They offer inference, too. I don't know anyone who uses this service, but presumably it must be reasonably popular.

This has become a common pattern in my Claude & Gemini usage. Always require citations, then check those citations to validate they actually contain the information/data the LLM's output claims. Claude, in particular, seems to make massive logic jumps and trust tertiary data sources way more than it should.

Honestly, there's just a huge amount of thrash across the cloud & SaaS industries as a whole. It's not a lot better at Google (way behind OAI & Anthropic on coding & local AI harnesses), Microsoft (no meaningful frontier model R&D to speak of, and a constantly eroding Windows business), the neoclouds (impossible to source adequate infrastructure and are existentially threatened by the hyperscalers (especially Google) if they can't keep up with demand, chipmakers (NVIDIA vs AMD vs Qualcomm vs Apple vs the long tail of specialty shops like Cerebras), data center pure plays (legislative & community pushback, natural resources, infrastructure availability), and the SaaS folks ("let's see if AI can do it").

It's both a great time to be gold mining in tech, and also a terrible time to be a bit employee.

I've been in Bay Area tech for 35+ years and this pretty much sums it up. AI has upset software economics at multiple levels, including forcing an increasing number of companies out of business. On the other hand it's hard to imagine a better time if you like new technology and have the grit required to win in very competitive markets.

Maybe, but GPU is just one aspect of NVIDIA's dominance. If you are buying Vera Rubin GPUs, you're getting an NVL72 rack, which is only one of several racks that you're probably buying. You'll also need your NVIDIA racks with NVIDIA networking & storage gear, too. At the end of the day, they're "vertically integrated" for your accelerated computing data center (e.g. the "AI Factory"). This doesn't even count the software layer, where CUDA + CUDA-X (not to mention the software for all the sysadmin pieces) has a huge first mover advantage over anyone else.

Is CUDA still a moat? Are we not at the point where frontier models can reimplement software stacks, given you throw enough tokens at the problem.

If you are making a decision to spend 50B on hardware, would you use the proven tech stack or rely on engineers taking an unspecified amount of time vibecoding your software stack while the hardware sits idle?

How about in a month or so when you have to run a slightly different workload?

Hyper scalers like Google or Microsoft, which are the big spenders, have all the incentives in the world to get more out of their gargantuan spending.

In fact both of them, actually Amazon too, invest in their own inference hardware and owns the stack.

You can't possibly think that these companies will keep shelling 50-100B per year in hardware alone where 60%+ is margin for Nvidia and not invest there.

if you are developing your own hardware, you provide your own stack to avoid lawsuits with nvidia. i don't think it's a technical problem at all, but a legal one. this is probably why zluda was scrapped by AMD and Intel. Nvidia technically bans the creation of CUDA reimplementations in their TOS if i remember correctly

Doesn't Google v Oracle provide protection here? Copying APIs is fair use.

If it's patents that are the problem then presumably all these large semiconductor companies have defensive parent portfolios.

I don't know then, amd and Intel decided to get rid of Zluda before that lawsuit ruled that apis are fair use. And now, their cloud customers are ok with using their own stack such as rocm, and Intel got rid of their Garuda ai chips. Those things also happened before the lawsuit was settled

It would be a breach of contract not a copyright issue.

You still need experts to know what is good and what is not.

i haven't seen it. see the recent browser attempts.

You just proved that AI cannot currently do that

It can definitely create a software stack for you if you hold it right, but the software stack supported by a trillion dollar company with decades of expertise, that also uses AI to improve its stack is probably gonna be better.

On the reverse side, there are fundamental limits to the number of ways you can perform certain actions, and agents are both diligent as well as able to swarm. If you have your tests beforehand, there is a chance.

How did you effectively download 500GB of photos? Did you just use Takeout?

I did, takeout worked pretty great actually.

Remember to choose the `tar` version, not the `zip` one, unless you like downloading ten thousand zip files one by one. The tar files can be up to 50GB, the zip ones are much smaller.

The zip files can be up to 50GB as well, but they recently changed something on the backend so that if you do use the zip option you may still get however many 50GB files then maybe a 1MB, a 7kB, and a 1GB file, I assume because of the way they parallelize the tasks. I don't know if the tar option behaves in the same way.

Oh, interesting - I wonder if that's new. The last time I checked (granted, a few years ago), it was limited to something much smaller.

I have noticed that the tar files toward the end tend to be smaller, but they still typically weigh in at multiple gigs apiece - 37, 14, etc.

Did the same. One issue was the links had some # of retries you can do before they become invalid. Solved it first by exporting them to google drive instead :D

I think it's arguable that this "intuition" you call knowledge could very easily also be called "opinion".

The interesting thing is that China began as an agrarian economy and quickly modernized via influence/direction of the CCP.

The United States used to have a dominant middle class that was geographically distributed (cities & rural areas inclusive), but the advent of the tech economy has also been having a similar effect here as it did in China: massive wealth accumulation in Tier 1-3 cities and everyone else being largely left behind.

It'll be interesting to see if there's convergence in the next decade or so, especially with the GOP reducing regulation and increasing the explicit capitalist priorities here. TBH, though, as a generally well-informed political outsider, the feeling I have is that the US government is slow, bulky, inefficient, balkanized and overall poorly run compared to the Chinese government. We'll likely either slowly improve authoritarian efficiency and become more like the CCP ... or we'll pivot left and move more toward the EU model, but we're floundering around right now paying lip service to both.

USA has to fall behind China much further before its people can make up their minds to pick a road forward.

As a business user, the same thing is true for non-code documents. The biggest exertion is reducing the excessive slop down to concise, clear points.

My question is this: why is Palantir required for this kind of fairly basic data analysis? Couldn't they use much cheaper, probably open platforms for data science & analytics?

I don't see where it says who's paying whom. Maybe the deal is that Palantir gets more personal data to throw in a national profile database.

FWIW, your pessimism isn't entirely well-founded. If you're in a hot, tech-centric market, then yes, staff will be far more inclined to job hop for comp increases. But if you're hiring remote or you're located outside one of the very few tech hubs, employees value quality of life and workplace culture/relationships much more highly relative to comp. E.g. If you're hiring a SWE in Nashville or Portland or Columbus they're going to behave differently than if you're hiring in the Bay Area, Seattle area, NYC, Boston or Austin.

I think those sorts of priorities come later. I've never known a junior engineer that prioritized anything much beyond "holly shit, someone finally gave me a job, and I'm making money!". Then, with something on their resume, they realize how they're currently being taken advantage of, and hop to a significant pay increase.