h14h

5 ms·

31.
by h14h 4mo ago
Fair points all around. Ultimately it all comes down to execution. In theory, Apple SHOULD have an advantage given they have everything they need in house and can all pull in a unified direction. In practice, it's not always the case t
32.
by h14h 4mo ago
I think a major incentive could be to sell hardware. If Apple is able to get their hands on a local LLM capable of covering a significant % of what people use ChatGPT for, the pitch they can offer is: "Free, private, offline ChatGPT so
33.
by h14h 4mo ago
One can only hope. That said, Apple's vertical integration is a massive competitive advantage here, IMO. Nvidia's reliance on Microsoft & Windows for software support likely makes competing w/ Apple an uphill battle. If&#
34.
by h14h 4mo ago
IMO it's only a matter of time before "self-hosting local AI" is as complicated as installing an app and clicking a download button. And when that happens, the pitch to non-techy users is "Free ChatGPT you can use offlin
35.
by h14h 4mo ago
This feels like getting a foot in the door to ensure Apple doesn't entirely eat Nvidia's lunch if AI inference workloads start to shift from cloud to local. With MLX, Apple is building an answer to CUDA, and if people start switch
36.
by h14h 5mo ago
That's not all that surprising, IMO. From what I understand, LiquidAI is focusing pretty narrowly on building models that operate as the "agentic core" of a larger system. If I were going to use this model, I'd be lookin
37.
by h14h 5mo ago
Huh. I don't understand the hate because I think this looks incredible. The interior is head and shoulders the best I've ever seen in a car too. Might not look like other Ferraris, but why should it? It's NOT like other Ferra
38.
by h14h 5mo ago
The RK3576 is a really interesting/versatile chip and it is awesome to see major effort going into baking full support into the linux kernel. I could see it opening up a ton of doors for awesome FOSS hardware projects w/ AI accele
39.
by h14h 5mo ago
Given how widely varying the amount of tokens each model uses for a given task, "price-per-token" is essentially meaningless when doing this sort of comparison. Artificial Analysis's "Cost to run" model (aka num_tok
40.
by h14h 5mo ago
Is the website broken for anyone else? I'm clicking on "Join the livestream" and nothing is happening.
41.
by h14h 5mo ago
I'm worse at producing code by hand, but feel smarter overall. 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 f
42.
by h14h 5mo ago
I recently set up my own self-hosted "Knot", but haven't spent much time with other features: https://tangled.org/h14h.com/knot Overall, I think the platform looks really promising. The AtProto separatio
43.
by h14h 5mo ago
I'm currently setting up a self-hosted "Knot" for use on tangled.org. 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 th
44.
by h14h 6mo ago
A lack of tool & die isn't the West "forgetting how to make things" so much as it is the West choosing not to make the things that make things. To the extent that AI is analogous to automation in manufacturing, and "
45.
by h14h 6mo ago
Been meaning to check out Railway for a while, but now feeling happy about dragging my feet. 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 r
46.
by h14h 6mo ago
What sells me on it is I get to take a spare Gen4 m.2 ssd out of my gaming PC I wasn't fully utilizing instead of paying for 1TB of storage. Being able to drive the price down by re-using parts I already have is a pretty big selling po
47.
by h14h 6mo ago
This seems huge for subscription customers. Looking at the Artificial Analysis numbers, 5.5 at medium effort yields roughly the intelligence as 5.4 (xhigh) while using less than a fifth the tokens. As long as tokens count roughly equally to
48.
by h14h 6mo ago
For some real data, Artificial Analysis reported that 4.6 (max) and 4.7 (max) used 160M tokens and 100M tokens to complete their benchmark suite, respectively: https://artificialanalysis.ai/?intelligence-efficiency=intel...
49.
by h14h 6mo ago
If your use-cases don't benefit from RSC performance characteristics then they probably aren't outright better. But I do think they're a compelling primitive from a DX standpoint, since they offer more granularity in specifyi
50.
by h14h 6mo ago
TanStack uses streams as the basis for loading RSC data, and recommends using a route loader to access them: https://tanstack.com/start/latest/docs/framework/react/guide... AFAIK, at least when usin
51.
by h14h 6mo ago
"Critical" even feels strong. The article was essentially a collection of statements others have made about Sam.
52.
A quick introduction to "gemtext" markup(gemini.flounder.online)
2 points by h14h 6mo ago|0 comments
53.
by h14h 7mo ago
"Despite unprecedented capital investment in our R&D, our core product isn't getting meaningfully better so now we're building an app." Doesn't really strike me as the kind of statement that comes out of a compa
54.
by h14h 7mo ago
I read the README and it wasn't clear what the material benefits of these features were over e.g. having the agent write an organize markdown files with guidance, incorporating README.md with indexes/ToCs that act as lookup tables
55.
by h14h 7mo ago
> My main issue the article saying "RAM prices are crashing" (which I can't find any evidence of) and linking to an article that doesn't even repeat that claim That's totally fair. The article is written in a ver
56.
by h14h 7mo ago
My (potentially naive) take is that open models will save us. The biggest markets for LLMs (e.g. coding) are narrow-enough to be served well by smaller models with proper RL. Cursor's Composer 2 (created from a Kimi K2.5 base) is a gre
57.
by h14h 7mo ago
I do wonder how closely prices consumer RAM kits follow the wholesale prices for NAND chips manufacturers see internally. The pcpartpicker graphs you linked show consumer prices have leveled out and may even be starting to fall. Depending o
58.
by h14h 7mo ago
What does lat.md do better than a docs/ or specs/ directory you manage yourself?
59.
by h14h 7mo ago
Have to imagine each tinybox is targeting different tiers of startups trying to fine-tune/RL their way to custom models for narrow use-cases. Maybe the target profile for exabox looks like a smaller/younger Cursor? If you're
60.
by h14h 7mo ago
Would be very curious how RL benchmarks shake out vs M5 Pro/Max. Doubt local inference is the target use case near nearly as much as post-training. I could totally see something like this being super appealing for a startup looking to

More ›