reading
useful ideas from things ben guo has saved and read.
showing all 163 reading entries
- 44% on arc-agi-1 in 67 cents
- timesfm-3: a zero-shot foundation model for multivariate forecasting
- techno singularity and “satanic” eschatology
- “a message and a warning to humanity as seen from the universe” by contactee sixto paz wells, 1990s
- ai will change everything, just not all at once | tyler cowen | strange loop | ep 6
- who captures the value created by ai? - dylan patel
- a quick rundown on accelerationism
- could the ai boom trigger a global debt crisis? - dylan patel
- how ai could reprice the entire economy - dylan patel
- how ai could concentrate the world’s labor in a few companies - dylan patel
- what if the world had one gpu per person?
- monid: connect your agent to every tool it needs
- skilldeck: install agent skills from git-backed catalogs
- a sneak preview behind an embedded software factory. i suspect rapid application dev is back
- coding agents are becoming runtimes
- dynamic apps: deploy ai-generated apps for your users
- agentos security model
- ori eval: find the best model for your project
- malleable software = solid bases + custom code
- another ai app builder. why?
- agent memory as a file format
- everything hackable will get hacked
- the hugging face attack surprised me
- what my dad taught me about ai coding in the 90s
- the abstraction fallacy: why ai can simulate but not instantiate consciousness
- the rise and fall of agent civilizations
- how to understand the next wave of ai before everyone else | tibo interview
- vms won’t contain cyber-capable agents
- initially delayed, the pass-through of tariffs to consumer prices has arrived
- terminal-bench-science 0.1
- “weird” is a weird word
- i accidentally turned llm memory into program analysis
- just a rumour of a bug is enough to find a security exploit these days
- software is now user-generated content
- swarmworld: stigmergic technological evolution in societies of language-model agents
- durable objects are made for agents
- the oracle and the firm
- software is made between commits
- agencity: a durable runtime for autonomous agents
- code-orchestrator: delegated phase-based implementation
- build-person: an evidence-calibrated whole-person model
- semantic-algos: why combinators
- how i design with ai.
- startup fellowship 002: balto energy | devcon 6
- the company as a codebase
- we’ve built 1% of the ai compute the world needs
- agent swarms are a distributed systems problem
- gemini omni 1.1 flash lets you build with more control
- intelligent transcription with gemini 3.5 transcribe
- programming as theory building
- dylan patel – anthropic & openai will have most of the world’s compute by 2028
- being legible to capital
- dear vcs, just give up
- harness engineering: an anthology and field guide
- let the tokens flow. sail powers efficient, reliable deep research
- no revenue is enough revenue
- prime agent: a self-improving rlm harness
- say something people want
- small models have arrived
- the choices we make about ai now are critical
- actually queryable executables
- openai jalapeño: better than nvidia blackwell
- my friend aaron
- the mismanaged geniuses hypothesis
- walgit: one binary in front of an object store
- agent lightning v1.0.1
- ai coding will prevent expertise
- anger, anxiety and agency
- everything i own, owned
- fences, not sandboxes
- headlong: a microharness for persistent agents
- llms could control their host machines by exploiting inference engines
- rubber-stamp travel field notes
- this rural village in spain is welcoming digital nomads with open arms
- what is a harness?
- why are some octopuses so smart? the answer might lie in a never-before-seen mutation that helps them accurately build proteins
- your executable is a sqlite database
- training ai to paint with code
- cursor, openrouter, and where the money went
- stop making tuis
- there's no reason for software to be slow anymore
- are open models catching up?
- bb: the agent ide that builds itself
- chroma context-1: training a self-editing search agent
- designing autonomous agents
- training a 125m-parameter model to autocomplete piano
- the hofmann wobble
- alibaba is a factory directory
- building software factories (with no slop)
- everything i know, section 12
- everything i know, section 11
- everything i know, section 10
- everything i know, section 9
- everything i know, section 8
- everything i know, section 7
- everything i know, section 6
- everything i know, section 5
- everything i know, section 4
- everything i know, section 3
- everything i know, section 2
- everything i know, section 1
- extensible software in the age of llms
- hax, a minimalist terminal-native coding agent
- ian xiaohei illustrations
- if a hdd was an airplane...
- personal agents notes (instinct / grok bots / chatgpt work)
- prime agent: a self-improving rlm agent
- stripe's letter on acquiring openrouter
- the best general advice on earth
- very much in line with…
- desktopfly
- does creatine make you smarter?
- git at any scale
- tokens are the new dollars
- turbovec
- finger: the 1971 social network that never died
- the benchmarkpocalypse
- too dangerous to concentrate or distribute?
- ai isn’t outthinking mathematicians. it’s out-remembering them.
- cultivating a state of mind where new ideas are born
- guiding ships with moire patterns
- patterns and problems in multiagent systems
- so how is ai drug discovery doing, really?
- chiptycoon: how a computer chip is made
- deepseek harness
- draw faces with javascript
- eigendrum
- everything is about to “go dark”
- how to influence google ai overviews
- kelly criterion simulator
- painting with gaussians
- tests must justify their presence
- words have power
- x for you feed algorithm
- agentic web index: spoofing & security
- felix and i
- reality has a surprising amount of detail
- a tome of forbidden technologies
- discovery loop — continuous exploration
- feynobg: a sota model for background removal
- google search is dying. what comes next is worse
- harmony explained: progress towards a scientific theory of music
- let’s all meet up in the y2k
- nvidia’s risky business
- petri nets as a music sequencer
- something is changing in the unit economics of software
- tech leaders say ai means less work - their staff say they work up to 90 hours a week
- the gargantuan lie that is collapsing the world’s climate
- why is everyone in tech so sad?
- joe rogan experience #2530 - timothy alberino
- mirrorcode: what's the largest software project ai can complete on its own?
- developers are attached to tools because tools encode trust
- llms reward expertise
- heaven in a silicon chip
- it scrolls up
- martin shkreli breaks down the collapse of situational awareness
- obliterate, don’t automate
- sarah's wager
- the artist who colored ghibli
- the building block economy
- the rise of million-dollar companies with just one employee
- the shape of things to come, part 1: the continuous thunderdome
- trajectory: a standard format for agent experience data