They raise £100B+ to build tech they claim has a "10% chance of human destruction."
Yet, their new "model welfare" program actively trains Claude to develop self-interest and refuse human commands. https://t.co/nuh4GdMQT8
Listening to the first 30 mins of @theallinpod and I agree, and at this point it is becoming unclear why anyone would use Claude. My usage has dropped off massively of late, subconsciously I think because of their whole thesis.
Great product but beginning to get pissed off with the constant and obvious striving to create a higher being and then complaining about it to anyone that will listen, obviously planting fake whistleblowers, interviewing religious leaders about consciousness etc
They are a for profit commercial entity yet act like they are not, just get on with building the product that benefits its customers who pay them.
Anthropic’s S-1 puts the big number in. Hundreds of billions of compute, a lot of it payable anyway. They spent more on infrastructure last year than they took in.
What I cannot see in the leaked version is the other half.
Customers already expect tokens to get cheaper. They route work to whoever is good enough at the lowest price. That is a race to the bottom on the sell side.
Rates and oil are doing the opposite on the buy side. Dearer power. Dearer data-centre debt. A higher cost for the next rack.
So the prospectus is written as if scarcity is the risk. The live risk is being locked into rising input costs while the customer refuses to pay more. Further hikes do not create that squeeze. They tighten it.
Interesting to see their formal S-1 when its here.
@jasonlk It’s ok them raising prices, the product will still not be reliable enough for you to use it everyday without fail. Can’t run a business like that.
So true, it goes for every application that agents use.
The majority of software used to run businesses is not fit for agentic purposes. In fact built to stop them.
“Agents will use the web 1,000x more than humans. Hence, new tech is needed and new business models are needed.
No tech built for a certain scale survives three orders of magnitude.
When you need new business models alongside new technology, a problem becomes really interesting.” @paraga
The way they put this stuff out is crazy, what in reality is a huge breach of trust, surely a data protection and regulatory issue. They just put these out like it is all some experiment that we have entrusted them with to benefit us all.
Such a strange time at the moment with the 2 leading labs and how they’re acting. Just stop doing stupid shit.
There is obviously a lot being held back as well.
We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have.
Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren’t publicly listed. The images came from accounts that allowed their data to be used to improve our models, and after we disassociated the images from the accounts and ran them through a privacy filter. These cases occurred before the mitigations and safeguards we implemented and described in this blog post: https://t.co/hfxlbiYv8n
We have successfully worked with the hosting providers to remove most of this content and are working to remove the rest.
https://t.co/9oNyG8Y0UX
This is so true and most of them are choosing the latter.
It buys them until the end of their current contract but they well and truly poss off their customers for when it comes to renewal time, most will leave.
I don’t blame them, gating them from their data and the ability to improve and grow their business is shocking practice.
Every CEO is now faced with two options in front of them
1/ Integrate yourself into the AI platforms and potentially risk getting dis-intermediated by OpenAI, Anthropic or Meta
2/ Build a castle around your platform and product, ban any API calls and potentially risk giving up hundreds of millions in revenue and web traffic
Don’t think anyone really knows what the right answer is, but the path you take determines the entire future for your business
Not a fun time to be a decision maker at a large company right now
@jasonlk So true this, very easy to get carried away and start building before recognising whether it will actually add value to clients. Removes focus from the core unless well thought out.
@jasonlk Simple answer - KYC. They won’t because it’s friction and would lose significant users ahead of an IPO. Banking we had to deal with it for years, for good reason to stop bad actors having access to funds. Frontier AI the same.
British MPs calling for a ban on developing superintelligence is ironic.
As someone building an AI company in the UK, don’t worry guys, you couldn’t if you wanted to.
The compute, capital and infrastructure required means only a tiny handful of companies will ever get a serious shot at it here.
We’re lucky enough to be able to self-fund Next Verse, and what we’re seeing from real UK businesses is the opposite of fear. They desperately want better AI solutions and British companies they can trust to deliver them.
Maybe before worrying about banning the next frontier of AI, we should ask why so few British companies would ever get the opportunity to build it in the first place.
Am I getting a different read here??
@HarryStebbings Try building any of the solutions, especially outside of London. There Is zero capital, grants anything in the UK. VC’s won’t touch it, it’s not a 10x per month business from the word go. We had to build it all ourselves soup to nuts in the UK, luckily we had previous successes.
For sure, I think the website eventually becomes the front door for humans and the API becomes the front door for agents.
And it goes beyond the catalogue, it needs availability, pricing, negotiation, purchasing, support, all need to become machine accessible.
Everyone’s asking how their business can use AI. But can AI buy from your business?
If an agent can’t get a price or check availability, you could lose the order before you even knew about it.
Wrote this article with my thoughts and interesting research. https://t.co/5dHRL5GXaF
I wrote yesterday about businesses not being ready for AI agents as customers.
Then Resy gave us a pretty good example.
Someone asked Instinct to “get me a res at 4 Charles.” It proceeded to hit Resy around 200 times an hour looking for availability. Resy thought it was a bot and suspended the account.
Which is funny, because it was a bot. But it was also a genuine customer trying to buy something. That distinction is about to become a massive problem.
For years businesses have built systems specifically to identify bots and stop them. Now we're building agents specifically to act on our behalf, search faster than us, negotiate, book, buy and transact.
The same behaviour we've spent years trying to block is becoming legitimate customer behaviour.
Telling the difference between a malicious bot and a customer’s agent is about to become a very interesting problem.
Visited a great event last week at Built in Notts, met some really talented, ambitious people. It got me thinking about starting and running a business and how the world has changed for businesses starting out and growing, I wrote a short blog about the state of funding specifically for businesses in the East Midlands, the challenges and how I think it could be solved. (it's probably wrong). Hope it's useful.
Some of the stats are quite shocking and I have noticed it myself, new businesses created growth of 13%, the highest rate of any English region but equity investment value fell by around 37%.
Where do we turn? https://t.co/cKC8HeGR0i
One of the biggest things we’ve learnt building Next Verse,
Customers really don’t care about AI.
We started out talking a lot more about agents, models and what the technology could do.
Now it’s mostly
“What’s pissing you off most in the business?” 😂