Dan Abramov
Co-creator of Redux and long-time core contributor to the React team; writes about JavaScript and software design at overreacted.io.
Dan Abramov did not write this page. ppll compiled it from their own words, each linked to the source. Is this you? · Report a mistake
Latest Bluesky
What they believe
-
Novel mathematical claims produced by an AI must be formally verified, such as in a proof assistant, before they can be trusted as legitimate results.
I needed to back it up with Lean before laying more bricks on top of existing bricks, and I definitely needed to verify things with Lean before making any novel mathematical claims.
overreacted.ioSaid 18 Sept 2026This on korrents
-
Separating where content is hosted from the apps that display it fixes the incentive problems in both closed and federated social networks.
Separating hosting and apps fixes broken incentives in closed and in federated social. Coupling hosting and apps was the original sin, and the fix is simple.
overreacted.ioSaid 19 Jun 2026This on korrents
The bold line is korrents’ wording; the quote is theirs. These 2: what they wrote before what they said aloud, one per source, newest first.
Feed
As its own page →Hiding
20 November
19 November
17 November
1 November
21 October
19 October
4 October
18 September
From one piece How I Vibed a Proof of Conway’s Conjecture 2 beliefs · overreacted.io
-
korrents.com
A language model cannot be trusted to judge which of its own results are actually significant or interesting.Their words
I could not trust the model to evaluate which results were interesting.
-
Their words
I needed to back it up with Lean before laying more bricks on top of existing bricks, and I definitely needed to verify things with Lean before making any novel mathematical claims.
-
Likedrcmnd.app
CodexTheir words
Codex has a really nice "Goals" feature that periodically reminds the sessions what they're supposed to be doing, which makes it easier to prevent drift.
-
Likedrcmnd.app
ChatGPTTheir words
While Claude made important contributions, on average ChatGPT seemed better at new mathematical thinking, and definitely better at coordination and adhering to goals.
-
Mixed onrcmnd.app
ClaudeTheir words
Claude was good at writing Lean when there was a clear unambiguous goal. While Claude made important contributions, on average ChatGPT seemed better at new mathematical thinking, and definitely better at coordination and adhering to goals.
+ 1 more
-
Usesrcmnd.app
LeanTheir words
It took me an entire month of my free time and a boatload of tokens, but I believe I've obtained a Lean proof of this conjecture posed by John Conway 50 years ago
3 September
9 July
19 June
-
Their words
Separating hosting and apps fixes broken incentives in closed and in federated social. Coupling hosting and apps was the original sin, and the fix is simple.
-
Recommendsrcmnd.app
TangledTheir words
If you want to try new apps or make new apps, you can do that too! Check out Tangled and Semble, which have nothing to do with Bluesky.
-
Recommendsrcmnd.app
SembleTheir words
If you want to try new apps or make new apps, you can do that too! Check out Tangled and Semble, which have nothing to do with Bluesky.
-
Usesrcmnd.app
EuroskyTheir words
If you want to swap your hosting, you can. I literally did this today. Aside from three or four UX snags, it was all automatic. My atproto stuff is at Eurosky now.
16 June
24 March
18 January
From one piece A Social Filesystem 2 beliefs · overreacted.io
-
Their words
It's actually good that different products can disagree about what a post is! Different products, different vibes. We'd want to support that, not to fight it.
-
korrents.com
A person's creations should live somewhere they control, not be owned by the apps used to make them.Their words
Since files represent your creations, they should live somewhere that you control.
-
Recommendsrcmnd.app
pdslsTheir words
Open pdsls. If you want to explore the Atmosphere (at://-mosphere, get it?), pdsls is the best starting point.
13 January
5 January
19 December 2025
30 November 2025
11 November 2025
21 October 2025
From one piece How to Fix Any Bug 2 beliefs · overreacted.io
-
Their words
This is why, whenever you narrow a repro, you should also confirm that a positive result ("everything works") is still possible to obtain with the new repro.
-
Their words
I don't know any other method that's so effective after you've run out of theories.
2 October 2025
From one piece Where It's at:// 2 beliefs · overreacted.io
-
Their words
However, in theory, whoever operates the PLC registry could deny my requests to update the DID Document, or refuse to serve some information about it.
-
Their words
But by standardizing how these pieces fit together-putting the user in the authority position, separating identity from hosting, and making data portable-it turns the web into a place where your content belongs to you, not to the apps that display it.
26 September 2025
From one piece Open Social 3 beliefs · overreacted.io
-
korrents.com
Aggregation features such as notifications, feeds, and search are essential to any modern social product.Their words
Social aggregation features like notifications, feeds, and search are non-negotiable in modern social products.
-
Their words
If you can't leave without losing something important, the platform has no incentives to respect you as a user.
-
Their words
Despite the grip of today's social media companies, I believe open social will eventually seem inevitable in retrospect-just like open source does now.
2 September 2025
16 August 2025
30 July 2025
11 June 2025
5 June 2025
1 June 2025
Nothing matches.
Beliefs
Korrents What they believe 12 beliefs — each backed by an exact quote.
Each is a — compiled by korrents.com, not by them: the one-line wordings are korrents', the quotes are theirs.
Recent
Novel mathematical claims produced by an AI must be formally verified, such as in a proof assistant, before they can be trusted as legitimate results.
I needed to back it up with Lean before laying more bricks on top of existing bricks, and I definitely needed to verify things with Lean before making any novel mathematical claims.
How I Vibed a Proof of Conway’s Conjecture Said 18 Sept 2026
A language model cannot be trusted to judge which of its own results are actually significant or interesting.
I could not trust the model to evaluate which results were interesting.
How I Vibed a Proof of Conway’s Conjecture Said 18 Sept 2026
Separating where content is hosted from the apps that display it fixes the incentive problems in both closed and federated social networks.
Separating hosting and apps fixes broken incentives in closed and in federated social. Coupling hosting and apps was the original sin, and the fix is simple.
There Are No Instances in atproto Said 19 Jun 2026
Show 9 more
A person's creations should live somewhere they control, not be owned by the apps used to make them.
Since files represent your creations, they should live somewhere that you control.
A Social Filesystem Said 18 Jan 2026
It is good for different products to disagree about what counts as the same kind of content, rather than converging on one shared definition.
It's actually good that different products can disagree about what a post is! Different products, different vibes. We'd want to support that, not to fight it.
A Social Filesystem Said 18 Jan 2026
Removing parts of the code one at a time while confirming the bug still reproduces is the most effective way to find a root cause once other theories are exhausted.
I don't know any other method that's so effective after you've run out of theories.
How to Fix Any Bug Said 21 Oct 2025
When narrowing a bug reproduction case, it should still be possible to obtain a passing result with the new case, not just a failing one.
This is why, whenever you narrow a repro, you should also confirm that a positive result ("everything works") is still possible to obtain with the new repro.
How to Fix Any Bug Said 21 Oct 2025
Separating a user's identity from where their data is hosted, and making data portable, gives people ownership of their own content rather than the apps that display it.
But by standardizing how these pieces fit together-putting the user in the authority position, separating identity from hosting, and making data portable-it turns the web into a place where your content belongs to you, not to the apps that display it.
Where It's at:// Said 2 Oct 2025
Whoever operates a did:plc-style identity registry retains some power to deny identity update requests or withhold information about an identity.
However, in theory, whoever operates the PLC registry could deny my requests to update the DID Document, or refuse to serve some information about it.
Where It's at:// Said 2 Oct 2025
Open, user-owned social networks will eventually become as standard and inevitable as open source software already is.
Despite the grip of today's social media companies, I believe open social will eventually seem inevitable in retrospect-just like open source does now.
Open Social Said 26 Sept 2025
A platform has little incentive to treat its users well when leaving it means losing something they can't take with them.
If you can't leave without losing something important, the platform has no incentives to respect you as a user.
Open Social Said 26 Sept 2025
Aggregation features such as notifications, feeds, and search are essential to any modern social product.
Social aggregation features like notifications, feeds, and search are non-negotiable in modern social products.
Open Social Said 26 Sept 2025
What is a korrent?
A korrent is a belief a person has stated in their own words: one sentence stating the claim, backed by a quote and a source, kept at korrents.com.
Under a name here, the quoted block is what they actually said. The korrent beneath it is the claim those words support, in korrents' wording — tap it to see the record, its source, and who else holds it.
Nobody here wrote their own korrents. They are compiled from public statements, and a person can change their mind, which is recorded too.
Recommends rcmnd
rcmnd What they recommend 8 things — each backed by an exact quote.
Compiled by rcmnd.app, not by them: the quotes are theirs, and each links to the post it was copied from.
Recent
Uses Lean
It took me an entire month of my free time and a boatload of tokens, but I believe I've obtained a Lean proof of this conjecture posed by John Conway 50 years ago
overreacted.io 18 Sept 2026
Mixed on Claude
Claude was good at writing Lean when there was a clear unambiguous goal. While Claude made important contributions, on average ChatGPT seemed better at new mathematical thinking, and definitely better at coordination and adhering to goals.
overreacted.io 18 Sept 2026
Liked ChatGPT
While Claude made important contributions, on average ChatGPT seemed better at new mathematical thinking, and definitely better at coordination and adhering to goals.
overreacted.io 18 Sept 2026
Show 5 more
Liked Codex
Codex has a really nice "Goals" feature that periodically reminds the sessions what they're supposed to be doing, which makes it easier to prevent drift.
overreacted.io 18 Sept 2026
Uses Eurosky
If you want to swap your hosting, you can. I literally did this today. Aside from three or four UX snags, it was all automatic. My atproto stuff is at Eurosky now.
overreacted.io 19 Jun 2026
Recommends Semble
If you want to try new apps or make new apps, you can do that too! Check out Tangled and Semble, which have nothing to do with Bluesky.
overreacted.io 19 Jun 2026
Recommends Tangled
If you want to try new apps or make new apps, you can do that too! Check out Tangled and Semble, which have nothing to do with Bluesky.
overreacted.io 19 Jun 2026
Recommends pdsls
Open pdsls. If you want to explore the Atmosphere (at://-mosphere, get it?), pdsls is the best starting point.
overreacted.io 18 Jan 2026
Where they publish
Site danabra.mov Link list: github, bluesky, overreacted, talks.
Link verified 25 Sept 2026.
Blog overreacted Essays on JavaScript, React and software design.
Recent
- How I Vibed a Proof of Conway’s Conjecture 18 Sept 2026 You can just prove things, apparently.
- There Are No Instances in atproto 19 Jun 2026 Like RSS and Google Reader.
- A Social Filesystem 18 Jan 2026 Formats over apps.
Show 17 more
- Introducing RSC Explorer 19 Dec 2025 My new hobby project.
- Hire Me in Japan 11 Nov 2025 I'm looking for a new job.
- How to Fix Any Bug 21 Oct 2025 The joys of vibecoding.
- Where It's at:// 2 Oct 2025 From handles to hosting.
- Open Social 26 Sept 2025 The protocol is the API.
- A Lean Syntax Primer 2 Sept 2025 Programming with proofs.
- Beyond Booleans 16 Aug 2025 What is the type of 2 + 2 = 4?
- The Math Is Haunted 30 Jul 2025 A taste of Lean.
- Suppressions of Suppressions 11 Jun 2025 I heard you like linting.
- I'm Doing a Little Consulting 11 Jun 2025 Personal update post.
- How Imports Work in RSC 5 Jun 2025 A layered module system.
- RSC for LISP Developers 1 Jun 2025 Quoting for modules.
- Progressive JSON 31 May 2025 Why streaming isn't enough.
- Why Does RSC Integrate with a Bundler? 30 May 2025 One does not simply serialize a module.
- One Roundtrip Per Navigation 29 May 2025 What do HTML, GraphQL, and RSC have in common?
- Static as a Server 8 May 2025 You wouldn't download a site.
- RSC for Astro Developers 6 May 2025 Islands, but make it fractal.
Link verified 25 Sept 2026. Recent items update automatically from the channel.
Bluesky danabra.mov Short-form notes on JavaScript and React.
Recent
- lean zulip discussion 20 Nov 2026 leanprover.zulipchat.com
- i was literally in a bath 20 Nov 2026
- my favorite take on all of this, from @xenaproject.bsky.social a few weeks ago. 20 Nov 2026 xenaproject.wordpress.com
Show 14 more
- oh well. end of an era. i thought this was a good video about it, certainly helped me understand a bit 20 Nov 2026
- i think with coding it can be hard to tell but with formalization the gain in capability is staggering. i mean higher level skills (how to organize proof) and not just writing Lean itself. in my experience Fable 5.2 blows Astra 6 and Fable 5.1 out of the water. subjectively it feels like 5x better 20 Nov 2026
- if you take AI math to absolute it seems like there should just be a social content-addressable space of proposition nodes (true, false or unknown), proofs as directed many-into-one pathways (these nodes together imply this), and many points of view (theories) that select interesting nodes and roads 19 Nov 2026
- oh uhhhhhh good to know 19 Nov 2026
- @samuel.fm seeing a jumpy frame when switching to Notifications where list is under pill section 17 Nov 2026
- python3: what is my purpose old="""...""" new="""...""" oh my god 17 Nov 2026
- fable 5.2 > astra 6 > fable 5.1 competition is good! 17 Nov 2026
- im pretty sure Amelie and Matrix happen in the same cinematic universe 17 Nov 2026
- @samuel.fm plzzz 1 Nov 2026
- this was very good and articulated a lot i've been feeling lately. in formalization, this effect is even more noticeable: decoherence is punished by the agent being stuck in the maze for days when otherwise it would've made progress in 30 minutes but i think this applies broadly to software 21 Oct 2026
- rip to the real one 19 Oct 2026
- i’ve been doing some research on using Lean with AI to prove software properties (in my case, pieces of React). i’ll try to share lessons if it gets to a useful state. the main thing i’m learning is that, with this setup, proofs are *grown*. you can’t “just build them” 4 Oct 2026
- fun read 4 Oct 2026
- been unironically upping my p doom over the last few months 4 Oct 2026
Link verified 25 Sept 2026. Recent items update automatically from the channel.
GitHub gaearon Redux, React core work, and smaller tools.
Recent
- overreacted.io — Personal blog by Dan Abramov. 18 Sept 2026
- conway-refinement — A proof of Conway's refinement conjecture in Lean 3 Sept 2026
- next-cc-activity-problem 9 Jul 2026
Show 5 more
- turbopack-seal-bug 16 Jun 2026
- react-udv-bug 24 Mar 2026
- woodshed — Create, run, rate, and iterate on your Claude Skills 13 Jan 2026
- rscexplorer — A tool for people curious about the React Server Components protocol 5 Jan 2026
- next-bug-repro 30 Nov 2025
Link verified 25 Sept 2026. Recent items update automatically from the channel.
About the English under a post
Some people here publish in a language other than English. Where they do, this site shows a machine translation beneath the post, in this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the serif above is exactly what the person published, and it is what to quote them on. A translation can be wrong in ways that matter, especially about tone.
Only the post's own words are translated. A quoted post, a linked article and a belief on korrents.com are left in their original language.