Joshua Andrews

@api_assasin Others have said it but a low interest rate mortgage is one of the few, if not the only, real hedge that middle income folks have against inflation.

if cash is worth less over time due to inflation - is it "cheaper" to finance everything as long as the interest rate is not greater than inflation?

@StefanFSchubert Competence is not the same across fields. There are some areas of life where it’s not that applicable. Competence in something like “life coach” for example is laughable. Elon’s point is that what passes for competence at Harvard is really just skill at climbing a hierarchy.

This claim flatters most listeners, which probably partially explains its popularity. But it's not true. Elites are much more competent than randomly selected people.

@_Mira___Mira_ I share your frustration with “just call sales” and bloated corporate bureaucracies and wasted lunches. But… Good engineering at scale in a Demming sense requires not cutting corners, solid process, and a bigger picture respect for the impact of what you do beyond short term $.

If you are a rich VC-funded SF startup, then you are "price-agnostic" and should buy from American companies. Their fat margins are fed by your inflated valuation, so it somewhat balances. But if you are a true underdog, you study the Chinese. You replace the memory chips on used or smuggled GPUs and patch the driver to recognize the increased memory. You make custom motherboards with 16 PCIe slots and PCIe splitter chips. You hang them open-air on racks and use riser cables to connect them. You locate yourself next to a gigantic dam with excess capacity, and tell them "we would like all of your electricity at current price". You use a VPN to prevent the government from easily identifying and raiding your operations. You do whatever it takes to bring the cost as low as possible. Do not hold back at all for fear or shame. Americans like to put "Contact Sales" on every website, but they won't even respond back if you're "not worth their time". They will take your management out to lunch, laugh and joke, and charge your company a price with 500% margin. Every project has dedicated engineers getting on calls to help you(and billing you many hours by the hour). Chinese companies will make you whatever you want, at COGS + a modest margin, in bulk. If you incorrectly state what you want, you'll get something you don't want as well-deserved punishment for your stupidity. No refunds! There isn't enough margin for refunds or support. You should know what you want. NEVER trust an American who talks about "commoditizing" anything, much less a petaflop. They literally don't know how to cut manufacturing costs, because all people that may have known have been fired and replaced by outsourced manufacturing in China. They can not give you a low COGS-adjacent price even if they tried, it is not within their - or any American's - skillset! The necessary concepts are outside their vocabulary(and inside the Chinese's). Notice that the cost of LLM inference has decreased only after @deepseek_ai released their 14 cents/million token API! You will never get a competitive price from any American company, simply purchasing a product - you must make a trip to the Holy City of Shenzhen and prepare to roll up your sleeves and WORK alongside your purchase! Nobody here can help you.

@inerati Can you give some examples of why it is so bad? I’ve been meaning to try it but haven’t had a chance yet.

streamlit: an abomination before god?

@matsonj I also found when I switched to Mac for work that I hadn’t realized how frustrating Microsoft’s disrespect for users was in terms of constant patch reboots which I could never seem to control, overriding of settings, etc. Much more respectful on Mac.

one week into being a mac user, as i suspected, the only thing i'm missing really is excel. everything else is electron or web-based

@sv_techie @laurenbalik got it sorry about this. Saw “distributed SQL” mentioned and was not clear.

@Jra_tech @laurenbalik This is not about BQ This is dist sql - spanner - operational DB (oltp) not olap.

I have a lot of respect for what Amazon/AWS has done but there are two major areas they missed the boat : 1. Dist. SQL 2. LLM This also happens to be the strength of GOOG/GCP. They got Data and AI stack, including HW (TPU) right.

@pdrmnvd ask a developer to solve a problem for you and he writes a bunch of code. have to go back and say “no I meant the most efficient way to solve the problem and if you must , code”

@ergestx Agreed. Which is why I always think timestamp data types that explicitly force/include definition of time zones are superior to those that don’t (if you have a choice.)

Whenever you see a timestamp field in a table your first question should ALWAYS be “what time zone is this in?” Trust me on that one 😀

@NateSooter And we are going…

@NateSooter We are a services firm so we are a vendor but we are also practitioners… I’m just as happy to talk the finer details of SQL and query plans and data modeling without anyone buying anything…

@mikpanko Need the ability to version code and write automated tests in code. Then whole process of PR reviews, merges, etc. A lot of these GUI tools don’t support that at all, or if they do in some way it doesn’t really fully and cleanly integrate with git.

Folks who use SQL or Python for data analysis, why do / don't you use point-and-click product analytics tools like Amplitude, Mixpanel, Heap, Pendo, etc? I observed a clear divide here between technical and non-technical folks in my career so far.

@NateSooter I disagree on this. I see tons of cases where the source application where data entry is done has no validation, no reference vocabulary, no referential integrity, and no responsibility for data quality.

Interesting pattern I see now that I'm off the data team and in the business: Near universal blame on "data quality". Which isn't the root problem at all! Rather, the problem is with how data moves (or doesn't) from raw to the product a business user is consuming.

@natmiletic that…. is so true

The longer you work in tech, the stronger the urge to quit and start farming

@SashaMTL Worse than that. Underlying all of that, everything, is philosophy. You can’t make any claims about any of these things without a philsophical grounding for them. And most modern STEM folks have no education in this and don’t think they need it.

Having studied cognitive science, I am regularly appalled by the way many AI folks see cognition/intelligence. Literally ignoring centuries of research in everything from psychology to education and neuroscience..

@medriscoll @pwang I don’t understand this because, for example, everyone in a business wants one “customer” table. Nobody wants to go look at a bunch of Kafka streams to deduce their own definition of that. And getting that kind of table requires tons of complex modeling.

@pwang @Jra_tech Nicely put.

The "structured versus unstructured" taxonomy is outdated and unhelpful, a shibboleth of the Hadoop era. Data infrastructure today is dominated by *structured* data: JSON events, Kafka payloads, Postgres CDC logs, Hive-partitioned buckets of Parquet. "Modeled versus unmodeled" is a more meaningful taxonomy, and is at the heart of the philosophical divide between warehouse and data lake architectures. Data lakes store unmodeled data while data warehouses love modeled data. But as companies and their data scales up, the models required to run the business increase combinatorially. For warehouse architectures, where (most) models are materialized as tables, what's the result? Instacart's 1000s of models, massive data team, a $28m annual payment to Snowflake. This is why in the long term, unmodeled but well-structured data lakes will dominate. Their flexibility and efficiency -- given that most models can be built at query-time, and can be ephemeral -- is aligned with cloud computing's evolution towards fully serverless systems.

@sherrod_im hah I was at Caesar’s in June and the same thing was showing in the elevators…. terrible that they still haven’t fixed it.

August 13, 2023. Caesar’s palace elevator.

@imrobertyi Hard part is when you are both AKA “Head of Data”

you either die a data person or you live long enough to see yourself become the stakeholder

@antonycourtney @dhh it’s funny how this is so similar to a lot of SQL/NoSQL arguments. For a while a lot of the NoSQL folks were saying “you don’t need schema!” (hence no types.) Wellllll… You don’t hear that as much now.

I guess every generation needs to re-learn the value of types. I take solace in the fact that most in my feed aren't buying @dhh's argument. For an old school perspective, this is from the classic 1989 paper "Typeful Programming" by Luca Cardelli ( lucacardelli.name/Papers/Typeful… ):