Yeah I was wondering about this too. It seems like way too much per GPU considering the purchase price of a B200 was around $40k last time I checked. So if we naively ignore the price of electricity and maintanence, it would only take 5 months before renting is a worse deal than buying outright.
It’s a supply and demand thing. Google would definitely be buying from nvidia and setting up themselves if nvidia had the capacity.
SpaceX/xAi/musk are currently in a good market for “happening to own 100k cards we have nothing to do with”, and are exercising that control as hard as they can.
> A University System Went All In on A.I. Now It’s Tearing Itself Apart.
The NYT often has a bias against AI, but the article's contents are actually a pretty reasonable summary of the different attitudes towards AI in academia. Then they went and slapped a terribly sensationalist headline on it, which doesn't seem well supported by the actual article.
If the NYT is the same as British newspapers, the person who wrote the article doesn't write the headline. It happens here too, and its really annoying because often the article has a much more nuanced picture than the headline would have you believe. In an era when people do read the article after reading the headline it's somewhat forgivable - getting someone's attention then they get the nuance, but in the internet era when people just read the headline it's anachronistic.
I don't discount that it's possible that NYT headline policy could have changed in the last decade, but sensationalism when it comes to newspaper headlines is the historical norm. "Clickbait" is an ancient phenomenon:
"In A History of News, Mitchell Stephens notes sensationalism can be found in the Ancient Roman gazette Acta Diurna, where official notices and announcements were presented daily on public message boards, the perceived content of which spread with enthusiasm in illiterate societies."
Sure, if it bleeds it leads. But I've been a news consumer since the 1980s and a reader of the NYT almost as long as it's had a web site. It is my strong impression that its headlines have gotten noticeably worse. The main one that annoys me, and I don't actually see an example of it on the main page right now, is the teaser headline that forces you to click through to know what the article is even about[^1].
Edit: here's an example. Headline on the front page is "A Chaotic, Confusing Campaign: Here’s Who Should Be the Next Governor of California". Makes it sound like you're clicking through to an endorsement, right? Nope, the article is actually a voter guide. It's a completely misleading headline.
[^1]: You can often inspect the URL to see the original descriptive headline before the clickbaiters got to it which makes it even more annoying.
Blame the economics of the Internet. Companies use clickbait because it works, there have been many examples of this, and if a company wants to stay revenue generating in this day and age it must use clickbait.
Quite funny to think that we might have AI models meticulously nudging newspaper editors in order to carefully control the public's Overton Window about AI, playing some 5d chess.
Yeah that title is absurd, tho it did make me read the whole thing out of pure incredulity. The “tearing itself apart” apparently refers to the fact that the CSU system spent $16M on AI tools during a $2B+ budget deficit, which… yeah. Doesn’t take an economics professor to see the problem with that thesis!
The author does seem interested in supporting the headline, but I think they're too good of a journalist to pull off the outrage. It mainly comes through in passages like this:
After I pointed out to Janos that Marx himself would have had a field day with MarxGPT, he laughed… by interacting with ChatGPT, he and his students solidified its role in the public education ecosystem; and their ability to do so was the result of the transfer of almost $17 million of worker-generated public funds to a private, for-profit company.
If this wasn’t the NYT, I’d assume this was a joke. Sadly, I think it is indeed intended as something of a slam dunk…
They do get to AI critics eventually, though obviously ‘activists dislike X’ isn’t really proof for ‘X is tearing us apart[, Lisa!]’. Namely,
“We feel like a guinea pig for what A.I. is going to do to higher education,” Kenney said. The embrace of generative A.I., she went on, is “a step down the path of creating a really different kind of future citizen and worker.” This kind of student would be intellectually passive, less likely to see themselves as agents of their own lives.
I think everyone would agree they’re “guinea pigs”, as are we all in a way — such is the curse of living in interesting times. The rest seems pretty plainly speculative, though.
This winter, the [critics at SFSU] circulated a petition asking the chancellor’s office to invest in protecting faculty jobs and academic programs rather than renew the OpenAI contract.
Hey there, I'm going to check out your project because the comments here have me a little worried that OP's project might have some quality issues.
Two things I found a little confusing from the docs though:
I couldn't easily find a page describing what it can't do yet. I saw that it only works with a "strict, deterministic subset of TypeScript", but is there a page showing what's included and not included in that subset?
Also, what's an "ambient surface" in this context? Is that a compiler term I'm just not familiar with?
I'll add that page, thanks. Today, almost all of idiomatic TS is supported including most of its utility classes. Dynamic JS-style code is not supported, for example adding a function or a field into an object, prototype-based class modifications etc. I'll compile a list, and include it along with the large docs cleanup planned before v1.
> Also, what's an "ambient surface" in this context?
The idea is that when JS gets transpiled into C# (or Rust, upcoming), JS globals and built-ins are invalid. The native "surface" is C#, meaning the string is .Net's string type and the methods that you expect on JS strings would be missing. But when you opt in to a surface, such as the "JS surface", the compiler applies surface defined translations such as substring becoming SubString, either directly or via a companion helper class. This allows you to write against standard JS and Node APIs, instead of relying on the stdlib/builtins of the target framework (currently CLR). And you get the JS "stdlib" - console, JSON, Date, Map, Set etc.
> Dynamic JS-style code is not supported, for example adding a function or a field into an object, prototype-based class modifications etc.
FYI, just in case you didn't know, there is an ExpandoObject type in the System.Dynamic namespace that you could use to do this. IDK if you want to, but it's one of those less common .NET features that people tend to not be aware of.
tsonic uses NativeAOT by default, and DLR/System.Dynamic isn't supported in NativeAOT. There are switches in tsonic to give you MSIL, but that's used mostly for testing or for compat reasons.
This is such a refreshing policy. AI code is welcome as long as it's good, but comments have to be human.
If someone can't take the time to write their own replies (in their own words), then it feels fair to assume that they didn't take the time to test, review, and clean whatever code they submitted.
Yeah I really agree. There's no overage pricing on their website's main pricing page, which makes me think the jump from "Team" to "Pro" to negotiating an enterprise contract will really hurt. Going from $39/month to $199/moth because you needed slightly more is a really big jump in pricing. It's pretty much the opposite of what I would expect from a service that lets you scale to 0.
Props to them. That makes DeepSeek v4 Pro extremely cheap compared to others, even in the same category. Look at these prices per million outputs tokens:
It's actually even cheaper when you look at the cache read costs. Those costs can dominate in agent workflows and DeepSeek's cost for cache reads is insanely low comparatively. At $.003626/M tokens, the cheapest other thing on your list is >$.2/M tokens. That's on the scale of 100x cheaper.
Also, deepseek cache hit rates are pretty good. I use deepseek v4 flash model regularly for agentic tasks (more than 20 tool calls on average per run), and 70%+ of input tokens get served from cache.
The speed is absolutely bonkers too. I once misconfigured a mcp I was developing locally, and told it to use the tools provided by this mcp to get certain task done. It figured out that the mcp is misconfigured, and then automatically went ahead and started to fix the mcp, fixed it, and then started using it by passing raw jsonrpc messages using stdin/out, bypassing the harness integration (since it would have needed a restart).
It did all of this in under 30 seconds and made over 15 tool calls in all of this (yes, I use yolo mode in a container, so my agents have full access to everything in the container).
OpenCode Go is not DeepSeek. They may host the model but they're run by an entirely different organisation.
I imagine when onlyrealcuzzo said "they don't make the model worse once you have a subscription", he didn't mean OpenCode Go, otherwise they would have probably said so.
Right, but then whether the service degrades with a subscription or not is up to the service provider - not DeepSeek themselves. In this case, it would make more sense to say "OpenCode Go don't degrade the quality of the DeepSeek offering with the subscription", etc.
I'm aware I'm slightly nitpicking but your message more or less implied you could get a DeepSeek subscription.
Seems like a branding mistake to me. This new device doesn't target the same group of customers that the Flipper Zero did, so it will be much harder for them to market it effectively, since the first device is already famous enough that people know them for that market.
reply