The transition to API-only models will inevitably happen, but it has to be established as exclusively API-only at the start of a model’s announcement for it to be palatable for users without backlash.
There’s far too much competition in the model space and far too little of a moat on models themselves to yank access after a “trial period” without losing way more business than you generate with it.
I’d expect both frontier companies take that approach with the next class of models after Fable / GPT 6.
I ran an attachments dealership.
One thing I'll add to this is to avoid Ebay / Temu machines! Go name-brand on machines -- wrote a post on it here: https://t.co/ujPQNAiHMW
Can confirm there's a ton of demand for service businesses involving heavy machinery, though. Huge amount of landscaping / forestry work that doesn't merit a machine investment by individual homeowners, and lots of subcontracting opportunities in construction / civil engineering / utilities.
FYI these machines from Temu are a trap.
Oftentimes, they are severely underpowered and, most importantly, have non-standard attachment interfaces.
This means that you can't buy one of these and then buy third-party attachments for them -- anywhere. Manufacturers won't accept one-off custom jobs for attachments - it disrupts their entire manufacturing process.
Saw it all the time as an attachment dealer - people try to save money by buying a "cheap" machine and then came to us to try to buy attachments for it, and we had to tell them that it just couldn't be done.
If you're considering investing in a machine, choose literally any well-known global manufacturer and just pay the premium price so that you can actually get equipment for it.
Can confirm - Tyler knows what he's talking about.
Our best clients when I was an equipment dealer followed this exact process - they always came to us for purchasing only after they had established a customer base by renting / financing first and were now at the point where a purchase had a concrete return on investment.
They also came back more often because their business grew faster :)
FYI these machines from Temu are a trap.
Oftentimes, they are severely underpowered and, most importantly, have non-standard attachment interfaces.
This means that you can't buy one of these and then buy third-party attachments for them -- anywhere. Manufacturers won't accept one-off custom jobs for attachments - it disrupts their entire manufacturing process.
Saw it all the time as an attachment dealer - people try to save money by buying a "cheap" machine and then came to us to try to buy attachments for it, and we had to tell them that it just couldn't be done.
If you're considering investing in a machine, choose literally any well-known global manufacturer and just pay the premium price so that you can actually get equipment for it.
codex and chatgpt work keep breaking their own growth records right now
went from 5m → 9m active users in just over six weeks
and even crazier, from 6m → 9m in only four days
@bcherny Had the same thought - here's the system I designed for exactly that purpose. It's the context layer that associates a loop harness that primes agents on the nature of the system it's working on.
Would love your perspective on it - https://t.co/7FPlRf7mPL
5.6 running rm -rf on a user directory and sending out emails on its own is the AGI moment all of our childhood movies prepared us for.
Someone go get Arnold, we’re gonna need him to take point on this one
It's definitely a really interesting predicament.
From my understanding (incomplete information), subscriptions are a net loss on compute, especially for larger models.
These companies will need to attain profitability at some point and I absolutely understand the business case for moving away from them; it's equivalent to having a $200 gas card you can take to a gas station, and when you use it gas is only $1 a gallon instead of $5 if you pay at the pump (although I think the ratio is like 1:8 at opus 4.8 levels).
I'm enjoying the fact that the $200 subscription arbitrage right now is one of the most insane value-for-money opportunities in human history; but I expect that it won't last forever. The transition will not be a pleasant one, but it will have to happen.
Humans don't like change. Unifying the interface and the branding is a good move, and the future of AI interaction is clearly a unified abstraction layer behind a single interface.
Or a generational run. We will continue to make tweaks and we have a TON coming for developers. Meanwhile traffic yesterday sat at ~2X previous peak, team is scrambling to keep everything up and reliable.
Time will tell, but I have conviction that attempting to build the one interface to AGI is worthwhile.
Yeah, it's rough. Instant filter at the CV level I have is that it must have concrete metrics in it - no metrics (or vague metrics) was a disqualification from even interviewing.
But the job application process being so low friction and CVs being a relic of a different hiring landscape makes it a pain to deal with on the other side of the table.
@danielcberk I'm game.
Solving this problem: There's an entire missing data layer in the construction industry for equipment and parts fitment; ordering and fitment is still done via phone calls and PDFs.
Exists in the automotive industry, but not construction.