Brooklyn, NY
- Tweets 1,125
- Following 373
- Followers 567
- Likes 4,217
it took about ~6hr and ~$133 in tokens, but its not vaporware - this pipeline is live and processes new recordings within 5min of hitting the GDrive
1
42
you too can save $200k/year on conversation intelligence tools by creating a @modal account, pasting this thread into codex, and getting started with these examples:
modal.com/docs/examples/batc…
modal.com/blog/low-latency-v…
55
totally agree - there’s no worse time in human history to be a technical middle manager
however, i think the skills you learn while building/managing teams are still extremely valuable, increasingly more so in the age of agents
ability to hire great people, give and receive feedback, connect work back to company strategy/impact, and break down multidisciplinary problems are arguably more important than ever
if you’re a data/analytics manager who’s excited to roll up your sleeves, you’d be a great fit for our GTM Engineering - Data Platform role
modal.jobs
Brutal truth from @Lethain
“If you're in [middle] management (…) you sort of have to decide, am I retiring now or am I re-skilling?”
He did just this: step down from leading a large team (250 ppl) to be hands-on with a small team (25 ppl), ship code & re-learn the new “rules”
4
7
616
i pride myself on having a strong bias to action, but i still find it really easy to spend a lot of time talking about role/responsibilities
it's obvious that DOING things outside of your role is a great way to increase scope
what's less obvious is that TALKING about your role can actually shrink it - you have less time to do scope expanding activities and less leverage to ask for more
1
9
468
the best time to hire a GTM Engineer to scale our data platform was before we hit a $1B revenue run rate
the second best time is right now
this person will be responsible for building our...
- company and people graphs, tracking companies, labs, researchers, and engineers
- forecasting inputs for both compute capacity (supply) and sales pipeline (demand)
- customer and user journey state machine, ensuring that we're surfacing the right action to the right person at the right time
you wouldn't be starting from scratch - we have an incredible team, a cracked engineering culture, the latest in agent infrastructure, and the mandate of Heaven to build the best GTM data platform the world has ever seen
apply at the link below and/or DM me a piece of technical writing or presentation about a data system/product you've built
jobs.ashbyhq.com/modal/dd6fe…
7
6
100
14,556
Chris Prinz retweeted
hello world! im so happy to announce that i’ve joined @modal!
when thinking about my next role post-VC, i prioritized the following:
+ a strong presence in NYC: modal and its NYC office are growing…FAST! i started yesterday and my desk has already been moved to accommodate more growth (lol)
+ talent density: even as an outside observer, the ambition at modal was abundantly clear - deep developer/customer love, a wide-reaching platform vision, an intense focus on cutting-edge research
+ a POV on brand: from its organic IC-led presence online to my conversations internally about everything from eng blog to socials to editorial projects, it was obvious that this is a team that deeply cares about how it shows up in the world
there has been quite a stir over the past few weeks (oops lol) of what “storytelling” and related work means in the world of tech... im excited to help shape what it means at modal :)
95
5
478
19,764
Everybody buying compute for RL needs to stop immediately and think about this result.
Big homogenous clusters are a sledgehammer for what is essentially an inference-shaped nail.
Fantastic work from @nanjiangwill.
Training large models with RL typically runs on a fixed homogeneous cluster of the latest accelerators for numerical stability and faster throughput. This throws away inference optimization work done for other vendors and generations, limits the capacity you can utilize, and is unnecessarily expensive.
So we trained a model on agentic coding tasks using fully asynchronous RL and mixed-precision, mixed-accelerator samplers. With off-policy correction, training stayed stable through convergence, on par with homogeneous samplers, at a lower cost.
3
12
210
19,392
ideal form factor for an agent controller imo - not always listening (hopefully), always immediately operable, discreet
super stoked about this as someone who spends a lot of time monologuing to ChatGPT on long runs
natura.inc/i/8s8r3u
2
4
344
“agent in the loop” is out
the agent is the loop, and we’re just living in it
“You can give the Engine a goal or an ambiguous workstream, and it creates the loop to understand, measure, iterate, and self-improve.”
Hear how @ChrisPrinz at @modal is rethinking GTM with Hone Engines that go beyond executing tasks to own entire workstreams, continuously measure results, and improve performance.
See what Engines can do for your organization at Hone.com.
14
929
don't get me wrong, i'm really excited that @hone officially launched today. but since we launched our engine AutoMOGS, it's taken both my job AND my best friend @ellenfclee!!
people used to come to me when they wanted to enrich lists, run campaigns, build automations, and do analysis
but now they just send a Slack to AutoMOGS. even worse - AutoMOGS will proactively Slack them with improvements to workflows and new opportunities!
i'd tread lightly around bringing Hone into your organization. and if you're one of our competitors - heed my advice and just stay away!
“Your best employees don't need to be micromanaged; your AI shouldn’t either.”
Some of our early users explain the difference they see with Engines, from proactive execution and multiplayer collaboration to continuous self-improvement and high degrees of trust:
6
4
27
3,363
Runtime is happening in two days.
Join us to hear speakers from @AnthropicAI, @typesafeai, @nvidia, @tryramp, and more.
4
22
80
14,033