blogMost AI apps should be agent builders, not agents.
Pete Koomen“Horseless carriage” refers to the early motor car designs that borrowed heavily from the horse-drawn carriages that preceded them.
Pete Koomenthis standard is currently used by Amazon for inter-datacenter communication – but for requests, not responses, and is a way to verify that the request you’re receiving is authentic and trusted, and not just anyone can query arbitrary data from AWS datacenters.
Advancing Standards for ZK and Provenanceevery device in Backchannel contains an Automerge document called the “system document,” which includes application settings and the contact list
Backchannel: A relationship-based digital identity systemPAKE generates a shared symmetric key that represents the relationship. This key is a globally unique and secure identifier that can be used to create a secure end-to-end encrypted communication channel
Backchannel: A relationship-based digital identity systema single symmetric key as a digital security token, proving that an out-of-band exchange was performed at a particular moment in time
Backchannel: A relationship-based digital identity systemPhonebooks are great tools for recognition because they are local, private, and memorable.
Backchannel: A relationship-based digital identity systempaperless workplace in the late 90s. From my experience, it took about 20 years for most workplaces to actually go (mostly) paperless
Why I think AI take-off is relatively slow - Marginal REVOLUTIONElectricity for instance arguably took up to forty years
Why I think AI take-off is relatively slow - Marginal REVOLUTIONsome Star Trek technology fell into our backyard, and how long will it take us to figure out how to integrate it with other things humans do
Why I think AI take-off is relatively slow - Marginal REVOLUTIONpolitical obligation ends when protection ceases
Hobbes’s Moral and Political Philosophy (Stanford Encyclopedia of Philosophy)These “precepts”, “conclusions” or “theorems” of reason are “eternal and immutable”, always commanding our assent even when they may not safely be acted upon
Hobbes’s Moral and Political Philosophy (Stanford Encyclopedia of Philosophy)what, exactly, it is about human beings that makes it the case (supposing Hobbes is right) that our communal life is prone to disaster when we are left to interact according only to our own individual judgments
Hobbes’s Moral and Political Philosophy (Stanford Encyclopedia of Philosophy)The bonds of affection, sexual affinity, and friendship—as well as of clan membership and shared religious belief—may further decrease the accuracy of any purely individualistic model of the state of nature
Hobbes’s Moral and Political Philosophy (Stanford Encyclopedia of Philosophy)