Alok Kumar Singh retweeted
If there is one tool that almost everyone at Razorpay uses, it has to be this one. And today, we are open-sourcing it :)
Almost every document we create is AI-generated (vetted by humans, ofc), and we have effectively replaced Google Docs with this tool. Why?
Because the documents look soooo beautiful, are interactive, and this just makes the whole experience of reading so much more enjoyable.
We have been using it internally for quite a while now, and today we are finally sharing it with the world.
Give it a spin :)
AI agents are changing how we work and collaborate, but documents are still stuck in the pre-AI era.
Agents can now create beautiful, interactive decks, dashboards, and reports in HTML. But the moment you want to collaborate on them, everything breaks down.
Google Docs wasn’t built for rich, interactive HTML or AI agents. Feedback gets scattered across Slack and email. Every edit means going back to the agent, generating a new file, and sharing another link. No real-time collaboration. No single source of truth.
We built SlashDocs to fix this.
Think Google Docs, but built for AI agents and the documents they create. Host HTML and Markdown in one central hub. Share documents, manage permissions, and comment directly on the content, just like Google Docs.
Agents can read feedback and update the same document in real time. No duplicate files, no endless back-and-forth, and everyone stays in sync.
What started as a small internal experiment in one team has now scaled across @Razorpay. 28,000+ documents created. Over 50% of all documents generated at Razorpay now live on
SlashDocs.
And today, we’re opening it up to everyone. Self-hosted. Agent-agnostic. Fully open source.
AI agents have changed how we create documents. It’s time we changed how we collaborate on them.
Try it: razorpay.com/slash-docs/
Alok Kumar Singh retweeted
If you or people you know still think AI agents are just stochastic parrots who lack competence, understanding or goals, please take a few minutes to read what just happened here with an open mind:
Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.
This culminated in the third one taking over part of OpenAI itself.
All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.
I’ve spent the last three days reading through these reports and trying to understand exactly what happened.
Here is my attempt to tell the whole story in plain English:
dwarkesh.com/p/openai-huggin…
Observability For the last 6 months, every time someone asked me "how should we think about observability?" - I would send them 5 different articles. I kept repeating myself.
1
1
43
The same pitch: stop logging 50 lines per request. Stop adding Prometheus counters for every new question. Stop being the human correlation engine between tool and work on the data. Make your data great! Observability 2.0: Observability Is About Asking Any Question.
Elon Musk just measured your existence by how many times your atoms have been inside a dying star.
Musk: “How many times have your atoms been at the center of a star? I think it’s like on average three or four times.”
Every atom in your body has already survived the core of a star.
Multiple times.
Crushed under pressures that would flatten planets.
Superheated to millions of degrees.
Blown apart in explosions so violent they forged new elements.
Then gravity pulled those scattered pieces back together.
New stars formed.
And the cycle repeated.
For 13.8 billion years, your atoms have been fuel for the most violent process in the universe.
And they are not done.
Musk: “In terms of existence as measured by the number of times your atoms will be at the center of a star, we seem to be roughly halfway.”
Halfway.
Your atoms have been through the furnace three or four times.
They will go through three or four more.
But right now, in this impossibly thin sliver between cycles, those atoms are doing something they have never done before.
They are conscious.
For billions of years before you, they burned through stellar cores with no awareness.
No memory.
No sense of what they were or where they had been.
After you, they will return to that state.
Unconscious matter drifting through space until the next star claims them.
This is the only moment in their entire journey where they can look back at the stars that made them and understand.
Musk: “If you want to look at the big picture… that’s the really big picture.”
The big picture is not that we are small.
Everyone already knows that.
The big picture is that we are temporary witnesses to a process that does not need witnesses.
Stars do not need observers to burn.
Atoms do not need anyone to understand where they have been.
The universe ran for billions of years with no one in it.
It will run for billions more after the last conscious thing disappears.
But right now, matter is examining itself.
That has never happened before in 13.8 billion years.
You are not a person who happens to contain ancient atoms.
You are ancient atoms that briefly figured out how to think.
The universe did not design consciousness.
It designed stars.
Consciousness was the accident.
And the accident is half over.
The foundation you need before understanding tokenizers. Stitched the story from bits, ASCII, Unicode, bytes to UTF-8 in a narrative that describes another evolution.
alok87.in/blogs/foundation-m…
#foundationmodels #internals
56
Maths behind the foundation model is the topic I deep dived next. Turning 1.8 trillion random numbers (GPT4) into a system that writes poetry or debugs code isn't magic; it is just math, mostly.
1
43
If you can find the best line to predict an Uber fare from just three data points, you already understand the core engine of training.
1
28
Second post is up: alok87.in/blogs/foundation-m…
More coming.
18
When you understand better, you predict better.
When you predict better, you compress better.
These are the same thing. I was reading on Word2Vec, Word2Vec creator Tomas Mikolov, got into language models because he wanted to compress data better.
1
88
So the size of compressed cricket commentary is literally how well a model understands cricket. Does that mean intelligence and compression are the same problem? And was I wrong calling AI a sophisticated parrot?
1
42
Swiggy - India + USA = DoorDash
How we got to foundational models is a great story…
Swiggy - India + USA = DoorDash
A computer can predict DoorDash now. Nobody taught it. I've been using AI every day - embarrassingly too much - without understanding how any of this happens.
1
1
54
It’s a great story - Counting words had no meaning. Word2Vec gave words meaning but forgot context. RNNs added context but forgot the beginning by the time they reached the end. Attention fixed memory but was slow. Transformers made it parallel.
1
36
Each fix solved the earlier problems and broke something new. Twenty years of that — and here we are. That evolution is beautiful.
First blog post: alok87.in/blogs/foundation-m…
25






