computomatic

It’s a business-scale cost. When I say materially useful, I mean replacing multiple salaries and improving quality at the same time. Not some variant of “make an existing employee 20% faster.”

The personal hobby version is to get the Claude max plan.

“Marginal” improvement isn’t really going to be enough. Products need to be closer to 10x better than the alternative to get traction (which can be divided across several dimensions). AI, I think, is just starting to get there.

I don’t know about you, but for me materially useful AI costs around $10,000/mo in tokens. But it is very useful.

Claude subscriptions help, but only if I come in with the expertise to build the entire system around the model and I don’t mind the whole thing grinding to a halt when my session limit is saturated. Anyone selling the system off the shelf is going to be charging API rates. Computers were cheap in comparison.

I came across this recently and was initially surprised too. It makes sense when you consider that any company trying to build models (or an agent/value-add layer on top of a model) needs benchmarks. The best funded will do the full vertical - build their benchmarks and models together - but many will outsource the benchmark. This is particularly valuable when the dataset is difficult to compile (robotics being a prime example) - it makes more sense to let the benchmark vendor focus on that.

Wouldn’t it just be easier to ask their team to upvote the post? They probably all have legitimate hacker news accounts already and are happy to upvote their work.

It’s more of a fun anecdote, but people forget that there was once a plague of stolen Apple devices. Apple customers were frequently targeted by pickpockets, etc. because of the value of the devices. I have no evidence but I’ve always believed this was the root of the privacy focus - Apple could lock devices to biometrics and eliminate various workarounds, making stolen devices worthless, and the thefts evaporated practically overnight.

Phone theft is a growing problem again, but it’s still less common than it was because they need to get an unlocked device, and I don’t hear them targeting Apple in particular.

> when you want to let them jump in and out of multiple apps (e.g. pick what game they want to play on their own)

The solution to that is Assistive Access.

It’s more painful to use than it should be, but it absolutely solves this problem.

I’ve had the privilege to work at a company with a highly aligned legal team. It turns out there are quite a few people who get law degrees and are just sensible, chill, fun people. And if you can hire them onto staff, there’s no limit to the awesome things you can do.

Which also seems a bit silly because rural areas have plenty of events and many use e-tickets. Maybe OP just doesn't like events.

GPUs aren’t becoming outdated in practice. An H100 still rents for something like $3/hour. The only reason to take them offline today would be due to limited data centre capacity. In space, that doesn’t apply.

For now, at least, this market doesn’t behave like the one for CPUs which are more fungible and general purpose.

Rather than thinking about the value of a GPU over time, it’s more useful to think about the value of a model (or generation of models) that will run on the GPU.

Sol 6 launched last week. People will build products/agents/systems on Sol 6. Many of those will end up staying on that model for the lifetime of the end product, which means there will be long-term demand for the model, and that means long-term demand for the GPU set that the model runs on.

For now, at least, models are tuned to a specific GPU and don’t move to the newest hardware very easily.

At the end of the day, will the lifetime value of a GPU exceed the cost of launching it into space? That’s TBD. But there’s growing evidence it might.

How do you evaluate that?

Zoom in on the large tree in the background. It looks as if it has ornamental balls hanging off it.