The Boba Guys result is a real kick in the nuts. Using a shitty paperish straw for the environment but the core product being so high in these tests.

The "paper" straw, that's still coated in plastic and becomes unusably soggy long before the product is consumed.

i mean you're lucky if it's just coated in plastic, most are PFAS.

20x variation in their measurements of "Black Tea Pearls" (and 8x or so in the tea juice). Would have liked to see more reflection on that.

(I feel like I'm still seeing plastic straws for boba everywhere in San Jose; but I'm far from a frequent consumer)

Think of the turtle.

Plastic straws are natural predators. Removing them from the food chain will simply cause overpopulation in turtles.

Cool. Rolling your own BI seems fraught with peril at most orgs where I imagine the buy vs build decision is always buy. How many PowerBI or Tableau seats do you need before rolling your own internal BI platform starts to make sense?

I don't think it ever makes sense because the large players will always be able to make new innovations (mobile apps, natural language querying, SSO integrations, etc) that, short of large corps hiring BI teams to invest in the open source ecosystem like superset, your open source solution won't have.

Yeah, unless your at unicorn scale, I don't see how self-hosting BI would ever make sense relative to other investments you could make.

I must like to live dangerously. In all seriousness though there are low cost alternatives to those mega BI tools that suit many use cases. If I wasn't rolling my own I'd probably start with Metabase or Superset. What I use: https://github.com/totalhack/zillion

I’m assuming this is visually just some racks in (probably) Azure data centers and is leased not owned.

This line makes me think this might be a physically-owned system:

>> Total mass of 3,291 kg (GPUs only; not counting chassis, system, rack)

That is also what I suspect. Otherwise where would they get two thousand H100s?

Maybe an AI startup that went under and the investors took the cluster?

Emad from Stability estimates this at 4M/month. https://twitter.com/emostaque/status/1668666509298745344

Or an upfront payment of ~$25m at 10k per H100.

H100s cost $35-40k each

My bad. I Googled around for the cost and misread things. $10k was the A100 cost...

Very impressive project sir.

And Nvidia has a very good business.

Per card? DGX is 8 H100, so ~$300k per node?

Yeah, ~250-350 per node depending on node spec, volume, etc.

Did you buy the cards or lease the system?

Tax payers end up fronting the bill for all this bad behavior :-/.

Good. Thank you. I won’t look a gift horse in the mouth here.

You, Daniel Stenberg, and the curl project (and I) seem to agree that's the appropriate reaction here.

Some other bystanders appear to disagree.

Really impressed with this. OpenAI has its foot on the gas!

There are tons of data brokers that get near real time user level location data from mobile apps (usually not from ‘name brand’ apps but from the long tail) and then sell this as aggregated data products to others: eg https://docs.safegraph.com/docs/monthly-patterns .

This definitely exists in the US as well. You can see a variety of caste discrimination lawsuits against tech companies.

Secondary sales and secondary sales rights generally impoverish the common engineer versus founders and exec management. It is extremely rare to find a scale-up that has had secondaries that put non-founder non-execs on equal footing. So sadly it is just business as usual in our industry.