Pinned Tweet
Thanks @rasdani_ for introducing me to @PrimeIntellect
Just finished 2 months, and I am feeling blessed. Working at this company has really updated my mental priors wrt what it means to work at a company with a mission.
Such companies are rare, but exist!
9
3
84
27,829
I taught myself programming at the age of 15.
My first project was a bash program to calculate Pi.
My second project was an Operating System.
7
68
2,206
my morning today
> wake up at 6am
> check Slack
> see a client request for more sandbox concurrency
> 2 mins later, authorize their concurrency upgrade
> go to the gym
2
47
1,352
I’ve spent much of the past year thinking about, building and sometime stressing about sandboxes.
I'm happy and proud to present you the result of our work:
Introducing Prime Sandboxes:
MicroVM sandboxes purpose-built for RL training.
Model training requires running tens of thousands of concurrent sandboxes, leading to complex and costly configuration. We built Prime Sandboxes for our own team. Today we're releasing them publicly.
14
11
144
10,709
We have some crazy features on our roadmap: GPU sandboxes, state snapshotting, sandbox forking to name a few.
This is co-developed with our researchers to stay at the frontier of what's needed.
Make this a reality with us: primeintellect.ai/careers/26…
15
224
Does anyone stop and think:
Why did the AI discourse go from: “It’s gonna take your jobs” to, all of a sudden, “it’s going to exterminate all of humanity” … and it still hasn’t taken our jobs yet?
285
340
4,180
103,794
This week’s AI safety discourse is a prime example of The Butterfly Effect.
It all started with a 28 year old leaving his job of 8 weeks.
1
3
401
.@github I appreciate the effort to improve the UX, but this unused, wasted space in the new PR sidebar is the opposite of just that.
2
349
i never fail to be impressed by the hubris of some people
bro was at anthropic for 6 weeks, barely finished the claude onboarding, doesn't know the names of all his colleagues
something is smells 🐟
2
2
38
1,468
i love how in a codebase, i can see which parts were written by Fable/Opus by the excessive use of redundant code commands
and which were written by Astra by the minified code
2
18
1,174
Can we put Astra/Fable behind the OSINT data collected on March 8, 2014 to narrow down what happened with MH370?
I think they could uncover hidden patterns in the data and help make progress, 12 years later.
Would shine a public, positive light on AI too. @OpenAI @AnthropicAI
4
1
41
3,961
astra is sure going to be the first model to make me go buy a 3D printer so that I can do my toying around in the real world now
1
16
731
i'm shocked that the AI labs don't have a custom harness + model-tune for reviewing (slop) PR diffs
devs would get addicted, burn millions more tokens and pay more money
1
13
1,524
It seems like Anthropic finally put a token length penalty in their Fable post-training. Long overdue.
The model finally doesn't blabber and reason for too long and just gets going on the task.
4
15
1,244
In preparation for the new Astra release, i'm building a PR review TUI that better visualizes the intent of your diff
This lets you review one level higher than the code
rarely will we need to read the code directly
We need a new tool, and GitHub isn't going to make it for us.
5
52
3,237
i wanted to visualize my tweet-in-jest, as it is technically true:
once the WAL is written to, every other structure, tables, indices, can be rebuilt. if the DB crashes, no data is lost
by this definition, the WAL is the true store and the tables are secondary indices
12
970