James Johndrow

@jakeserval @clay @vercel @togethercompute @getserval Congratulations! And much more to come I’m sure!

Serval is #1 on the Enterprise Tech 30. What a surreal moment. Special to see so many of our customers here like @clay, @vercel, @togethercompute, and many more we'll announce soon. I’m so grateful for the entire @getserval team that powers this rocket ship, especially my cofounder Alex McLeod and COO @TatianaBirgisso, who have assembled unrivaled teams across engineering and GTM. It’s Serval now. newcomer.co/p/mintlify-ser…

@jakeserval Congratulations @jakeserval and your entire team! It’s amazing what a great product you’ve built in such a short period of time. I can’t wait to see what comes next!

IT powers the workplace, and they’re hurting. Their tools are abysmal. Their workloads are unsustainable. They’re stuck in a vicious cycle of break/fix. But IT doesn’t just want to fix things. They want to BUILD things. They want to BUILD automations for: ☑️ help desk requests. ☑️ onboarding and offboarding ☑️ software access requests and reviews ☑️ asset management ❌ They DON’T want AI slopbots. ❌ They DON’T want tradeoffs between efficiency and security. ❌ They DON’T want legacy software that belongs in a museum. At Serval, we built the AI platform IT teams deserve: an all-in-one suite of AI-powered tools to automate help desk requests, build workflows, transform ticketing, and manage software access and assets. Serval is more than the next generation ITSM. It’s a reimagining of how IT operates. It’s rocket fuel for IT teams. It’s a glimpse into a future where IT BUILDS - not just for themselves, but for the entire organization. We heard from one customer “I would give up a kidney to keep Serval in our org.” The good news is, we won’t ask you to 🙂 But that’s how big this pain is and how much value Serval delivers. Today, Serval takes a huge leap forward, announcing our $75M Series B led by @sequoia, taking our valuation to $1 billion just 20 months after founding. This is still the beginning, and there is so much to build. But today I’m so proud of our team and all we’ve accomplished for our customers. Thank you to the servals that have worked so hard to make this possible, especially my cofounder Alex McLeod and our COO/my wife Tatiana Birgisson. Thank you to our customers who have believed in our vision from the very beginning. Thank you to our new investors, including Sequoia, Meritech, Evantic, Tenacity Capital, Sound Ventures, Radical VC, and Frank Slootman, alongside returning investors Redpoint, First Round, and General Catalyst. And most of all, thank you to all the IT teams that enable our most meaningful work. We’re working to repay the favor. ⬇️ See thread below for feature by Reuters, and our blog post on the raise!

@shortstein @KLdivergence I’m not sure what result you’re invoking here. Lebesgue measure is the right Haar measure for the additive group, but this isn’t a location parameter. I’m trying to guess why line 1 starts with P~uniform. The constant risk Bayes argument someone posted does prove minimax.

@KLdivergence @shortstein It looks like your proof is computing the expectation both over P and the data sequence X_i conditioned on P. This is the “integrated risk” not the MSE, it’s a fundamentally Bayesian quantity.

@KLdivergence @shortstein I’m uncertain where the Beta comes from. Are you trying to prove that the optimal estimator of p in a “Bayesian” setup where p ~ Beta and X_i ~ iid Bernoulli(p) has MSE Omega(1/n)? Your initial claim seems to have fixed p, and a “frequentist” setup would have fixed unknown p.

@KLdivergence Your co author also appreciated this comment. Thanks, yes, I am a statistics professor, but it’s always nice to have the validation of my statistics competency.

I take it as a compliment when I get reviews on my applied work like "the statistics seems competent" (which, yes, I do have a PhD in the topic) because it means my subject area understanding is so good they assume I'm an expert in the applied area and not statistics.

@mcmcgriffin I guess the editor is unfamiliar with no free lunch theorems.

Got a desk rejection and a major criticism was that that our method performed less well for a subset of the parameter values we considered in simulations. Feeling disillusioned - penalizing authors for being honest about when their methods struggle seems bad for science.

@PierreEJacob @junpenglao I couple everything these days

@GermsAndNumbers @NoahHaber The ihme strategy seems to be that when the prediction for total deaths ever is exceeded like 2 weeks later you decide that it’s time to tweak things a bit rather than be like “ok clearly we aren’t good at this, let’s do something else with our time.”

We’ve been working on a number of projects with interdisciplinary teams of people at Penn including at the hospital. Check out the latest.

@KarenVladeckEsq @steve_vladeck Excellent choice of shorts

When @steve_vladeck is destroyed after a 60 minute ride and trying to rest in our room, I tell the kids that daddy is playing hide and seek and that he really wants them to find him quickly.

@vnminin David Dunson and I wrote this last year, though it isn’t really meant to be historical: academic.oup.com/biomet/article…

@vnminin At least, as far as I know, the Hastings algorithm really was developed by Hastings

Critical issue right now. "Get it yourself" will lead to disaster.

@annabarryjester Thanks for tweeting this. Epi and stat people have been saying this for a while now but the message seems to not be getting out. Extreme shortage of tests = huge undercounting of actual cases. US is doing way less testing than most countries.

San Francisco update on #COVID19 cautions "the media and the public for relying too heavily on confirmed case count as an indicator of the situation in San Francisco." 40 known cases dont represent the threat to the public. Hygiene, distance and kindness are what we need now.

@ABindoff @DavidSFink @nyalborgesmd @EpiEllie @venkmurthy @NEJM It's hard to back these kinds of guesses up with data in the US because of the continued and ongoing shortage of tests. In the US, even people showing symptoms of the disease are often not tested. So the number of cases is certainly an undercount, but v hard to know by how much.

PSA: Uninterrupted exponential growth is NEVER a realistic assumption for an outbreak model and if you think it’s ok to use for a “first pass at the numbers” then you really really shouldn’t be posting about your model on the internet.

@KLdivergence Yeah me too. I think my research program needs to become covid-centric or I'm not going to get anything done.

I spent 2 years building realistic simulated contact networks to run epidemic (& other) simulations. Much of my recent work is on estimating mortality undercounts. If there are any covid researchers out there who could use a few more (well-washed) hands on deck, lmk.

@pollsstar @Laurie_Garrett Or it could be that the test showing the patient no longer had the virus was a false negative

@Laurie_Garrett What is more scary is the re-infection - the virus is not necessarily a one-off

36-yr-old Li Liang died of #COVID19 just 5 dys after being declared recovered & discharged from a hospital. He met the official standard of recovery: 2Xs tested negative for the virus. But 2 dys before his discharge, CT scans of his lungs still showed symptoms of pneumonia. MORE

@wesmckinn Let's be real though, they'll spend half of it on a new building.

Imagine if someone gifted $250M to fund 1000+ person-years of OSS data project development. One can always hope :)

@cmyeaton Right on Caitlin. We seem to really be dropping the ball on testing.

Growth in case counts in S Korea, Italy, Iran reflects testing strategy, in part. They are looking more so they are finding more. This is why expanding diagnostic capacity in the US is so urgent. We aren’t doing enough surveillance to understand when/if/how much community spread.