I write code that works, sometimes @modal
Joined June 2022
- Tweets 3,122
- Following 273
- Followers 2,633
- Likes 4,646
243 Photos and videos
Amit retweeted
Jun 10
Okay, that said:
we cut burst TTI on @modal sandboxes by over half!
no benchmark funny business, just hard systems engineering
Jun 8
One funny thing about sandbox startup latency is that “started” often means something much narrower than users think it means
dinner isn't ready just because the oven's on 🧵
8
3
57
11,955
> I've seen a Mythos output and it was the most incredible AI produced software I've ever seen.
Serious question: If this is true, why isn't Anthropic pumping out a bunch of incredible software?
Is Mythos better at smaller projects not worth publishing? Or mostly cybersecurity?
Jun 6
random shit
→ hvm5 done, ~5x faster than hvm4, supgen is happy
→ AGI = when you can train Mythos from scratch in your laptop
→ that is only possible through a whole new approach to AI
→ my goal is still to try. because why not? the space of ideas is vast
→ I've seen a Mythos output (not posted here) and it was the most incredible AI produced software I've ever seen. so I was right that these saying it is *just* marketing are in pure denial. it is marketing. on top of a massive breakthrough
→ I asked Opus 4.8 and GPT 5.5 to work in a long term impossible goal. 1 day layer, both failed, but Opus had a ~3k loc file that was savable. GPT 5.5 had a ~15k loc file that had zero value... :|
tldr: more free ad to a company that gives zero fucks about me (/us). it is the truth though, what can I do about it
2
8
1,024
cannot wait to make judicious use of elixir when type annotations land
Jun 3
Elixir v1.20 released! Now officially a gradually typed language: Elixir type checks every single line of code, finding bugs and dead code, without developer overhead (no typing signatures) and extremely low false positives rate. Plus a faster compiler! Links and reports below.
1
181
> π to four decimals in 16 seconds.
I don’t even think an LLM would market itself this badly
May 30
>backend has screenshots too. xd
What "run your code in the cloud" actually looks like:
50 isolated microVMs, every CPU thread pinned, one box. Threw a billion random samples at it → π to four decimals in 16 seconds.
no servers to set up.
just "burst run ."
ALT screenshot of htop showing 12 cores of a cpu maxed to 100% use
8
15
1,442
83,771
weekend plans just dropped 👀
May 30
1
580
I can totally see how people can get into timekeeping.
@geerlingguy was ahead of his time
2
131
time to buy some more GPUs
May 21
Today we're announcing our Series C funding: $355M at a $4.65B valuation, led by some great investors @generalcatalyst and @Redpoint.
We've had insane growth in the last year, but we're still very early. So proud of the team and what we have built so far!
28
2,356
Don't usually post my blog posts on here (maybe I should?)
amit.prasad.me/blog/sane-nix
1
4
593