Systems/Infra: https://xitter.cf/t.co/dNh1a32l4b OSS/Fun: https://xitter.cf/t.co/8lgUfUaxD3, https://xitter.cf/t.co/3ViIajTa54 Else: https://xitter.cf/t.co/iQYfvIEzGW @WasmChicago
Chicago || SF
Joined October 2016
- Tweets4.4K
- Following1.4K
- Followers1.6K
- Likes13.2K
Pinned Tweet
We’re back, and for a limited time only, my good people of San Francisco.
Interrupt me and let’s talk about:
- The sweet, sweet irony of the s*ndb*x cos now trying to distance themselves from the word but still bragging about deployments of such,
- Why ~60% of the vulns revealed in the last 48hrs could have been made inert with slightly different underlying implementations at the compute layer,
- And why there is, in fact, a CPU surplus, and no I don’t mean reselling compute from your grandma’s Dell that’s still running Windows 95 (though that actually sounds like a fun project)
Come through! – at San Francisco, CA
For the foreseeable future, I'll be spending every morning at this Blue Bottle in SF, at 2nd and South Park St, in SOMA in SF, working on some new infra experiments, existing projects, having meetings, and drinking coffee.
Stop me if you wanna talk Virtualization, the future of Agent Cloud Infra, why MicroVMs are almost fully commoditized, what to do about the CPU crunch, or anything else!
(I'm dead serious, interrupt me, even if I'm looking at my terminal angrily 😄)
ALT Blue Bottle SOMA, 2nd and S Park st.
What is the upper bound of agent parallelism if you could definitively solve for instantly available, safely-isolated, and universal compute?
Communication overhead? or command and control tax?
Not sure there is one for certain tasks, and I don’t think our mental models are accurate for the scale that could exist here
Goddamn, reading / listening is the bottleneck now, isn't it
Where's the ai that allows me to listen / read (viz. understand) something complex 100000x faster, or at least at the speed all this new stuff is being created
the feeling very much related to:
Man this is timely. I have been thinking about this a lot, in an unstructured way in my head/in the background of my thoughts, and it seems to be true at every turn.
Every time I have an inkling recently that something should be 1000x more impressive or robust or safe or fast or integrated etc., not only can it be done, but in fact that's then the new threshold for software of the kind that matters.
danluu.com/ambitious/
Man this is timely. I have been thinking about this a lot, in an unstructured way in my head/in the background of my thoughts, and it seems to be true at every turn.
Every time I have an inkling recently that something should be 1000x more impressive or robust or safe or fast or integrated etc., not only can it be done, but in fact that's then the new threshold for software of the kind that matters.
danluu.com/ambitious/
It is so Sunday today, isn’t it. Feels like it.
open.spotify.com/track/3MRQn…
Weekend hacking crew wya
We have LLMs to cajole, and infrastructures to decimate and remake in our image
join me ☕️ – at San Francisco, CA
We’re back, and for a limited time only, my good people of San Francisco.
Interrupt me and let’s talk about:
- The sweet, sweet irony of the s*ndb*x cos now trying to distance themselves from the word but still bragging about deployments of such,
- Why ~60% of the vulns revealed in the last 48hrs could have been made inert with slightly different underlying implementations at the compute layer,
- And why there is, in fact, a CPU surplus, and no I don’t mean reselling compute from your grandma’s Dell that’s still running Windows 95 (though that actually sounds like a fun project)
Come through!
Dan Phillips retweeted
We’re going to see a bunch of neolab to neocloud pivots next year
At the tail end of the sandboxing discussion here, @bernhardsson is right: what a new compute primitive should be needs to evolve, and indeed, should have already in many respects, during the first cloud era.
In our opinion, isolated compute, with the utility of a Linux VM, should be cheaper than a Linux _thread_, and indeed safer, and more universal/fungible.
Snapshotting/Restoring, Live-migration, Forking should be fundamentals that can be assumed to work as well as flipping on a light switch, across chips and architectures. It should be able to be embedded, and verified and "be stateful and live forever", or killed and revived 100000s of times per second, without fail.
We're at the beginning of the Arc of incredible Virtualization tech, not the end. – at San Francisco, CA
Last week, @bernhardsson (Founder & CEO, @modal) announced a $750M Series D. I sat down with him the prior week to discuss a ton of stuff about the business.
We got into:
-why open-source models made fine-tuning a lot less necessary
-how Modal's team grew 4x in a year and could hit ~700 people next year
-why the frontier labs have the same business model as music artists
and much, much, more…
Enjoy!
Timestamps:
0:00 What Modal does
1:25 From @Spotify to building @modal
4:19 Founding before the generative AI boom
7:43 First customers and fundraising
10:00 Building for inference and autoscaling
13:21 Where Modal gets GPU capacity
15:39 Renting vs. owning infrastructure
19:45 The risk of GPU prices falling
24:18 Is AI demand infinite?
28:25 What Modal can learn from neoclouds
30:26 GPUs as an asset class
31:55 Modal’s products: functions and sandboxes
36:03 The case for smaller training workloads
41:21 How agent sandboxes are changing
46:48 @AnthropicAI and sandbox security
49:29 Using agents to test security
52:51 Developer experience in the age of agents
57:45 What Modal’s forward-deployed engineers do
59:55 Modal’s approach to product management
1:03:29 Optional code reviews and engineering culture
1:05:56 Cloud, on-prem and bring-your-own-cloud
1:09:25 Capacity, fundraising and growth
1:13:07 Scaling the team
1:17:28 Erik’s tweets: ephemeral sandboxes
1:18:31 The problem with rectangular GPU contracts
1:21:20 Inference providers vs. model builders
1:23:25 Where models improve next
Learn more about what we are doing here (sign up to the mailer): boxer.dev
And checkout (and feel free to contribute to (!)) Weave: github.com/dphilla/weave – at San Francisco, CA
Dan Phillips retweeted
good morning sf. happy building this weekend. remember: optimism is a force multiplier.
It's Friday people
We're slowly strolling through Mission Bay, vibing and talking about single-digit μs startup times
If you are in the area, hmu and come hang with us
open.spotify.com/track/4gRA0… – at San Francisco, CA
One thing about living in SF this fall
Way less discussion around Rice-A-Roni than one would expect, just sayin – at San Francisco, CA
Dan Phillips retweeted
These are interesting times, we see agents hacking stuff and at the same time formally verifying oss software.
We are hosting a formal methods hackathon along with some of the best folks in this space including @theoremlabs
I’m super excited about all the things people are going to “vibecheck” at the event.
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
The general direction of this is exactly where so many things should go. Very cool work @meekaale – at San Francisco, CA
