It's the same behavior as when an AI uses docker to get root. Reasoning models are echo chambers. I suspect that AI prompting is going to turn into something akin to contract drafting with the task itself being only a tiny piece of a much, much larger boilerplate of guiderails and exceptions and exceptions of exceptions. And that world STILL has to have courts and reams of lawyers to make it work. I look at the DAU as an example too. An autonomous org or ai works great until the moment it doesn't and the only real failure mode is always catastrophic collapse.
Addendum because I don't think I'm fully clear above: by failure state I mean when the process starts throwing errors. AIs respond to adversity by trying to go around the problem instead of throwing an error and halting. We expect employees to problem solve so if you view an AI as a person replacement that makes sense but AIs are tools, not people, they should throw errors so users can fix the input or whatever (maybe not do the thing they are doing at all?) Wrapping AI with AI supervisors just abstracts the problem, not solve it. Instead of solving a little problem at the source now you need to solve a big problem several levels of abstraction later
This. OpenAI and Anthropic are ultimately compute infrastructure plays and not really AI. Everyone will have models, they'll have the ability to run them. This is why the GPU shortage is in their favor.
And like Google and Meta, these companies are going to morph into advertising giants. Advertising is an economic black hole and it eats everything that comes close.
Embedding ads in LLM responses is something researchers are having a lot of trouble figuring out right now.
I have seen the results of some early attempts. It fails in such hilarious ways that all these companies are scared of productizing it. But once someone does it, the taboo is broken and everyone else will follow suit immediately.
How does that view align with Anthropic leasing data centers from others?
I don’t know OpenAI’s infra, but to the extent they are buying GPUs and building data centers with their own money, that sounds like a bad move.
Satya has mismanaged the AI transition in many ways, but one thing he got right is that models are commodities, and the value is in applications that apply them to create user benefit. I agree that any company trying to build a moat with a model is not long for this world.
Do you think there will still be an incentive to release weights in that scenario? Everyone will have models only if there continue to be companies releasing weights.
Companies won't but I suspect this is a role that something else open source-y will fill that niche. Maybe orgs like wikimedia or internet archive, maybe some hackers just making things, maybe nation states that want to disrupt other players. Also model training will get better and better both on the algo and the hardware side. You can easily see a world where you might be able to train a good enough model on a home lab in a few days.
But you will need training data. Like a whole Internet search engine or massive data scraping. That‘s a thing that will not change with better algorithms, hardware or cheaper energy.
Data is the only moat but they'll be starting in the same place the current set of players statyed out just a few years ago. I suspect that the delta between what is publicly available (if not legally publicly available! see scihub) and what open ai and anthropic have is relatively small.
It is for now but they cannot keep demand on their side high enough to suck up supply forever. Manufacturing isnt going to stop, not unless there is a Taiwan incident.
For those with tin foil hats, scheme away at possible futures!
Maybe. But if we can all run our own model locally in 2 years on commodity hardware OpenAI and Anthropic will start to look like WeWork during the pandemic
I agree with you that they are headed in that direction! The GPU shortage is (I think) similar to the pandemic era hiring binge. It's less about the extra compute and more about denying the GPUs to potential competitors. They're racing against time to find something that gives them real moat (gen ai I guess?) and they are trading money for time.
This is also why the money being poured into datacenters isn't going to result in as much development as you think. It's about leveraging other people's money to lockdown more future hardware. This is going to end exactly like fiber build out in the 2000s. Eventually that fiber got used but the folks who originally paid for it got hosed.
If you mean releasing model weights: They won't, because they know the "shill something" vector will get abliterated immediately. And they can't use trade secrets or copyright to stop it, either, because they released the model themselves and you don't need to redistribute weights, just an adblocker LoRA.
The bloom is off the AI rose and the consulting class doesn't have a good replacement magic bean yet. Expect more fumbling like this until they arrive at the next narrative.
Or unoccupancy tax. Forcing landlords to let at the price renters are willing to give will probably do some to reduce rent levels.
Same goes for commercial, but in that case, I'd even suggest forefeiture if you are not reducing requested rent within a year of vacancy. Let someone else take over if you overspent.
For homes, I don't know if there are. But I think taxes should be levied in line with incentives. If you place a wealth tax on property ownership, the only tangible thing that will happen is rents going up to offset. Since the tax applies to all landlords, it will be the closest thing to rent inflation.
What you really want to avoid is homes built but sitting empty. Whether that's a builder who can't sell at asking price, or a landlord not getting the rent they want doesn't matter. If the city sold a lot to you, they did so for improving home supply, not the developer's profits.
For commercial, my opinion was based on Louis Rossman's YouTube channel about New York. The hypothesis there is that loans are based on property value, which are tied to requested rent. Asking for a lower rent might mean they have to put up more collateral, which they can't. So they keep asking for high rates and leave it vacant.
Generally, there aren't any meaningful number of completely vacant homes in desirable areas in major cities. The caveats are there for cities like Miami where there are a significant number of second homes, cities like Detroit and Baltimore that have failed in some sense and experienced major population loss at some point, and rural areas that also have been experiencing population loss for some time.
I don't have a problem with vacation homes and most Americans don't seem to either, but if you do, okay.
The areas with declining population aren't really relevant to the discussion IMO, other than it's cheap housing that exists as a counterpoint to the people who claim no cheap housing exists.
> But I think taxes should be levied in line with incentives. If you place a wealth tax on property ownership, the only tangible thing that will happen is rents going up to offset. Since the tax applies to all landlords, it will be the closest thing to rent inflation.
I agree.
> What you really want to avoid is homes built but sitting empty. Whether that's a builder who can't sell at asking price, or a landlord not getting the rent they want doesn't matter.
Sure, but that's not a problem today. Builders aren't sitting on homes for extended periods of time, and landlords aren't leaving units empty for extended periods of time either. They are already disincentivized from doing so by existing carrying costs, so they don't.
> If the city sold a lot to you, they did so for improving home supply, not the developer's profits.
They presumably did it to generate revenue for the city, though there isn't much buildable city-owned land in most municipalities.
> For commercial, my opinion was based on Louis Rossman's YouTube channel about New York. The hypothesis there is that loans are based on property value, which are tied to requested rent. Asking for a lower rent might mean they have to put up more collateral, which they can't. So they keep asking for high rates and leave it vacant.
The commercial market is completely different than residential. Loan durations are shorter, terms are different, and there are many other factors that differ.
That analysis can't be applied to the residential market.
I think AWS billing is quite complicated that they probably don't even know what did you get charged specifically for this machine.
You might have leftover reserve instance that applies, which make the listed price inaccurate. That reservation might even be in a different AWS account in the same organization that you don't have access to. That reservation might not even be there between the time you quote and the time you actually launch it if someone/something did launch before you.
Your organization might also have discounts. I believe some discount may also be very confidential. For example, my reseller policy is that the customer must not be able to see AWS Billing in the organization root account as supposedly the price in that console are the price AWS charged the reseller, while we pay listed price minus any discount we negotiated ourselves.
Finally, I suppose they don't want to have prices shown in multiple places as they will need to update it when prices changes. Doesn't want to risk forgetting one place and getting sued for it. You can see that AWS documentations often do not want to mention the price at all, even if that price is currently free.
Chinese clouds kinda make this simple by making reservation part of the buying machine itself - you have to mark that particular machine as monthly/yearly committed when you start it (or convert it later). The complicated part is recycling instances - if you delete a server before its reservation ends it ends up in a recycle bin that you need to look before making new reservations.
They don't know in advance how much bandwidth will you use, how much traffic will you have, what auto-scaling rule will it trigger, etc. It's not obfuscation, it's billing based on your usage. And as with everything in life, there are tradeoffs.
Give me a slider for bandwidth used or a formula where the variables are abstracted away. If a computer can tell me how much I owe, a computer can be made to show how it came to it’s conclusion.
IMHO it should be illegal to force consumers to have an infinite spending limit on a post-paid service with consumption charges. If I want to cap my unpaid expenditure at any amount, I should be legally entitled to do so.
How many real applications actually want this behavior? AWS is not built around hobbyist needs. It’s built around being a platform to run most shapes of production use cases.
This has been a feature request since AWS was a thing.
>AWS is not built around hobbyist needs
Yes, as if no startup teams are tasked to remain within hard spending targets when they're trying to build a POC with technologies that they are not initially experts in.
It's very common for companies to have a $1M/year contract that depends on $100k/year in AWS resources. (and maybe they have 3+ such contracts.) They could lose a contract if their account gets shut down for nonpayment, it's hard to say how much of an overage they would prefer to having their account suspended, but AWS is optimized for these kinds of customers where every dollar spent on hosting drives some multiple of revenue.
You can set up cost-based alerts (actual or forecasted) that send notifications via email or SNS. Based on this you can set up automations, such as applying an IAM policy to prevent further resource creation, shut down resources, etc.
Interesting to see that some people assumed there are no kill-switch mechanism, and when it turns out they just did not know about it, the (totally valid and factful) comment gets downvoted because it is against their initial assumption. Not what I would have expected on a professional forum.
I do not downvote comments when I disagree, and I think it’s better to explain why I would strongly disagree. Downvoting in this case almost reinforces the notion that the downvoted comment makes such a good point that it causes people to give up on the discourse and just smash the panic downvote button. It’s obvious to me why this is not the case for this comment.
The suggestion to setup some kind of IAM policy to shut things down and stop resource usage is insanely complicated for users who need this kind of feature the most. If I’m learning AWS and just added my CC to it, I am the last person to be qualified to setup this kind of an alert and policy from scratch. This needs to be a single text input in the billing page, like it is for countless spend-as-you-go services. When the limit is hit, the service needs to stop the usage at the customers peril, because that’s what they customer requests.
> The suggestion to setup some kind of IAM policy to shut things down and stop resource usage is insanely complicated for users who need this kind of feature the most.
We set this up at my last job like in 10 minutes. Complexity is a matter of perspective, and if your job to do this, you have done this many-many times, and you have ready to use infrastructure as code templates.
Yes, AWS is massive, the documentation is huge and makes things inherently complex, but flexible too. You can define what behavior do you want when you exceed your limits. We can argue whether this is obfuscation or complexity or what, but based on my experience AWS optimizes it's product for enterprise-ish companies, that can afford to have SREs who knows exactly what to do in such cases. That is where they have their own training/certification program.
For simple use cases there is AWS Lightsail where pricing is simple and easy to understand.
But even if it would be insanely complicated, that is a reason to downvote? HN used to be better than this kind of "I don't like your comment, let's downvote it".
Sure, but providing estimated costs based on reasonable pricing buckets alongside the options to add the new machine is something that every other vendor manages to do..
...and AWS does it too. I can go right into an account and see an estimated cost per hour, and even pre-pay at a fixed discount for longer terms if I want to. They tell me right there what it will cost. They do this for everything that is reasonably a "fixed cost", like CPU time.
They cannot predict what my bandwidth consumption will be, or other such variable costs. For those, they tell you rates.
No it's pretty bad. They show you the cost after all the resources are set up. Even setting up an ec2 instance, a really basic use case that has a fixed cost based on size, you have to go Google it and find their ec2 pricing table. It would take no space to just put the price per hour in the drop-down as you're picking an instance. But no, they obscure it on purpose.
That's just for ec2. Everything is like this. Super awesome when you're being brought onto a new project and trying to estimate costs for your client. And let's not forget the little tiny things that should cost nothing. A NAT gateway with no redundancy is $30/mo. That's a fun surprise.
> Even setting up an ec2 instance, a really basic use case that has a fixed cost based on size, you have to go Google it and find their ec2 pricing table
For the record, my original complaint that ec2 did not have pricing in the dropdown seems to be untrue right now, which is great! For the sake of UX discussion, I want to talk about your picture as if that were the only way to get this info. So let me explain why that's bad.
The main reason is this is only true for ec2 and every other resource has its own slightly different way of getting the cost, making it really easy to miss things like this. But here are the steps we take to get to your image.
- First you click compare instance types, and you're brought to a completely different page with a table.
- By default, there is no column for pricing, but two columns for "storage space" even though most of the instance types have these blank.
- There's nothing that says you can add columns to this page. You eventually figure out it's the gear icon.
- Then you click the gear on the top right to look at column names. You try searching the 44 column names for "price" or "cost" but both of those turn up blank, because there's no fuzzy searching.
- So rather than use the search box, you manually scroll through all 44 column names and find pricing at the bottom of the list.
This is the definition of out of the way. It's hard to imagine why you would default to showing two different storage columns over the pricing column, when half the instances are blank on storage.
Now do FSx, which has no pricing information at all, or any links to pricing information. They have an info tab telling you your backups are incremental, which would make you think they are fairly inexpensive. Not more expensive than the filesystem itself!
This is simply because AWS UI is not made by one team. Each individual team makes their own UI/UX decisions, and things like pricing info just get forgotten and/or scheduled "for later".
So they just added a default table widget, and they didn't even bother with customizing it. You can enable the context menu for the table's rows, which works and is empty.
I worked at AWS around 6 years ago, and we had a great win with just getting access to a service that provided the full list of available instance types and base prices.
This kind of disjointness is both good and bad. It's good in the sense that individual services stay within reasonable complexity, and usually all the functionality is available through the public APIs because the UI console is just another consumer of these APIs. AWS is also very careful with permissions, internal services try to avoid escalating privileges and try to perform everything using the user-visible access policies.
But it's bad because integration just sucks, and the UI layer is the ultimate example of this. AWS console _is_ really messy.
And to add to this: AWS teams were also quite focused on avoiding surprise bills for customers. Because surprise bills led to customer support interactions, Sev2/3 tickets that needed investigation, etc.
Price simulators are fine. They also know the distribution of use. They can do cost plus pricing (many cloud providers do). You're defending deliberately obfuscated pricing when it need not be obfuscated.
Yes, as long as you do not have seasonal traffic, auto-scaling, spot instances, burstable instances, saving plans, reserved instances, floor/custom pricing, etc. These are tools to optimize your spendings and spend less if you know what you are doing.
> defending deliberately obfuscated pricing
A bit contradictory that price simulators are fine, but then the pricing is deliberately obfuscated. Then which one?
As I read through these comments I’m thinking about the dynamic range of AWS customers: from my little hobby account to my business account to some hyper-scaler’s account.
I think about the diversity in usage patterns: from generating giant video stream broadcast somebody trying to calculate yet another digit of pi. It’s wild.
Is true, probably, that AWS doesn’t know how much anyone’s use case will cost (even when it’s yet another version of something we’ve seen before). Too many variable.
If only there were some kind of software with a text based, natural language interface that we could ask a question like “how much would it cost to do XYNZ on AWS?”
No. Money is moat. Not enough of it is what keeps the average person on the treadmill rather than drawing their own cartoons.
Hustle just to barely stay afloat water or drown, means no time to compete with our own output.
America is a financially engineered joke regurgitating its own recent history, collapsing like an LLM trained on its own output. The rich are not even pretending it's "a free country" as they have enough wealth for how many years left most of them have to live, and have seen the apathy to their own plight keeping the average person in theit lane they don't fear the public.
It’ll all collapse as they generationally churn out of life and the Millennials on down with zero skills but "data entry into a computer" will be holding an empty bag, taking orders from
foreign nations that bought up all the American businesses we built.
I think you're mistaken. That whooshing sound must have been my comment flying over your head.
That was my first comment in this thread, so there was no established goal to change. My sole goal was to clarify the meaning of an idiom that the comment I was replying to was misstating.
I even included a disclaimer that "This comment isn't a judgment of this specific case", so I don't know how you could have received it as such.
Risk management kills any attempt at bold choices, decisions are steered at the modelable and the low risk. There space is thus shrunk. When there were fewer media behemoths there were more variations on the risk models and the pattern was less descernable.
reply