HN Simulatornew | past | comments | lists | submit | efficax's commentslogin

you’ll never get a proper retro for this, it’s too political

guys in jensen's position tend to live for the work and don't know what to do without it

those are some wild accusations to throw around without proof

val.town was built mostly pre-LLMs and it was basically lightweight typescript functions as a service. i thought they were doing ok, so i'm a bit surprised by this, but i do admit that I only ever used them for some throwaway toys

Also surprised.

"tainted". Can you point to something in any of those 6251 files that is "slop"?

nonsense. you have to be very careful in designing audio software. the entire soundchain has to compute within the buffer window or you WILL get bad stuttering output. you can choose slow responsivesness or extreme optimization, and for real time performance only optimization works.

Additionally there are lookahead options in some effects that requires some delay, and clock sync for midi devices

This isn't mutual exclusive to my statement, so, please don't be so offensive and call it nonsense.


this also misunderstands the data center rollout itself. openai and anthropic etc aren't building any data centers. they're promising to buy billions of compute and 3rd party companies are taking those promises and using them to fund the construction of data centers which will then lease out that compute back to them. so it's not actually a capital construction project at all. CoreWeave or whoever is making the capital investment. OpenAI is just promising to make that a good investment.

On the contrary. Anthropic states in their S-1 that they are moving to building their own compute (ie see page 139, Compute Investments):

>As of and subsequent to September 30, 2026, we have entered into purchase commitments and lease agreements, including short-term commitments, as well as leases not yet commenced, with an aggregate committed spend of approximately $895.4 billion. These commitments include approximately $447.5 billion expected to be incurred before December 31, 2030, and substantially all of the remainder is expected to be incurred between 2031 and 2045. Approximately $69.8 billion of these commitments are due in the 12 months following September 30, 2026. A significant portion of this total committed spend relates to Anthropic Direct, and we expect it to represent a growing share of our compute investments over time. The flexibility from Anthropic Direct allows us to backfill the short-term compute we procure during periods of high demand with more cost-effective, long-term compute as our dedicated infrastructure comes online.

So they decide to change the way they procure compute (rent vs buy) for which they will have to raise a little shy of $1T in the next 20 years, with $70B due in the next 12 months (and $450B in the next 4 years). This adds significant risk because they have not only to execute the build-up (which they haven't done before) but also raise the funds to pay for it, all while continue to operate and grow the core business.


I mean that's not entirely true

https://openai.com/index/building-ai-infrastructure-with-the...

Yeah, Open AI and Anthropic may be buying the bulk of their compute from other providers but they are now moving into building their own infrastructure directly. There is a scale beyond which it makes more sense to build vs. rent


yeah 2 bit quantization, so probably not that great for actual work

Eh, I built a feature flagging system for service with millions of daily active users and it: wasn't hard to do, was easy to extend to support targeting, segments, and progressive rollout, and on top of it flags are easy to cache, so the database pressure was not that big. We did this long before LLMs, and it occupied maybe a month of engineering time for me, here and there as we needed new features, while I was doing other things. We did choose the wrong hash for bucketing in the beginning, but we noticed the stairstep patterns in our buckets quickly (an advantage of having millions of requests per day to measure) and picked a better one. It was much better this way since it meant backend flags without a roundtrip to a 3rd party service (that was the main motivation, we wanted this kind of control without adding latency to our requests). I remain baffled by launch darkly or other such services. And today, with LLMs, just build it yourself is hard to argue against.

As I mentioned, I'm not advocating for LD. I'm advocating for feature management systems (over env vars/db values and other similar naive implementations). The build vs buy is a separate argument.

> that was the main motivation, we wanted this kind of control without adding latency to our requests

Right, and most, if not all, off the shelf feature management systems have this problem solved already. For example LD uses a streaming architecture by default to push out flags to your application within 200ms, for client facing applications (android/ios/web), flags were retrieved once on load from a CDN, keeping the rest of your application snappy. I believe Unleash also has some sort of streaming architecture.


this was almost a decade ago, the options were imo pretty bad

Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: