woadwarrior01

Check out Clean Links[1]. It does everything Hyperduck does and more (supports multiple Macs) for free.

Disclaimer: I'm the author of the app.

[1]: https://cleanlinks.app/

OT: The 2nd link is the first time I've seen a Microsoft FrontPage website in decades. :)

I'm an app developer and all my apps have the "Data Not Collected" privacy nutrition label. I love the idea, but Apple's enforcement of it is very lackadaisical. I've reported dozens of apps that were blatantly lying on their privacy nutrition labels to Apple and I'm yet to see any such app's nutrition label change.

Here's a good overview of the problem: https://arxiv.org/abs/2206.02658v3

I have a co-worker who symlinks CLAUDE.md to AGENTS.md in all the repositories he owns.

There's an unspecified short/random delay between the user taking the screenshot and the notification firing. The trick used by Bluesky, Telegram etc is using a different technique. They (mis)use secure input fields to show an overlay when the user takes a screenshot.

Really happy to see this. Kudos to Qualcomm for releasing it under a liberal Apache 2.0 license.

Strategically speaking, I think this only makes sense as an anti-NVIDIA play.

If you're referring to GDPR, companies routinely evade such protections using "informed consent" / "legitimate interests" loopholes. The big ones get caught once in a while, get a slap on the wrist and continue to do whatever they were doing before, albeit with more safeguards.

Yes, iOS has userDidTakeScreenshotNotification for that.

https://developer.apple.com/documentation/uikit/uiapplicatio...

I'm only peripherally familiar with Go. IIRC, Rust defaults to move-only structs with the option to make them copyable with the Copy trait. Swift defaults to copyable structs (like Go), but has a ~Copyable generic type constraint to make them move-only.

Most languages would encode such behavior in a trait or a protocol instead of a zero-length struct field. It's type information and ought to be encoded in the type. From that perspective, I think it is a hack.

> noCopy is a special marker for types that must not be copied after their first use.

if it looks like a hack, walks like a hack, and quacks like a hack...

NVIDIA's 3090 was released in September 2020. Apple's M1 was released ~2 months later. Anyone with a 5 year old M1 Mac with 32GB or more RAM can run a 4-bit quantized version of Qwen3.8 27B on their machine. AFAICT, there are ~110m Apple Silicon Macs in the world. I'd wager that at least ~20% of those have enough memory to run this model. And if you account for gamers with NVIDIA and AMD cards, I'd wager that the segment is at least an order of magnitude bigger.

> Even if we assume that models will no longer improve and we reach a point where everyone can run Fable in their laptop, surely running 1000x Fable agents would give you an advantage.

IMO, asymptotic advantages are marginal. At least for coding, we got a glimpse into how much of an advantage it gives (or doesn't) when the Claude Code codebase leaked[1] ~4 months ago. :)

[1]: https://news.ycombinator.com/item?id=47586778

I agree 100%. IMO, it all started with ollama misrepresenting the Deepseek R1 distills as Deepseek R1, all for hype and marketing. I've had so many ostensibly technical people telling me: "I tried DeepSeek R1 and it was terrible", and every time when I probe further they'd tried the tiny 1.5B Qwen2.5 distill model that was further brain damaged by ollama's naive RTN quantization[1]. DeepSeek themselves were very forthright about it by naming it DeepSeek-R1-Distill-Qwen-1.5B[2].

I suppose the road to technical hell is paved with marketers and grifters. :)

[1]: https://ollama.com/library/deepseek-r1:1.5b [2]: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-...

I think writing is the killer use case for local LLMs.

We've had so many advancements in LLM samplers for improved text generation (off the top of my head: min-P, adaptive-P, XTC, DRY, p-less, Top-H, Top-n-Sigma, and so many more) but hosted LLM APIs only provide three basic knobs: temperature, top-k and top-p which are old as the mountains in LLM years at this point.

One thing that doesn't help the local LLM case is that all the popular VC backed local LLM wrappers also only support the same three ancient knobs because I suppose they're more preoccupied with their next fundraise than with keeping up with the advances in tech.

> The distillations of Qwen 3.8 down to a 27B model are good, but they're not on a par with frontier models.

Does it have to be? There are plenty of coding tasks, where it's good enough.

I think his point is hand wavy at best. It presupposes infinite scaling and ignores all the algorithmic efficiency wins that are being discovered. Ironically, many of which are being discovered with autoresearch style workflows, using the very LLMs that his company builds.

The #1 post on HN right now[1] is full of people jubilating about how they can run Qwen 3.8 27B on their > 5 year old GPUs. If that isn't democratization of AI, I don't know what is.

I'm sure he's smart enough to instantaneously realize this too, but as the famous Upton Sinclair quote goes, he won't mention it even if he does.

[1]: https://news.ycombinator.com/item?id=49324985

Because subtraction is not a total operation on positive integers. Negative numbers leave the domain.

Perhaps some of the people who got into traditional leadership roles got there due to factors besides their raw leadership abilities. LLMs are an equalizer, and with them, the inability to lead them simply becomes a "skill issue", for the lack of a better term.

There are some very interesting latent compaction approaches like this[1] for when you can control the whole inference stack. i.e in on-device and datacenter inference.

[1]: https://arxiv.org/abs/2602.16284

Evidently, Claude's tokenizer vocabulary size is ~15k[1]. On one hand, it's quite mind blowing. On the other hand, Anthropic models' token (in)efficiency makes a lot of sense in that light.

[1]: https://xcancel.com/magikarp_tokens/status/20878591737488549...

This is so reminiscent of Cider (Google's internal cloud IDE) which had the best multiplayer support that I've ever seen. It was such a joy to pair program with a colleague on a different continent.

All this posturing is all for naught, because people can do proper logit distillation into smaller models with open-weight models.

TBF, he has done more than his fair share of time working in BigTech.

Good luck writing fanfics with primitive top-p and top-k samplers.

Huggingface now owns llama.cpp, btw.

https://huggingface.co/blog/ggml-joins-hf

ICYMI, llama.cpp was also VC funded. Search for "ggml" on this page:

https://aigrant.com

For once. A former colleague once quipped: The only way to make decent money in the business of language runtimes is to get a job at a big company that depends on it.

The value has already been materialized. They were acquired by Qualcomm for $3.9 billion.

IIUC, it's the ability to export layered PSD files with images from multiple sensors.

When you get around to doing that, you might find the Github repo and its accompanying medium article useful.

https://github.com/0Itsuki0/SwiftUI_P2PWithNetwork