I think agents will get much better at solving these problems in the medium term. In the short term you should at least be observing what the agent is doing when vulnerabilities like this are so easy to create. Using AI to generate structured RPA tasks like with browsable.app or director.ai is still a better option for now for many tasks
They made $4 billion last year, not really "little to no money". I agree it's not clear they can justify their valuation but it's certainly not a bubble.
But didn't they spend $9 billion? If I have a machine that magically turns $9 billion of investor money into $4 billion in revenue, I need to have a pretty awesome story for how in the future I am going to be making enormous piles of money to pay back that investment. If it looks like frontier models are going to be a commodity and it is not going to be winner-take-all... that's a lot harder story to tell.
There is a pretty significant different between “buy $9 for $4” and selling a service that costs $9 to build and run per year for $4 per year. Especially when some people think that service could be an absolute game changer for the species.
It’s ok to not buy into the vision or think it’s impossible. But it’s a shallow dismissal to make the unnuanced comparison, especially when we’re talking about a brand new technology - who knows what the cost optimization levers are. Who knows what the market will bear after a few more revs.
When the iPhone first came out, it was too expensive, didn’t do enough, and many people thought it was a waste of apples time when they should be making music players.
It's a commodity technology and VCs are investing as if this were still a winner-takes-all play. It's obviously not, if there were any doubt about that, Deepseek's R1 release should have made it obvious.
> But it’s a shallow dismissal to make the unnuanced comparison, especially when we’re talking about a brand new technology - who knows what the cost optimization levers are. Who knows what the market will bear after a few more revs.
You're acting as-if OpenAI is still the only player in this space. OpenAI has plenty of competitors who can deliver similar models for cheaper. Gemini 2.5 is an excellent and affordable model and Google has a substantially better capacity to scale because of a multi-year investment in its TPUs.
Whatever first mover advantage OpenAI had has been quickly eliminated, they've lost a lot of their talent, and the chief hypothesis they used to attract the capital they've raised so far is utterly wrong. VCs would be mad to be continuing to pump money into OpenAI just to extend their runway -- at 5 Bln losses per year they need to actually consider cost, especially when their frontier releases are only marginal improvements over competitors.
... this is a bubble despite the promise of the technology and anyone paying attention can see it. For all of the dumb money employed in this space to make it out alive, we'll have to at least see a fairly strong form of AGI developed, and by that point the tech will be threatening the general economic stability of the US consumer.
Every new tech has companies start and fail, as the consumer market changes and things are tried and fail. There’s no way to predict ahead of time what will work, what won’t - and so a thousand ships are launched with only a few reaching shore.
Is that a bubble? I suppose it is; it’s also probably the right strategy.
> When the iPhone first came out, it was too expensive, didn’t do enough, and many people thought it was a waste of apples time when they should be making music players.
This comparison is always used when people are trying to hype something. For every "iPhone" there are thousands of failures
> I started a business that would give people back $9 if they gave me $4
I feel like people overuse this criticism. That's not the only way that companies with a lot of revenue lose money. And this isn't at all what OpenAI is doing, at least from their customers' perspective. It's not like customers are subscribing to ChatGPT simply because it gives them something they were going to buy anyway for cheaper.
How do people integrate steps on websites/web scraping into their larger workflows? I’m looking to try and integrate my own browser RPA tool [1] into n8n but I’m not sure how useful it is.
Notably not an AI agent like Operator, Manus, etc. which are largely unreliable for the time being. Instead this uses AI to turn your task into something repeatable and configurable.
Currently focusing on scraping use cases but hope to make it more powerful soon so it can actually do complex tasks rather than just extracting data.
It's RPA for browsers which is not fundamentally new, what I'm trying to do that is new is use AI to make it as easy as possible to create automations. Most of the existing tools require you to locate CSS selectors, XPaths, etc. whereas this is just point, click, type, describe data you want to extract in English, etc.
Still early days and it works much better for some tasks/websites than others but it's improving rapidly and I'm quite excited about it.
Also hoping that the likes of OpenAI Operator, etc. are rolled out in a way that I can use them to build a better product rather than being runover by them.
Just been playing around with the bounding box feature myself - do you mind sharing how you figured out how to translate the bounding boxes back to the correct dimensions/position on the original image?
I would say things are worse but it's difficult to disentangle it from the other major events like Covid and the war. Also worth pointing out that the UK's economic stagnation began well before Brexit and goes back to the 2008 crash: https://pbs.twimg.com/media/FvPzCWYXwAEoIRu?format=jpg&name=....
As someone who supports Brexit, I always thought there would be greater costs in the short term but getting out of institutional structure of the EU would be beneficial long term, as long as we take advantage of the institutional freedom and try to get onto a different path as a country (investing heavily in science and technology, having a more rational approach to regulation, having an immigration system that attracts great talent while also controlling the borders and thus making the whole thing less toxic).
I haven't overall changed my view as the EU still seems to be on a very bad path as I see it, I can't say I'm hugely optimistic about the UK either though. There are glimmers of hope like the new ARIA institution for blue sky science, but the main political parties are a pretty depressing spectacle.
The trouble is the UK electorate has no appetite for the kind of changes that would make Brexit a success. Instead we are going to slowly creep back in over the coming decades. Having such a large trading block right on our doorstep makes that inevitable.
In Australia no-one ever discussing anything about current propsperity in the context of "disentangling it from the other major events like Covid and the war".
Sure those are things, but we are not trying to make sense of current affairs through that lens as a major factor.
Same as another commenter, I try to make it a habit of either referencing documentation when I need to answer a question or making a note to fill it in when I can't find a reference.
For searching documentation there's a lot to be desired too so I'll make a shameless plug for my side project: https://neat.wiki. It's a simple wiki creator on top of Google Drive but I'm layering on some GPT-3 goodness for semantic search and question answering so you can always find what you're looking for.
BuiltWith.com is pretty much just one guy doing over $10m per year. At the notch below there are others like Pieter Levels who I think is doing over $1m per year with Nomad List and RemoteOK and I think it's mostly just him with some part time help. Probably a few more that no one has heard of.
