Again if you disagree, you'd better be prepared to produce birth certificates of all your ancestors to prove you're a "natural born citizen" born of citizens. That's where this leads.
Eventually, if they’re old stock European-American, that essentially means that they’re the descendants of invaders (i.e. illegal immigrants who never got the consent of the Native Americans to settle).
> They're in the "get 'em hooked" stage of the drug deal.
You're implying that people are selling inference at below cost right now. That's certainly not true for most third-party inference providers. I doubt API pricing at Anthropic or OpenAI is being solid below cost either.
The only place where you get what you're talking about are the fixed price plans OpenAI, Anthropic, Cursor, etc. sell.
OpenAI is claiming they'll post 74 billion in loses through 2028. Anthropic is on course to lose 3 billion by the end of this year, they lost 5 billion last year.
As far as I can tell the inference provider landscape is a fucking mess, and I can't find any decent financial information on any of the ones I tried. So unless you have something showing those companies are profitable I'm not buy it.
Even if you eliminate all of OpenAIs other costs besides inference they're still in the red. And they can't just stop training new models. That's like saying Honda can just quit designing new cars. They technically could, but it would destroy their business.
They have one method of monitization right now, and there is no clear evidence that their costs are suddenly going to decrease anytime soon. Despite claims to the contrary, no one has actually provided any evidence of a pathway to those costs magically cutting in half over the next few years.
The entire industry is being propped up by insane over investment and an obsession with growth at all costs. Investments will dry up sooner or later, and you can't grow forever.
Inference keeps getting cheaper, so "it isn't cheap enough yet" isn't an issue. Even with zero efficiency innovations from here, cost per instruction is the most deflationary commodity of all time.
So how was that ever going to be a problem?
The optimal choice for marginal costs, which will naturally drop on their own, at the beginning of a new tech cycle is to run in the red. It would be a sign of gross incompetence if they were fine tuning those costs already.
Training spend is the giant expense. And either training costs are unsustainable, and training spend will hit a pause, or it is not unsustainable and training spend will continue.
So, which is it?
Critical point: The majority of their costs are not required to serve the highest level of capability they have achieved at any given time.
That is unusual. In the sense that it is an exceptionally healthy cost control structure. Note that not even open source offers a cost advantage, for training or inference.
The stuff coming out of young republicans is beyond the pale. These kids are making vile jokes about doing a genocide / Holocaust of Democrats, minorities, their political opponents, etc.
And they’ve been consistently doing it for so long in this chat, that it’s hard to dismiss this as some crass joking. It seems like many of these people sincerely hold pro-Nazi views.
Yes, definitely. S3 has a time to first byte of 50-150ms (depending on how lucky you are). If you're serving from memory that goes to ~0, and if you're serving from disk, that goes to 0.2-1ms.
It will depend on your needs though, since some use cases won't want to trade off the scalability of S3's ability to serve arbitrary amounts of throughput.
In that case you run the proxy service load balanced to get desired throughput or run a sidecar/process in each compute instance where data is needed .
You are limited anyway by the network capacity of the instance you are fetching the data from .
These are, effectively, different use cases. You want to use (and pay for) Express One Zone in situations in which you need the same object reused from multiple instances repeatedly, while it looks like this on-disk or in-memory cache is for when you may want the same file repeatedly used from the same instance.
Is it the same instance ? Rising wave (and similar tools )are designed to run in production on a lot of distributed compute nodes for processing data , serving/streaming queries and running control panes .
Even for any single query it will likely run on multiple nodes with distributed workers gathering and processing data from storage layer, that is whole idea behind MapReduce after all.
That's an interesting concept. We could take that concept in general, and a person could write 100 essays (that they publish on whatever platform), and these 100 essays could comprise their magnum opus.
We need legislation mandating that all hardware[a] have at least one fully-functional[b] open source driver for any operating system[c]. And that any device with a microprocessor with writable memory permit custom software to be run on it.
[a] whether that's a single device like a fingerprint scanner, or a device like a phone or tablet
[b] no crippled or low-performance open source driver
[c] any OS, including Windows, Mac, Linux, BSD, or some obscure minor OS as long as such OS is readily available for free or for a reasonable price
What's the best place to submit issues, feature requests, etc?
Something I use on Obsidian is my own CSS as a theme (to override default editor styling). I see themes is a feature of Octarine. Will users be allowed to set their own CSS as a "custom theme"?
Currently the app supports 33 themes (3 free, 30 on the pro license). There's an item on the roadmap that would allow you to create your own themes (each theme is a palette of 10-12 variables - border/bg/text/outline)
How is Rust in terms of incremental compile time? Is incrementally recompiling on file in a large project quick? Hopefully link times aren't that bad.
One thing I like about the JVM is hot code reloading. Most of the time, changes inside a method/function takes effect immediately with hot code reloading.
Link times are the worst part but solveable with mold[1]/sold. Incremental compilations are usually an order of magnitude (or even two) faster than clean compiles but tbh that can still feel slow. Helped by using something like sccache[2] or even cranelift[3] when debugging. Still not as fast as having a hot-reloadable language but it gets you to a relatively pleasant speed still IME
I've never been successful in getting sccache to really speed up projects, but then again only release builds have _really_ been impossible for me, and that's only when I was working on Deno which was absolutely massive.
We had issues with long compile times at deno. Release builds were brutal, debug builds were OK as long as they were incremental. It was likely one of the largest open-source rust applications, but we were still quite productive.
Most likely you'll have years before it's an issue and there are mitigations.
Yup, that's what I do. Even for personal projects, with the flurry of changes Claude/other AI assistants make, a branch makes it easier for me to compare changes.
Often I have a branch with multiple commits on it, with each commit corresponding to a message in a conversation with AI on Cursor trying to get a new feature built.
In the end, I can diff the branch against the main branch, and see the sum total of changes the AI agent has made.
Maybe edit/improve manually on my own afterwards. And then, merge.
