This is why I'm bearish on Anthropic, OpenAI, and friends. I am not confident that we will continue to see the same pace of improvement in frontier model capabilities as we have seen over the past year or two - not using similar mathematics at least. But I think that getting results that are close enough to the same standard to be a realistic substitute but in a model small enough to run locally may well happen quite quickly. And if it does - where is the moat to defend these AI organisations with their astronomical budgets when they're already starting to price more realistically and that's already killing a lot of the hype they've enjoyed until very recently? They have an accidental moat because they bought up the global supply chain for storage but that surely isn't going to last once the data centres to hold that storage are becoming liabilities.
If model performance asymptotes and CPU/GPU and RAM keep growing, even slowly, then eventually we will have frontier models on desktop that are totally competitive with hosted. It’s only a matter of time.
You already can if you’re willing to spend many thousands of dollars on a beast of a machine. I’m talking about middle tier desktops and laptops here. Maybe eventually even phones.
The only way hosted stays strongly competitive in that world is if they can keep pushing the frontier or by playing the classic social media and SaaS games of network effect building and integrations.
Many people might still use hosted, of course, but what I really mean is that their multiples won’t be justified and they will have little to no moat. AI will become commoditized, like a sophisticated next generation form of an encyclopedia with search.
> This is why I'm bearish on Anthropic, OpenAI, and friends.
Just because you can do more and more things at home (thanks Moore and Dennard), doesn't preclude needing things also done remotely. The number of at-home systems seems to have fed a growing number of remote systems (especially once always-on connectivity became ubiquitous).
It's basically the angle Apple is going for: do as much locally (for the sake of privacy), and then offload when it becomes "too much".
I agree that one doesn't preclude the other. But the sky high valuations we've been seeing for the AI industry recently can only be justified if they bring about a fundamental change in our society and those companies continue to bring in the lion's share of the resulting profits. I don't see why everyone else in our society - particularly other large businesses with lots of money to invest - is going to play a game by the AI companies' rules once they can take their ball and go home and still have most of the fun without paying much for it by comparison.
That would seem like a logical solution. So wouldn't it be convenient for the expensive payment methods if legalities prevented merchants from charging higher fees to customers using them?
Indeed. It's a triumph of consumer protection laws failing to protect consumers. Merchants here have to set their prices a bit higher to compensate for the fees and you still have to pay those higher prices as a customer even if you're using a more efficient payment method. I will never understand why the law wasn't set the other way - requiring explicit disclosure of payment fees to end customers and prohibiting payment services from incorporating these kinds of anticompetitive terms in merchant agreements - so that everyone could make an informed choice and market pressures would push the transaction overheads down.
It might have been regulatory capture - though I have seen no specific evidence of that myself. It might simply have been the old story about a road and good intentions. At this point it doesn't really matter how it happened - it would be better if the situation were fixed in any case.
The problem is that the medium term prospects are irrelevant to all the businesses that won't be around long enough to enjoy them.
Smaller businesses in particular - not so long past the COVID disruption and already facing significant challenges in areas like logistics and energy supply costs - will not necessarily have the reserves that older and larger businesses often do to withstand another multi-year price shock.
If there is one good thing that the generative AI tools have shown beyond any doubt it's that the classic "good programming" practices are still useful and effective. Self-documenting code. Modular design. Clearly defined architecture. Incremental development. Coding standards. Automated tests. Automated everything.
If there's a second thing the generative AI tools have shown beyond any doubt it's that many of the more modern (relatively speaking) "best practices" that have always been over-hyped and questionably-evidenced really do tend to produce worse results. LLMs take these methods to their logical conclusions and show us the end result much sooner. You can't just iterate your way to a solution when you don't even know what problem you're trying to solve. If you don't have a clear spec then you don't know what a correct product looks like. You need to invest time in reviewing code properly. If you don't keep the big picture in mind then the big picture becomes a mess.
Maybe one day the LLMs will leave me out of a job but at least I'll feel validated first!
> If there is one good thing that the generative AI tools have shown beyond any doubt it's that the classic "good programming" practices are still useful and effective
If you apply those practice, then quickly you find yourself using the agent as merely a writing boost. And there’s an inflexion point when coding is no longer a bottleneck. Instead, you spend more time on thinking about design. You can see it in open source projects where most PRs are just a few line diffs. The bottleneck is knowledge and problem solving talent.
If you apply those practice, then quickly you find yourself using the agent as merely a writing boost.
I don't know what that means but I have seen no evidence so far that if you don't apply those practices then your code will be anything other than unmanageable spaghetti if you leave AI to maintain it for long.
Coding has never been the bottleneck for good developers. Part of the reason for that is that good developers know how to isolate different aspects of a system and so keep each individual aspect relatively simple and self-contained. Another part is that good developers were already standardising and automating a lot of the grunt work. These traits are also advantageous for keeping generative AI on the right track and keeping its proposed changes manageable.
Yeah and that design and insight is the tiring part and while fun a bit less satisfying in the way that writing a nice bit of boiler plate or populating the struct members for your data type can be. One thing is you can work on design and insight while taking a good walk around the block, which is nice.
I spend that time mostly on the sofa, or in front of a whiteboard. Or sometimes a live brainstorming. Typing code is actually relaxing. What looks like relaxing is actually hard thinking.
I try to explain this to my family (currently in a pure remote job) but it is difficult to make the case persuasively. Honestly tempted to start recording voice memos to self to capture the ideas for next steps, which I can run five of in tmux if i keep the directories straight.
I'm not sure FAANG does look good on a CV any more. The skill set to be effective in those environments is quite specialised and crucially it's very different to what you need in a lot of other software development organisations. There appeared to be a happy cycle for a while where very well paid devs working in one of the few FAANG or FAANG-adjacent companies could jump to one of the others because they were "in the club" and had experience of working at a truly global scale that most software never needs. Those days seem to be over with the mass layoffs and hiring limits. And if you're not working at that scale - and outside that small world almost nobody actually is - those skills aren't always very transferrable and other types of experience often have more value.
When I worked in aerospace one of the surprising things to learn was we had an unwritten 'max allowed' percentage of people hired from Boeing overall and applied to each team because we did not want to incorporate Boeing's culture. If I was still in management at a software company I'd probably apply that as a consideration with regards to FAANG. Even for those hired ex-Boeing had a much stricter vetting not on skills but personality/approach/vibe(?).
I don't know exactly I never worked there and I was IT not design/qa/factory floor. My understanding was slow, non innovative, not efficient, and prone to complain. The ex-Boeing people we had were great though. I think it's probably just normal 150k employee companies end up toxic stuff. Just know we had a quota.
I am also increasingly worried by the potential for violence here. This is a social experiment that is harming the daily lives of millions of people in very obvious ways already. The environmental costs for the data centres are not insignificant. The economic damage from allowing AI to have so much funny money when it doesn't make much real money to justify it could be disastrous on a generational scale. Governments aren't making any serious attempt to regulate and if anything are drinking the Kool-Aid. We might be on a path that literally collapses the established Western capitalist order within a generation but historically societal change of that scale usually has a body count and I have no idea what comes afterwards.
AI "support" bots that just attempt to read the published documentation for you are possibly the most annoying thing to have come out of the current AI plague.
Even Stripe - once legendary for the quality of its support - has apparently given up now. I had to deal with it recently over a case where the merchant was seeing an unexpected change in the way it was collecting payments and the AI bot was worse than useless - it actively suggested incorrect explanations and resulted in several days of trying to change the wrong things while the problem persisted.
For my own businesses we give this issue a heavy weight when choosing which services to use. We have even seriously considered moving existing integrations to different services over this one issue recently. If we're integrating with a service then we want to know there's a real person who can actually help if we have questions or anything goes wrong. Failing to provide that because it's cheaper to push everyone through the AI bot is a statement of intent about how much you value your customers.
The problem is using AI to “push” the answer to an asker.
Unless the company has hidden docs they use for support (which why would that ever benefit them), I could get just as good of an answer if I point my LLM at your docs (“pull” an answer). In fact, the response might be better because I have context set up to tune it to my understanding.
Instead, you (company / support agent) have decided that I should instead have a conversation with an LLM through a worse, more opaque harness to the detriment of all of us.
It would be interesting to hear the other side, the people running support. I wonder what fraction of requests are answered by basic knowledge and stuff clearly in the docs. At some very high fraction I could see a lot of pressure/incentive to optimize for these cases.
From talking to people I know on support teams at hyperscalers and other tech companies: A mind-boggling huge portion.
But even as an AI believer and thinking it's great for this sort of thing, I also think if it's good enough for this sort of work, it should be good enough to identify when it's not capable of providing value, too.
Gosh, that's kind of sad. I hate having to wait 12 to 24 hours for an answer from AWS support - there's no way I'm raising the ticket without exhausting all avenues from my end.
> AI "support" bots that just attempt to read the published documentation for you are possibly the most annoying thing to have come out of the current AI plague.
I disagree, personally. The way they were employed in GP's example, yeah, that's annoying and hostile.
But almost every time I go to some documentation page for a commercial product now and it doesn't have that, I find it pretty annoying (I don't expect an open source library to wire this up and pay for it, ofc).
It means I have to convey these docs to my own LLM, I will waste cycles and tokens fetching them, etc.
Your version sounds like it's potentially useful. The thing that winds me up is when the online chat that used to be talking with real support people gets quietly replaced with some LLM-backed noise generator and there's no way to contact real support people any more (possibly because 95% of them were laid off).
Yeah that's a totally different thing, agreed that's really hostile and companies over a certain amount of revenue should be forced to have human support agents.
> AI "support" bots that just attempt to read the published documentation for you are possibly the most annoying thing to have come out of the current AI plague.
Meh. It's no big change to before, where you had first-level support search for something vaguely related and just dumping a template to the user.
The thing is... it actually works better than I'd like, because in a lot of cases it turns out that you (as the user) forgot to follow one tiny step in the documentation.
What I'd actually like to see as the user is the AI actually going over my AWS account, looking up the resources and their state on its own, and figuring out what exactly I missed, but (un?)fortunately that cannot be easily done for IAM permission reasons...
The big difference IME is that first-tier support at decent companies at least had the grace to recognise when they couldn't help and needed to escalate. I can count the number of times I've seen an AI bot automatically escalate when it was unable to find the solution on zero hands.
Agreed. I remember a couple of years ago raising an AWS support ticket and eventually got an answer about it being some undocumented limitation in Aurora Serverless v1. Now I get equivalent of googling my question before I can even raise a ticket.
Even more annoying, is when the integrated "chat with AI" boxes don't actually have full knowledge about the website. Tried a couple of different such boxes, and in the end I still had to crawl the website on my own to find the information.
And presumably the instructions for this have been on display on our local planning department in Alpha Centauri? If a user isn't even aware that their local disk is being encrypted without their knowledge or consent then why would they think to set up recovery keys?
It's not a matter of belief. Signal does not provide a way for me to download my own messages off my own devices and safely store them using my own secure backup facility.
Obviously Signal don't owe me anything. I'm not paying for the product and I appreciate what it does offer and makes available for free. But it would be much better if it also supported local backups under the user's control.
The GDS is one of the more credible parts of government IT in the UK and IME generally well respected. The government websites and online services have largely been well done. But there are limits on how much that organisation can take on with the resources it has and it's still subject to the same challenges around compensation and working environment I mentioned in another comment that make it difficult to hire and retain good people. Unfortunately it's not realistic to build all government IT projects in house that way at the moment.
reply