HNHacker News
TopNewBestAskShowJobs

carlineng

656 karma · joined March 12, 2015

Working on the Malloy query language at Meta. Our mission is to replace analytical SQL. Check us out at https://malloydata.github.io/documentation/index.html

[ my public key: https://keybase.io/carlineng; my proof: https://keybase.io/carlineng/sigs/SxKhzboFFXEbC-x-9-NGEtke2DKLBS2wJCzYxL2vPV8 ]

submissionscomments
carlineng··on A sequel to SQL? An intro to Malloy
PRQL is very cool, but I think Malloy is meaningfully different (and more useful) because of its inclusion of a semantic layer into the core language.
carlineng··on A sequel to SQL? An intro to Malloy
I think Malloy is meaningfully different from previous attempts (e.g., PRQL), and describe why in the post, namely the inclusion of a semantic layer as part of the language. Take a look at the post, and would love to hear if you agree or not.
carlineng··on A sequel to SQL? An intro to Malloy
My musings on why SQL has been so hard to displace, and what it might take to do so. Any and all feedback appreciated!
carlineng··on A Critique of SQL, 40 Years Later
Author here. Fixed the mistake. You may now proceed with reading the remainder of the post ;)
carlineng··on Why is Snowflake so expensive
[Disclaimer: former Snowflake employee]

Snowflake is not expensive because of perverse incentives, which is the primary claim of the article. It is expensive because it is a highly differentiated and very sticky product.

As others have mentioned, competition is the ultimate incentive to work on performance. Every dollar of Snowflake revenue is a dollar of revenue that Amazon, Google, Microsoft and Databricks are fighting for.

carlineng··on Unprecedented 100% of First 14 Patients with Cancer Respond to Dostarlimab
Clinical trials for cancer treatment rarely use placebos. Instead, the control group is the current standard of care:

https://www.cancer.net/research-and-advocacy/clinical-trials...

carlineng··on How to structure your sales compensation plan to deliberately undersell
For businesses where a significant percentage of customer LTV is in the expand phase, I'm surprised it's not more commonplace to have AEs take on the role of customer success. The AE usually has a lot of context about the customer that is likely to be lost in the transition to a CSM. Doing this also has the benefit of not needing to pay two people for the expansion.
carlineng··on Ask HN: YouTube Channels for the Intellectually Curious
MIT OpenCourseware: https://www.youtube.com/c/mitocw Yale Courses: https://www.youtube.com/user/YaleCourses
carlineng··on The Creepiest Cycling Condition: Shermer’s Neck
Radiolab did an episode profiling Jure Robič [1], a former member of the Slovenian special forces who took up ultra-endurance cycling and won RAAM 5 times. It's a mind-boggling story. Tragically, Jure was killed by a car while training in Slovenia in 2010.

[1]: https://radiolab.org/episodes/91709-limits

carlineng··on The Creepiest Cycling Condition: Shermer’s Neck
Good thing there is an enormous list of reasons to cycle that far outweigh the downsides.

Shermer's Neck is a condition that impacts a very small percentage of cyclists, and only those who choose to participate in the most extreme of events like Race Across America or Paris-Brest-Paris.

carlineng··on Three arrows founders break silence over collapse of crypto hedge fund
"But then yeah, after that day, when, you know, Bitcoin went from $30,000 to $20,000, you know, that, that was extremely painful for us. And that was in, that ended up being kind of the nail in the coffin."

Bitcoin has had 10 drawdowns of 40% or more in its short history of 13 years [0], and yet a mere drop of 33% caused them major problems. Doesn't sound to me like they can blame "a systemic failure of risk management"; more like a profound lack of understanding by these two individuals.

[0]: https://www.livewiremarkets.com/wires/analysing-bitcoin-draw...

carlineng··on Friendlier SQL with DuckDB
I love these updates. It would be great to see some of the major data warehouse vendors (Snowflake, BigQuery, Redshift) follow suit.
carlineng··on The fastest tool for querying large JSON files is written in Python (benchmark)
Snowflake’s handling of JSON data via the “variant” data type is quite nice to work with, both in terms of speed and convenience of the access methods.
carlineng··on Why is it hard to buy things that work well?
"Variants of this idea that I frequently hear engineers and VCs repeat involve companies being efficient and/or products being basically as good as possible because, if it were possible for them to be better, someone would've outcompeted them and done it already"

Reminds me of the old joke -- an Economist walks past a $100 bill on the sidewalk, but doesn't bother to pick it up. He turns to his friend and declares smugly: "If it were real, someone would have taken it already."

carlineng··on Apache Arrow Flight SQL: Accelerating Database Access
Mark Raasveldt and Hannes Muhleisen at CWI wrote about these challenges in their 2017 paper Don't Hold My Data Hostage [0]. It's great to see progress on this front.

[0] https://ir.cwi.nl/pub/26415

carlineng··on Databases in 2021: A Year in Review
Cloud SQL data warehouses like Snowflake and BigQuery.
carlineng··on The Importance of Price Signals
> You can tell because she gets a lot of things wrong.

That puts her in good company with pretty much all of the world's most respected economists.

carlineng··on Japan to implement bank-deposit-backed digital currency in 2022
I'm somewhat bullish on the promise of permissioned blockchain systems, but the Facebook/MySQL example is misleading --

* Facebook doesn't need MySQL to interop with other systems in the same way that blockchain enables.

* Relational semantics are way more complex than what most distributed ledgers technologies offer. A blockchain is more analogous to the MySQL binlog than to MySQL itself. I would guess that Facebook's DB transaction log is orders of magnitude larger than all the blockchains in the world combined.

* Facebook's challenges with MySQL are most likely due to their unprecedented scale. If one were to layer on the governance challenges associated with distributed ledger systems, I would think an upgrade like that would basically be impossible.

carlineng··on Malloy: An Experimental Language for Data
A project from the creators of Looker. From the docs [1]:

> Malloy was designed by a team of people with a lot of experience in understanding the task of extracting meaning from data. Years of constant exposure to SQL resulted in a tremendous sense of wonder at the power of SQL ... and a tremendous source of frustration at how bad SQL is at representing the types of operations needed to get meaningful data out of relational databases, and how difficult it is to maintain and extend a complex set of transformations written in SQL.

> Malloy started as a "clean slate" thought experiment, if we knew all the things we know about data, and about programming with data, and about programming languages in general, and we were designing a query language today, what would it look like.

> In it's earliest form, Malloy looked a lot like SQL, and much of Malloy continues to be influenced by the overall design of the SQL language. It was always the intention that there could be a document that would contain both Malloy and SQL and those would make sense together, perhaps parallel to the way Javascript and Typescript work together.

This statement resonates a lot with me. The VSCode plugin already looks fairly built out, and there's a native BigQuery integration.

[1]: https://looker-open-source.github.io/malloy/documentation/la...

carlineng··on App Store chief to Tim Cook: “we should not use these download figures publicly”
Next time I’m asked what data scientists do, I’m sending them this email exchange. This conversation likely resulted in a few sleepless nights for a couple of Apple data scientists.
carlineng··on Amplitude (YC W12) IPO S1
Gross Margin for Calendar Year 2020 was around 71%, so no more than 29c.
carlineng··on Confluent S-1
A few tidbits after a quick reading:

* Confluent Platform (on-prem) accounted for 85% of revenue in 2020, whereas Confluent Cloud (managed SaaS) accounted for 15%. The S-1 makes it clear that Cloud is the future for them. This split makes it a bit tough to interpret their Gross Margin numbers though; as there's no breakout of Cost of Revenue between the two.

* Confluent Platform is subscription-based whereas Confluent Cloud is consumption-based, which further complicates modeling the business.

* Fiscal Year 2017 Cohort must have had some whale customers -- it looks larger than both FY18 and FY19 cohorts (p. 86)

* Expanding net loss -- 97% of revenue in 2020 vs 63% of revenue in 2019. This looks primarily driven by $111.9 million worth of stock-based compensation (tender offer, secondary sales) in the lead-up to the IPO. Excluding that (which is questionable, but still tells us something about how company operations are developing), net loss is ~50% of revenue

* Revenue Growth from 2018 to 2019 was ~130%, but they weren't able to maintain that meteoric trajectory in 2020; revenue growth has slowed to ~58% from 2019 to 2020

carlineng··on Launch HN: Mozart Data (YC S20) – One-stop shop for a modern data pipeline
Reminds me of Panoply, which had native SaaS integrations, a managed Redshift instance on the back-end, and a BI layer on top. Basically a fully turn-key "modern data stack" [1]. The stack is way easier to operate than it has ever been before, but still requires folks with expertise to manage each of the components.

[1] https://blog.getdbt.com/future-of-the-modern-data-stack/

carlineng··on Sovereign writers and Substack
I would argue that yes, the fundamental incentives are different. Traditional media relied on an advertising-based revenue model, that needed to maximize eyeballs to be attractive to advertisers. Substack relies on subscriptions, which can operate at a much smaller scale. Matthew Yglesias's 10k subscribers is not nearly enough to be attractive for an advertiser, but more than enough to support a single writer's subscription business. Through Substack, writers are incentivized to write things they think that people will pay for, not just things they think people will click on.
carlineng··on Instacart mulls direct listing in snub to IPOs
Practically speaking, how is "raise money on private market, then quickly go public via Direct Listing" different from a traditional IPO? Presumably the late stage private investors get the company at a steep discount relative to what would be available to retail investors, so it doesn't seem that different from an IPO where the majority of shares go to select institutional investors. If the shares pop 50% after the Direct Listing, then couldn't one argue that the company left money on the table during the pre-Direct Listing private round?
carlineng··on Police misconduct settlements
This is fantastic. When people ask me "what do data scientists really do?" I am going to send them to this repo and the accompanying article. This looks like a massive effort to send out each FOIA request and then manually process each individual report. At the end of the day, their conclusion has little to do with actually investigating trends about police misconduct, and is almost entirely about standardization of data collection practices.
carlineng··on Coinbase S-1
Some sources say Coinbase has already been trading at >$100B on private markets [1].

I find it interesting that about 2/3 of their trading volume (and thus revenue) is from institutions. Also that 44% of their trading volume is from "Other Crypto Assets", which implies that there must be a rather large bit of institutional activity in "Other Crypto Assets", and it's not just the stereotypical retail crypto investor trading in alt-coins.

[1]: https://www.axios.com/coinbase-valued-100-billion-direct-lis...

carlineng··on Coinbase S-1
Trading volume of "Other Crypto Assets" (non-BTC/ETH/LTC) made up 44% of revenue in 2020, which they attribute to DeFi crypto assets. I wonder how 13% of their assets drives 44% of their revenue? The section on Applications talks about DeFi as peer-to-peer financial agreements, which implies there's some underlying economic activity going on, as opposed to just speculation, but I haven't been able to find anything else that digs deeper into that piece.
carlineng··on Coinbase S-1
In a Direct Listing, the company does not issue new shares. This doesn't mean they can't in the future, but it's just not part of the company's initial listing on the public markets.
carlineng··on Coinbase S-1
From "Use of Proceeds" on p78:

"To the extent any registered stockholder chooses to sell shares of our Class A common stock covered by this prospectus, we will not receive any proceeds from any such sales of our Class A common stock."

This is a direct listing, so I think its primarily a liquidity event for investors and employees.

← PreviousPage 2 of 4Next →