Joel Labes
LightNews LightNews
banner
joellab.es
Joel Labes
@joellab.es
1.4K followers 90 following 84 posts
I don't know how to be data driven and at this point I'm too afraid to ask.

DX at dbt Labs | Wellington NZ 🥝
Posts Replies Media Videos
joellab.es
Imagine buying out every ad space in the NYC subway or whatever when you surely must know @vicmsong.bsky.social is gonna get your ass because your thing doesn't work

www.theverge.com/column/79101...
A screenshot of the linked article: Another time, I asked what it thought of my day. (Because really, how does one strike up an organic conversation with an always-listening AI?) It responded by saying, “What makes you think I have thoughts on your day, Vee?” That’s a weird response considering it’s the entire pitch behind the device.
October 4, 2025 at 12:58 AM
joellab.es
You did say "simple" and this is absolutely not but 🤷‍♂️
May 9, 2025 at 8:34 PM
joellab.es
🦷
A modified screenshot of a tweet. [Compiler waiting room] Me: [chanting] types, types– Other logical plans: types, TYPES Execution engine: [pounding her clipboard] TYPES, TYPES, TYPES!
January 28, 2025 at 11:34 PM
joellab.es
Hmm have you got multi-project going on? I'm seeing everything filtered out except for `customers` and `stores.v1` which are coming from an upstream project (unclear to me right now whether that's a feature or a bug).
A screenshot of dbt Explorer with the `+transactions` selector. It shows all parent nodes of the transactions model, plus a purple node showing that transactions is used by a downstream project. A screenshot of dbt Explorer with the ` stg_tpcds_core__web_sales+ +transactions` union selector. It shows the same nodes as the first screenshot, plus orange nodes downstream of stg_tpcds_core__web_sales identifying exposures detected from Tableau. A screenshot of dbt Explorer with the `stg_tpcds_core__web_sales+,+transactions` intersection selector. It shows `stg_tpcds_core__web_sales` and `int_sales__unioned` feeding into `transactions`. The upstream and downstream project references are also displayed.
January 17, 2025 at 3:45 AM
joellab.es
I was going to shoot off a fast dumb comment yesterday about how that's incompatible with the expectations of users today, but wasn't happy with the phrasing. The good news is that this nails it: www.secoda.co/blog/dbt-sdf...
This is exactly what's happening with dbt. When it started, it was beautifully simple: just write SQL with some Jinja templates. But as data teams started building more complex systems with dbt, their needs adapted with it. We needed better error messages, faster compile times, better IDE integration. The acquisition of SDF is dbt really growing up. And this isn't a bad thing – it's necessary. When tools become critical infrastructure, we need to grow more sophisticated to meet the needs of their users. But there's something subtle and important happening here that most people are missing. The really interesting part isn't that dbt is adding features – it's that we’re adding understanding. SDF's technology doesn't just run SQL faster (which solves a core user problem in dbt), it also makes the process of developing with dbt much more pleasant and metadata rich.
January 16, 2025 at 4:10 AM
joellab.es
Got ur back. Exactly how we do it still tbd, but improvements 100% coming for core users!
All of the benefits laid out above are being realized in production today by SDF’s existing customers. But it will take some work to get all of these capabilities integrated into dbt, and this won’t happen overnight. Our first goal is to get SDF’s SQL parsing capabilities integrated into dbt. These capabilities will enable meaningful improvements to the dbt developer experience, and we want everyone to have access to them. While SDF won't be included as part of the Apache 2.0 code base, we plan to make meaningful parts of SDF’s capabilities available to all dbt users—whether you’re using dbt Core or dbt Cloud. As we work through integration details we’ll share more about how this will work. *But if you use dbt today, you’ll be able to use this new tech.*
January 14, 2025 at 10:45 PM
joellab.es
I found this a really useful catchup on a year's worth of LLM news: simonwillison.net/2024/Dec/31/...
A screenshot of the table of contents: The GPT-4 barrier was comprehensively broken Some of those GPT-4 models run on my laptop LLM prices crashed, thanks to competition and increased efficiency Multimodal vision is common, audio and video are starting to emerge Voice and live camera mode are science fiction come to life Prompt driven app generation is a commodity already Universal access to the best models lasted for just a few short months “Agents” still haven’t really happened yet Evals really matter Apple Intelligence is bad, Apple’s MLX library is excellent The rise of inference-scaling “reasoning” models Was the best currently available LLM trained in China for less than $6m? The environmental impact got better The environmental impact got much, much worse The year of slop Synthetic training data works great LLMs somehow got even harder to use Knowledge is incredibly unevenly distributed LLMs need better criticism
January 4, 2025 at 12:46 AM
joellab.es
logging into LinkedIn like
A screenshot of a tweet by audrey farnsworth which has been edited. It says "(opening my god damn bad take website) there better not be any god damn bad takes in here". bad take has been edited over the original text of the tweet.
December 11, 2024 at 10:33 PM
joellab.es
Looks like I'm stuck in traffic behind Cory Doctorow?
A car with the number plate DRM666
December 9, 2024 at 3:42 AM
joellab.es
While reviewing @faithebear.bsky.social's post, this forced its way into my brain so now you're stuck with it too.

If you have literally no testing, the classics are classic for a reason. But when you're ready to go to the next level, make sure you've read this article first.
The drowning high five meme. The drowning hand is labelled "low data quality". "testing" reaches out to help. In the third panel, it high-fives, captioned with dbt's testing syntax to add unique and not_null tests. The hand slips below the surface.
November 11, 2024 at 5:56 PM
joellab.es
The three scourges
A photo of two signs at the entrance to a store. One says "firearms prohibited" and the other, larger sign says "we don't accept Apple Pay or Google Pay"
November 5, 2024 at 11:24 PM
joellab.es
Just watched today's MBP announcement, and completely baffled that Apple thinks an AI-generated cartoon jellyfish would be an appropriate thing to insert into an *academic paper about jellyfish*

#appleevent
A screenshot of the Apple Event video at https://youtu.be/G0cmfY7qdmY?t=347 of a woman sitting in an aquarium environment writing on a MacBook Pro. The caption says "for example, if you're a researcher" A screenshot of the Apple Event video at https://youtu.be/G0cmfY7qdmY?t=347 of a woman sitting in an aquarium environment dragging an AI-generated image into her paper. The caption says "that you created in Image Playground"
October 31, 2024 at 1:32 AM