Artificial Analysis's "Cost to run" model (aka num_tokens_used * price_per_token) is much better, but even that is likely problematic since it's not clear whether running a bunch of benchmarks maps cleanly to real-world token use.

I've learned an insane amount in a very short period of time, and have been engaging in much more challenging problems.

Instead of "what's the right syntax for this for loop again?" I'm asking "what's the business critical module in this system and how do I structure the test suite to prove it's working to spec?"

https://tangled.org/h14h.com/knot

Overall, I think the platform looks really promising. The AtProto separation between Personal Data Servers, Relays, and AppViews seems like the appropriate set of trade-offs, IMO.

Being able to host my git repos as a headless, data-only server is about as painless as self-hosting can get. Compared with ActivityPub solutions (like Forgejo), it's great that I get to side-step the tedium of hosting and scaling an entire webapp when all I really care about is controlling my data.

Since the initial setup, the only ops maintenance I've had to do is bump the knot-server version and redeploy (tangled.org displays a banner warning notifying me when it's outdated).

Excited to spend some more time with Tangled on other projects and test out their other features! I'm particularly interested in their native support for jj and stacked PRs.

Mainly doing it because I think AtProto is cool and self-hosting is fun, but also because owning the infrastructure that hosts my projects is definitely the direction I want to move in.

Tangled's Knot system feels like a really strong abstraction for this. I host the data in an AtProto Repository, but can rely on a third party to host/manage the AtProto Application that presents it to the rest of the world. If Tangled goes under, I can happily take my AtProto login to a different platform and point it at my Knot without changing a thing about my hosting setup.

Much more convenient that hosting an entire, siloed webapp on my own corner of the internet.

To the extent that AI is analogous to automation in manufacturing, and "writing code" to working on an assembly line, it's hard to argue the West is any thing other than a global leader in "software tool & die", so to speak.

As flashy as their DX seems to be, the fact that a sketchy single VPS node with a server, a SQLite instance, and a LiteStream hookup has a better recovery story really makes me not trust their platform.

Being able to drive the price down by re-using parts I already have is a pretty big selling point, IMO.

Also, I think Apple is benefitting from scale, since they're able to maintain the (usually too high) storage and memory prices they've had for years. At this moment in time, framework have the misfortune of being forced to pass inflated wholesale prices onto the consumer.

Make this comparison one calendar year ago and the F13 Pro could very easily beat the MBP on price spec-for-spec.

As long as tokens count roughly equally towards subscription plan usage between 5.5 & 5.4, you can look at this as effectively a 5x increase in usage limits.

https://artificialanalysis.ai/?intelligence-efficiency=intel...

Looking at their cost breakdown, while input cost rose by $800, output cost dropped by $1400. Granted whether output offsets input will be very use-case dependent, and I imagine the delta is a lot closer at lower effort levels.

But I do think they're a compelling primitive from a DX standpoint, since they offer more granularity in specifying the server/client boundary. The TanStack Composite/slots API is the real selling point, IMO, and as far as I can tell this API is largely (entirely?) thanks to RSCs.

https://tanstack.com/start/latest/docs/framework/react/guide...

AFAIK, at least when using TanStack Router, this RSC implementation seems just as capable as the others when it comes to reducing server round trips.

Comments URL: https://news.ycombinator.com/item?id=47644891

Points: 2

# Comments: 0

Doesn't really strike me as the kind of statement that comes out of a company that can sustain a ~$1T market cap...

Not trying to poopoo your project or be dismissive, just curious whether I'm missing something.

That's totally fair. The article is written in a very odd way where it makes a bunch of authoritative, factual-sounding claims and then throws a "this is all very speculative" line right at the end.

It's very interesting speculation, but can't really be considered anything more than that, despite the prose it chose.

The wealth of great open models provide an excellent base for fine-tuning, distillation, and RL. I see a lot of untapped potential in the field of bespoke, purpose-built models that can be served far more cheaply than the frontier competition. I would not be surprised if we see frontier-adjacent experiences running comfortably on a Mac Mini by year end.

With frontier models seemingly hitting diminishing returns in quality, I struggle to see a world in which gigantic, expensive, general-purpose models don't become increasingly niche.

My personal prediction is that once the VC bill comes due and prices for frontier models starts to climb, competition for efficiency will heat up. The main AI use-cases seem to be falling into buckets, and I doubt serving gigantic, do-it-all general models for every use-case under the sun is remotely cost-effective.

If common use-cases start to be more efficiently served by smaller, more efficient purpose-built models (or systems thereof), it'd make the big frontier models increasingly niche. Cursor's Composer 2 model is a great example of this.

In any case, I think it's pretty fair to speculate we may be seeing RAM prices start falling sooner rather than later.