Claude Fable 5 and the Future of SketchUp Modeling (+ general discussion about AI)

SketchUp
21

If you have got to that stage, there won’t be an approvals process… If it’s designed with AI, then why would it need analysed by another AI, that’s not efficient? In that scenario building would be managed by a single AI which operates as designer and approver, and I’d take bets that it would be some corporate entity that is in control rather than a democratically elected human committee!

“Please submit your house requirements for plot 666, together with your payment and our AI will forward an approved proposal shortly. No further negotiations on alternative designs will be considered”

I do think the current AI trajectory poses a significant existential threat to Homosapiens, IF we let it!

22

If AI is going to take all the jobs and therefore people have no money to buy goods and services, then who’s going to buy the goods and services?

Other robots?

23

well that’s how the oligarchs build their technocratic serfdom.

24

I guess ya’ll failing a little bit to realize how much of hype is in this thing and how much of actual value it brings.
Claude Fable costs twice the API tokens of Claude Opus which is already insanely expensive, and is still failing to get decent results on trivial model like the ones in the example and the ones that @JustinTSE showcased in his video yesterday..

When this whole bull$h1t will stop getting subsidized (because even on the more expensive subrscriptions you are still paying a fraction of the actual cost) people will realize that this stuff is crazy expensive and will be only used at corporate level and mainly for code (and even there is starting to expose its weakness).

But of course the hype-generating machine will go on until Open AI and Anthropic will go IPO.
Then we will see the actual cost and probably most people will backlash that. :smirking_face:

25

Yeah sure.. for free.. they’ll only need to pay a gazilion API tokens to get a messy knockoff :joy:

For reference, I spent 25$ of Claude Sonnet API tokens in order to fix a bunch of bugs on a couple of my plugins.. and they are just maybe 0,01% the complexity of a software like Sketchup.

26

Lets face it no one has a crystal ball but the trajectory has been nothing short of amazing. From my own perspective I’ve realized that if I don’t use AI to help me I’m too slow and my competition will outpace me. Every other company is realizing that as well. The AI race is not just the frontier engines competing for supremacy but everyone else is competing to adopt AI and monopolize on its efficiency so they don’t get left in the dust.

Where will we be in five years or even two? I really have no idea but I’m already seeing non-programmers threatening to knock me off my precarious perch. The only thing I can do is leverage AI myself and try to innovate even faster.

27

I think it’s more about the speed of improvement than what the model can do today. If it can generate a 3D model from a very complex floor plan and only be off in a couple small areas, at the rate of improvement we’ve been seeing, I suspect that will not be a problem much longer. This tech is only getting better at what it does.

And btw - I’m not trying to be a fanboy on this stuff - I think we’re being VERY irresponsible with the way these models are trained and rolled out. The security implications and the risk to humanity as a whole (in my opinion) are very real, and aren’t being taken seriously enough.

However, it’s here, and pretending it isn’t wouldn’t do me or anyone else any good.

Crazy expensive? I was using it to create extensions on the $20/month plan. Granted I was doing enough that I jumped to the $100/month plan, but if you compare that to the cost of hiring a developer to work on code, it’s WAY cheaper. That’s $1,200 / year vs paying a developer anywhere from $50k-$150k or more. It gives me a way to create tools that I want in a way that I couldn’t before as well.

I’ve come to that same realization. It’s here and it’s not going away.

I’ve spent a lot of time wondering about this question. Still don’t have an answer, but wherever we are, it looks a LOT different than today

28

I’m not sure about that.. I myself am not a programmer (even if I studied Computer Science at college, but never graduated, because I started to work and left).
I did 5 plugins for myself, basically porting of quad-modeling related functions pretty commons in software such as 3dsMax and a couple of utilityes.

I have basic code understanding, but I can’t code at any meaningful level. I basically don’t even know ruby at all.
But still I managed to do something useful.

That said, I’m quite advanced in quad/subdivision modeling with more than 15 years of experience doing it every day, I work a lot with programmers for VR and web 3d applications, I’ve done tons of beta testing.. basically i was in the beta for most of quad-modeling related Sketchup extension.

I follow the rules quite strictly (analisys/design/implementation/testing), I care about namespaces, avoid global variables, I read the best practices about security and stability on the API documentation, I run RuboCop after any AI session, I use harnesses, I use diffs, I define strict markdown specs, I keep up to date all rules, skills, error logs, architectural documentations, changelog and such, I use modular structure, I care about recursion complexity, I add tons of guards and this kind of stuff..

And still.. there are a lot of bugs and I constantly need to go back at fixing and make sure that AI doesnt fix a bug introducing 3 or 4 new bugs..

I saw a few plugins from other people who literally have no clue about technical modeling or general computer science.. they are doing a lot of messy spaghetti code full of bugs, namespaces violation, javascripts vulnerability, O(n) recursion for massive data sets (or even worse) and all sort of amazing stuff..

I mean.. I don’t see regular people stealing the job of real developers anytime soon.. And with the raising cost will make it even less likely..

On the other hand real developers like you can of coursetake advantage of this tool (because it’s just a tool there’s no real “intelligence” in it).
At least until it’s heavily subsidized.. and I fear this is not going to last long.. that’s the real problem.

29

I did five extension made out of about 15 ruby module for each one using only free plans.

But you completely missed my point.

The subscriptions are heavily subsidized, even the Max plans.
And that is not going to last forever (they are actually gradually reducing quota on all plans and the new moels are burning more quota).
Currently the Anthropic and open AI spend 1,5 dollars for each dollar they earn (and they earn from enterprise API and lose A LOT on subscriptions).
It’s not my opinion. Is a fact.

Your 20$ Claude Pro is not 20$ of API tokens.
Try to buy 20$ of API tokens and try to start building a serious plugin using that.. it will last less than one hour of coding using Sonnet 4.6.

Now, consider that the API cost per token of Opus is roughly triple the cost of Sonnet.
And the API cost per token of Fable is double the cost of Opus.
And on top of that, the more powerful the model, the faster it burns tokens.. so the cost is basically exponential.

That’s how it works.
You can do some reserch and verify it.

30

Claude Code is included with the Claude Pro and Max memberships - I do all the coding work that I’m doing directly inside of Claude - I wouldn’t go through the API, so if I ever hit a limit, which is rare (I don’t think I’ve hit it once inside the Max plan, though I did hit it a lot on the Pro plan), I don’t get charged more, I just have to wait until my usage resets. I never code outside either the Claude Code Desktop app or the website app.

I never pay more than whatever my monthly membership cost is - I don’t get charged more for more usage. Now if I exceeded that amount and had to buy additional tokens I suppose that might be a thing, but it’s unlikely I’ll have to deal with that.

If the point is simply that the subscription gives me much more usage than I could purchase with the same amount of money through the API, I agree. It is a very good deal for my use case. If we’re considering this being “subsidized,” and you’re saying this isn’t sustainable for Anthropic, I suppose time will tell - as long as they’re offering it, and OpenAI is offering a similar product, I doubt they’ll slash usage rates too much as this would lead to a lot of people moving to competitors.

As for quota being gradually reduced across all plans, do you have a source for that? At least as of May, Anthropic was publicly announcing increased usage limits and additional compute capacity.

31

We are in the free trial / subsidized trial FAFO phase of this experiment. There is no way this will last - as soon as the real costs of this outstrip the imaginary money shell game that is happening the most advanced models will be under lock and key available to the highest bidder or to governments that force access / regulate / negotiate under threat, etc.

There is no way (if you believe the hype) that the promises of this will trickle down to common folk. We are all helping build this for the people who control the black box.

Even if the hype is 25% real that 25% will be reserved for those who can pay or those who can brute strength to just take / force it.

32

The goal is for money to trade hands between a small number of companies while paying off a smaller number of politicians. The rest of us will fight each other because we’re told to believe that the guy on the corner asking for food is stealing from us because he’s not paying his fair share, he is a different color, or he is from a different culture. While we are distracted fighting each other the wealth transfer will continue until something breaks. The will of the small folks or security of the wealthy and in power will break, and as they say where @ateliernab lives ‘have a look at how we treat kings’. But by then it may be too little too late.

This isn’t being built for ‘us’. We are helping build and train the next tools of our oppression under the guise of the democratization of this new power.

Dystopian view? Yes. I am not hopeful.

33

CC is basically free. You can install VScode and get the free plugin. Then you have to fuel it with tokens (API or subscription).

Of course is an excellent deal for you, and even a better one for me, being that I basically don’t pay. The problem in this business model is that is not a good deal for them.
That’s why this will not last forever.

Yep, true. In fact they are in big troubles and they already dropped Sora in order to focus on the most profitable enterprise API business.

It’s not me or time to tell that.. it’s public data.
Here’s a couple of links (that I just quickly googled for) which explain the thing, but you can literally find tons of sources if you search.. I followed this kind of things for months.

Intersting summary

Detailed article

Go to the Claude Reddit and you will find tons of info and (measured) sources about that. Of course they increased the limits after that they dramatically reduced them.. I guess it was in April.. then they rased limits again a bit.

To be fair, they literally don’t give a sh*t about “people”.
And also the competitors are raising prices vs quota accordingly.

Just in the last couple of weeks GPT codex lowered the quota again.
And you can also search for the GitHub Copilot (which used to be the best deal for 10$) that jumped to 40$ reducing quota and restricting the most powerful models.
Also.. Anthropic removed Opus from Pro Plan..
Not to mention Google Antigravity which was super-generous and now is a joke even on Ultra Plan.
And Gemini CLI which was basically unlimited free is going to be removed althogeter on 18 june.
The whole subsidized business model is collapsing.

They are doing kinda “the drug dealer trick” with regular users..
And it will go even worse when they will go IPO.
They are actually trying to retain individual users just another bit until they go public (because you know, they can say “look at the trend, people want this stuff, let’s invest in our companies”).

The point is: when they go public, they will HAVE to raise money.. and the only reasonable way is to sell enterprise API tokes to the big corporations who can afford it.
They cannot afford to lose money to me and you for a long time.

34

Big tech is investing an insane amount of money on AI R&D, massive data centers, etc. These are for-profit businesses, not charities; they will eventually need to show a decent return on all that spending else they will have failed their corporate missions regardless of how well the stuff works. Waiting for the other shoe to drop/the bubble to burst.

35

Well I think there are 2 things at play here - one is what you mentioned - they are going to have to show a profit, especially after IPO, which will definitely affect their business models (I don’t fully agree with the other posters on exactly how this looks, but I think we can all agree it’s going to get more expensive, not less).

The other thing that I find much more concerning is the race to AGI. Every single one of these companies is trying to race to being the first company that can create an AI model that actually thinks for itself - they’re betting that once they have that, they can do whatever they want - they think it’s a winner take all kind of thing.

I’m VERY concerned, especially after all these IPOs, that safety is going to take even more of a backseat than it already has. It’s odd to me how these companies seem to be in a massive hurry to create something artificial that is smarter than us, and has already been proven to not be 100% under our control. That feels like a really bad idea to me

36

100% agree. I do think though that they’re going to try to avoid losing a lot of paying customers (VC money may be a source of a lot of their money, but they can’t just cut out regular subscribers that aren’t enterprise entirely), so I don’t think the reductions to what they offer will be THAT drastic. I’ve been wrong before though

I do agree that after IPO they’re going to raise prices, and make it less accessible than it currently is. It’s either that or they’d have to keep building data centers at the current rate, which I at least HOPE is unsustainable and it slows down.

37

The AGi thing is even more monumental bullsh*t.
They WANT exactly that. They want you to be concerned.
This is because it generates even more hype in investors’ perceptions.

Anyone with even a passing knowledge of computer science knows that we’re light years away from AGi, and if we ever get there, we won’t get there through LLMs, which are just stochastic models that “improve” only by dramatically increasing the data set and the computing power used to improve their statistical behavior… which has NOTHING to do with “reasoning” or “efficiency” or “intelligence”.

Usually, when customers costs you twice what they pay you, do you want to keep them?
I don’t.

And you have to consider that chips, electricity, water, and space, which are consumed exponentially, are not infinite resources.

38

You do when you’ve already sunk the cost into the datacenters and need the revenue to actually pay for them.

A huge percentage of the costs for these companies has already been incurred by building the datacenters. That’s why they’re operating at a loss - to get as much infrastructure in the ground as fast as possible.

Maybe, maybe not.

Whatever you want to call it, we are building these AI models, and giving them the ability to interact with the world, without making sure that we have the ability to keep them aligned and under our control.

Whether they are “statistical models” or something else, we continue to make them smarter, faster, and more powerful, then we’re plugging them into things like cars, weapons, using them for biological research, and other things that actually interact with things in the real world.

Even if they’re not “AGI,” it still seems like safety should be a much higher priority than it is.

39

They won’t take safety really serious until we’ve had a significant incident. I don’t know what that incident will be but when it happens you will either know about it or you will be dead. I’m betting on someone coming up with a bio weapon that is far worse than what Covid ever was. And even though many of us shrug our shoulders at Covid, if you look at the numbers it was the worst disaster America has ever faced when it comes to American lives being lost. Now imagine that times ten or twenty or even a 100.

I hate to be a doomer but there is some significant risks that should not be taken lightly.

40

The “100% Human Made” and “No AI Used” product labels may become the new ethical consumer movements like “organic” and “fair trade”