You can now run Unison Cloud on your own infra. A distributed cloud computer you can program simply and directly, and talks to all your existing stuff. Super easy to set up too

OpenAI – now worth half a trillion dollars – infamously started as a research lab Only 14 companies ever reached a trillion dollar valuation. Two of these – Apple and Microsoft – were also built on the work of a research lab, XEROX PARC This is not a coincidence In 2017, Alan Kay gave a two-part talk, "How to Invent the Future", hosted by Sam Altman In this talk, Alan bemoans how XEROX PARC was "the most successful generator of wealth in computing history, [but] almost nobody wants to know [about it]. Sam [Altman] did." How did Sam Altman replicate the results of PARC? 1. Vision to get the best researchers: PARC = Office of the future OpenAI = AGI 2. Exponential to accelerate with money: PARC = Moore’s Law OpenAI = LLM Scaling Laws 3. Don't expect results for 5 - 10 years: PARC = Built the Alto computer in 5 years, inventing personal computing as we know it – the GUI, the “desktop" metaphor, overlapping windows, WYSIWYG, object-oriented programming – inspiring the Mac and Windows OpenAI = Founded in 2015. In 2017 – the year Alan gave this talk – the Transformer paper (Attention Is All You Need) came out. Five years later, ChatGPT launched. Why there aren't more research labs in the image of PARC given that it's one of the few ways we know how to create this much wealth? We aren't greedy enough. Startups want to make meager billions. You know what's cool? Trillions.

GPT-3, if prompted well, could write incredibly convincing poetry. The new models with all their RLHF and fine-tuning to be chatbots can’t do it to save their lives. It was great, and I kind of miss it.

Everyone who lives east of me got hustle and everyone who lives west of me is lazy

1985: Sinclair BASIC 1986: Commodore BASIC 1988: STOS 1990: AMOS 1992: Turbo Pascal, QBasic 1994: Delphi, C, Pascal 1998: Java 2000: Java, Smalltalk 2006: Scala, Haskell 2009: JavaScript, Ruby 2010: Scala 2016: Haskell 2018: Unison, Haskell

Whenever someone suggests changing waking and work hours instead of changing the clocks twice a year, invariably there’s a chorus of “and show up for work at 10 AM? What are you, lazy?” But like, it’s the exact same instant you’d be showing up if the clock were an hour behind.

🚀 New deep dive: Distributed System Engineering in Unison In our latest blog post, take a look at the design behind Volturno, a distributed stream processing library built atop Unison Cloud. unison-lang.org/blog/volturn… unison-lang.org/blog/volturn…

“[Unison] doesn't pretend you can ignore the challenges introduced by distribution, but it gives you a wealth of tools to tackle them with the full expressiveness you're used to having when programming any other type of software.” - Distributed Sytem Engineering in Unison link:

Fil-C is pretty cool. A garbage-collected fully source-compatible C that turns unsafe memory access into segfaults? Freaking great. It's not compile-time checking, but it is something, and something is often better than nothing.

🔎 As you're writing applications in Unison, you might need to search for a substring (where is this error coming from!?) or numeric literal (who put this magic number in my codebase!?) Here's a quick workflow video for that using the text.find command: piped.video/e3o7tjNnR18?utm_sou…

there are adjoint functors everywhere for those with eyes to see them

There are no "apps" on the Vibe Computer. Instead, there are intentions and capabilities: read mail, analyze data, render text, simulate worlds. It composes the interfaces for these on the fly, just for you. It's all perfectly malleable.

Vibe coding is all the rage, but is anyone thinking about vibe computing? A machine that adapts its interface to the user's intent and context. Where current OSes fix tasks into static apps, the Vibe Computer fluidly generates the right interaction for each moment.

what I wish for all of you is an inexhaustible fruitfulness but you have to want it too

A lot of actions across my whole computer are very low entropy (predictable) just like Cursor predicts next-edits. I wonder which OS (android, iOS, Mac, Linux) will figure out the equivalent tab, tab, tab UI pattern within and across all apps

We're hiring a Growth Engineer to build side-by-side with me at Val Town This is a very personal search. We're a small, ambitious team on a mission to spread the joy of programming & help more people build with code This role spans product, sales, and community. You'll talk to users, design and ship growth experiments, code in TypeScript, write blogs and tutorials, and advocate for our builders For example, on one afternoon you might run some SQL to diagnose a problem in our onboarding flow, design & launch a small fix, and see hundreds of new users succeed with it the next day We're looking for someone who will thrive in a startup environment. A true owner. Maybe a former or aspiring founder, excited to take responsibility and make things happen. Someone who learns and grows quickly. We work hard because we care about our mission. We're not 996, but we're also not clocking in and clocking out We're in-person here in Brooklyn. You will get creative freedom, fast feedback, and me as your daily collaborator If this feels like your next adventure, please reach out. Amazing people make amazing companies, and we have so much ahead of us

Are you using Kinesis? Would you be interested in a drop-in replacement that drops your bandwidth costs to zero (which is most of the cost for a typical Kinesis setup), with the trade-off that it has 10x higher write latency? DM me if interested.

🤯 Unison in the browser!? Dan Freeman's new library, Proscenium, runs Unison code in WebAssembly. Its companion library, @dfreeman/playbill, contains a helpful walk-through of various features and larger examples live on Unison Cloud! 🤩 Give it a try! share.unison-lang.org/@dfree…

Here are the deets: unison.cloud/kinesis-on-s3/?…

An 85% cheaper Kinesis-compatible replacement that just uses S3, has much larger batch size limits, allows huge records, and can remember partial results of map-reduce jobs to avoid redoing work. Link in reply.