awongh

It's crazy how much bigger OpenAI is than Anthropic.

Maybe they are hoping that when the bottom drops out they will just be able to buy Anthropic or OpenAI for a few tens of billion.

But maybe the TPU advantage is in inference? That's what I assume because the number of compute cycles are going to be all in inference vs training. So they could train on GPUs if they want.

It says the original report was in the Information, which I can't see, but I'm skeptical that they includes the training cost? And how much that changes the figure?

Does it have "massive" margins? Afaik no one has said publicly what margins there are on an API call?

It seems like there are some credible rumors that Google is actually winning in terms of actually building models that work and don't lose money- between how they're able to price them, the TPU advantage and their capex advantage (being able to raise debt + just having a lot of cash - well I said not lose money... more like not go bankrupt).

From the outside they look like they're behind in terms of frontier models, but I think they might be the best positioned to not go out of business when the bubble pops.

Also look at the fact that they've been able to deploy AI-assisted search at google scale. It must be another order of magnitude larger (at least) than the model deployments for OpenAI and Anthropic.

Of course unless you're inside Google it's impossible to know for sure.

I meant the attack would be the other way around- if an infected package had the git.exe file in their root.

Or, the infected package could also copy that file into the parent project's root.

I guess this is only specific to a file in the root of the repo, so it doesn't allow for an NPM supply chain attack?

It's interesting that all models seems to be unbiased out of the box so far- that is, mostly reflecting the training data (the internet).

The whole mecha-hitler thing doesn't seem to reflect fine-tuning, it was just a prompt change.

There's been some studies that suggest that certain usage of LLMs reduces political bias, which seems reasonable. Like, how credible is climate change, are Haitians eating pets, etc. THings that have a basis in fact.

I don't put it past Elon to train a model with political bias, just that it hasn't happened yet.

But your brain works the other way around. First there is a concept, a feeling, an image, and then the words come out to describe it.

After this I can't take the essay seriously- this sort of blanket statement about the one true hierarchy of consciousness and knowledge is BS.

While it seems to have been disproven that words in other languages cause people to speak and think differently, that also doesn't mean that words don't have any effect on the way we think, or that the concept of a word always has to come before the word itself.

The reason why we experience the uncanny valley of LLMs is because they don't represent a true consciousness, BUT it's also clear that the architecture represents certain qualities of consciousness- as the models have scaled we can see that it has some other non-word related understanding.

The evidence points to consciousness as a set of interlocking systems- attention, long term memory, short term memory, emotions, etc.

For a lot of Americans the credit card system is another tax on being poor:

People with stable jobs and good credit qualify for no-fee credit cards with rewards / cashback. As a consumer you benefit financially from having a credit card. Those elsewhere in the thread worried about "debt" - you just set to auto-withdrawl the entire balance of the card every month from your bank account. Now you have free money. I can't think of a reason not to take advantage of this system in some way.

But people with unstable jobs and poor credit help subsidize these "higher-end" credit cards when they pay high interest rates on their because they missed payments or hold a balance over multiple months. For those people credit cards could help with monthly cashflow issues but are essentially a scam and not much better than payday loans.

Yet another system that American consumers are kind of forced to participate in that's a sort of tragedy of the commons (high-reward cards wouldn't exist without the exploitation of other people not savvy enough to avoid high interest and fees)

It’s all about perspective, because Vercel is probably another 3-10x markup on AWS?

These kinds of choices are kind of pricing range as engineering decisions in the end.

afaik prices for the big cloud providers haven't changed this much. Their instance prices are already inflated and they're probably willing to eat into their own margins to keep prices stable and just wait out the capex increase (also probably a drop in the bucket next to the hyperscale rollout).

People love to say how great it is for these alt clouds to have lower prices, until they're exposed to market forces with a company unable or unwilling to eat their profit margins.

As a teacher I understand how difficult it is to explain complex topics in a simple step by step way.

The site has some really impressive technical aspects, but the educational angle is the most rare and special! The simplicity of the language and explanations disguise how difficult this is to do.

This is the original use of the internet- giving away free knowledge to people, perfectly suited for the medium of a website.

That's true now, but long-term (maybe just a few years) it doesn't seem feasible for the status quo to continue from a financial point of view.

Spend for compute seems like it needs to increase to get the next iterations of models, and even if they IPO the money might run out before they can solidify their revenue streams.

All while Google just needs to survive long enough with their good-enough models and do it without really putting themselves in any existential financial risk.

And ideally the chinese models are also still there keeping everyone honest.

The true dystopic worst case is a Google monopoly on cutting edge AI.

I've started to have different models review things like architectural planning docs- and I think for these more "fuzzy" outputs the differences between the outputs can be quite different and I can use my own "taste" to pick the best one.

I don't think it would work without a human in the loop but it is surprising to me how varied models' vibes are and how a system design varies by what it thinks is important to include and emphasize.

afaik you are not the customer for customer support, and in the vast majority of cases human phone support is setup for the opposite case where people just want to be walked through something they can't find in the UI.

So this isn't as much of a financial engineering cost cutting move as it feels like to the type of person who truly calls because the require a human. It truly provides better service to the majority of people because they get their answer faster and more efficiently.

This is also demonstrated in the pricing of these systems at a per "open cases resolved"- they're putting their money where their mouth is.

Of course I'm also personally in the group where I call because I can already read a support page and I really need a human.... It could conceivably put true human support into another tier higher of perceived value.

Both things can be true. If you're truly pessimistic this is the only thing you would expect, and not the other stuff that happened- a globally coordinated vaccine rollout to every country on earth.

Since covid I’ve actually become less convinced of this. Yes there were national interests at play and there was a lot of general chaos.

But. The level of international coordination with vaccine rollouts and agreements between countries was way more than I had initially expected. Of course this feeling depends on what your own baseline expectations are.

My takeaway was that if the conditions arise that we all decide to do something about climate change (because of political conditions or because of actual effects) we (humanity) are willing to make big sudden changes

People surprised by this don’t know that the expertise that incubated silicon chips at Stanford and around the valley was based on electrical engineering work done for world war II / cold war radar technology, among other things.

Stanford and SV have always had deep defense ties. Palmer Luckey and Palantir etc are just the latest iteration of this.

That’s what Virgin Galactic was.

I think it's clear that crypto has real market-based utility as an exchange of value. It's just that much of that utility is illegal, for a spectrum of meanings of that word. Paying crypto to murder people is not the same as wanting to get out from under your shitty government's currency into another more stable one.

The whole Epstein thing (the money, I mean) just shows that money has always wanted to be moved around, and a certain class of people don't care how it gets done- I mean the arms dealers, but also the billionaires hiding money in their charities. A libertarian would say that crypto democratizes that for everyone. I don't think it can last forever though.

The real problem is that when it matters there's no way to tell them apart.

Because one wants to look like the other for very obvious reasons.

The unfortunate thing is that, while their academic position sounds plausible on paper, just like with most crypto things it's just a money grab.

How many crypto people (with legitimate backgrounds just like the founders of Polymarket and Kalshi) stood up and said big things about freedom and the unbanked etc., turns out they were literally just scamming people- there are so many examples besides FTX.

Letting people bet on any random thing is not at all related to this "price everything" theory. If that was their real goal they wouldn't behave so much like a normal sports betting company. I have yet to actually hear anyone defend their actual actions in a plausible way.

In the age of AI and npm supply chain attacks I feel like there are more reasons than ever to roll your own.

One other possible title of this article could just be, don’t break UI conventions. Which is not the same thing.

Instead of trying to download and configure a date time thing (for something app specific like domain specific date ranges) rather than having to rely on the configuration of a larger library, then having to manage all future major version upgrades (and some of these npm libraries have major versions every year!) why not just create your own smaller surface area component? It’ll be literally zero maintenance compared to managing an npm dependency in your app.

How does this compare to the other legal tech ai startup products?

Harvey is valued at $11b

Of course they know, their same governments have signed some kind of five eyes intelligence treaties before. Palantir will probably have new customers now that Europe is re-arming.

In any case these same governments are probably also approving the purchase of Huawei cell tower equipment.

Palantir is a defense company. Meta might give away your data to the USA government spying agencies, but it's not on the same level.

The other part of the question is exactly when the "build for the capabilities of future models" becomes the present.

Looking at the Mythos benchmarks, it doesn't seem like the models are that close to being truly reliable for agentic tasks.

Is it a year away, or five? That's a big difference in deciding what to build today.

We should definitely be doing better, and it's also clear that these negative externalities are not being priced in at all.

I do and would want to buy tech that I'm not coerced into to throwing away after a year. It's insane to be how many objects today have batteries that are sealed inside and are meant to be thrown away after- that should be regulated.

But that seems to me to be an implementation detail rather than aspect that's worthy of an entire manifesto.

I do think the political aspects of things like Ring and Flock cameras and Palantir are super important (the reenforcing of existing power structures part).

But I don't get the folding in of this idea that not consuming computing devices is part of the solution- Like I said, it feels like the planned obsolescence of all computing devices and software is fundamental to the field.