dannyw

people like ease of use, UI, docs, and community.

after all, why use opnSense; just use suricata and $PROPER_X! why use TrueNAS; just use $DISTRO and ZFS!

We _think_ this power / divide feels harmless right now, but I'd bet money that NSA, CIA, etc have access to the latest and greatest unrestricted models; and massive compute. At least for OpenAI, and even if not willingly for Anthropic, I'd bet money NSA has it too. (After all, when Google decided to migrate to HTTPS, the NSA decided to hack Google's internal network to preserve their taps).

Who knows what they are up to.

Mossad/Israel at least:

> Documents supplied to the Hamburg court indicated that Hanning was the go-between with the Israelis. According to media reports, these Hamburg court documents contained clues to his other business activity with classified BND information.

https://www.irishtimes.com/world/europe/2026/10/06/former-ge...

Where there’s smoke there’s fire, and western intelligence precedent (eg russian “illegals program”) is that when it’s Russia/China, it’s named.

So the omission here itself (and denial of Russia) is some additional speculative information, on top of his already known ties to Mossad.

That’s kinda very small and light for modern trillion-param LLMs.

Crows can remember a human face for 17 years, and pass this knowledge through their murder (group).

We know animals communicate with each other, and the trend is we generally discover more and have underestimated what many animal species practice and can do.

Everyone agrees that homo sapiens and homo neanderthalensis have talked together and lived (and bred) together; what makes inter-species communication (assuming a translator or device) impossible?

No, I don’t think “universal translators” are outside the domain of science fiction; but I think us building a device that enables two-way communication between one species and us, is probably more likely than not.

[1]: https://urban.uw.edu/news/crows-hold-grudges-against-individ...

Yes, the timeframe does matter.

An actively and easily exploited (just a port scan), and high-impact root RCE needs faster patching than one week.

When there was a more user visible bug (2017, empty root password gives you root, CVE-2017-13872), Apple managed to remotely patch this across all supported macOS versions in about 26 hours + next macOS online, end to end, without user intervention or manual updates.

And that was a decade ago before Apple built “rapid response security updates”

If you’re willing to pay and don’t mind something gray market, I’m a happy customer of SerpAPI

Usually luxury stores have sprinkler protection, availability of fire hose reels, etc.

In Australia this would not comply with our fire code. Putting it behind a cabinet is fine; but it must have visible signs that can be seen from 20m away with a 20/20 eyesight (assuming no obstruction etc).

This applies to all portable extinguishers; even if it’s additional ones you’ve added beyond what’s required. The idea and intent is simple: train people and first responders to locate portable fire extinguishers by the sign.

I agree there’s zero chance of written documentation that the building isn’t up to code.

That’s very different to it actually complying with code however. And to your second example, if I was physically attending WWDC back when they did they keynotes, I’d much rather seeing illuminated exit signs than having them (illegally) turn them off in case of a fire.

Much of building codes is either outdated, no longer the most effective way to achieve something, etc. But these building codes are written in blood. It’s why we have them.

It’s a LLM generated article, and from the sentence style, I think a GPT. It looks like it’s been put through a “humaniser” or had less copy editing.

Just about all of gadgetreview.com is LLM writing these days.

It shouldn't be this way, but DDR4 is perfectly fine and will last for years. RAM generally has low failure rates; and prices are relatively better on the used market. If you want ECC, many used DDR4 ECC sticks seem like a 'relative bargain' compared to how much DDR5 ECC goes for.

As for GPU, AMD hasn't raised prices by as much; and honestly my 3060 12GB from many, many years ago plays absolutely every single game I want to play at 1440p; with nice visuals. Yeah, I can't pump everything to ultra, and maybe I need some DLSS for the most demanding games, but it's far cry from "old" or "obsolete".

In terms of actual experience or enjoyment I get from video games, I'm not missing out on anything.

It's legitimate to feel disappointed and upset at the current state of the PC market, but slightly older desktop hardware is still more than plenty fast enough.

I own both a MBP M5 Pro 48GB, and a MacBook Neo, and I use my Neo probably 2x as much as my MBP.

This looks just like financial engineering / creative accounting; to comply with things like GAAP rules and appearing less capex heavy, while functionally having the same effect.

In the end, Amazon is still going to be the ones leasing and hence using the chips; nominally the 'owner' changes hands, and it looks like a bond (with a small equity component) in just about every way, except through some legal and accounting magic, it doesn't go on their balance sheet.

For additional context, Greg Kroah-Hartman has been contributing to Linux for 3 decades, and is _the_ maintainer for Linux's stable branch, as well as many other core parts of Linux.

My takeaways:

* Do not panic. Do acknowledge that LLMs are sycophantic, and LLM companies are trying to sell their stuff. "Yes, push back hard".

* If it smells like AI slop, treat it like AI slop and relax. If someone sends you 50 security reports, and a few look wrong, just calm down and ignore them; or ask for proof-of-humanness.

* A report without a patch is, for better or worse, worthless if you maintain widely used open source software.

* Of course, there are real vulnerabilities being discovered and reported. Keep calm, keep fixing real bugs, and carry on.

* Delete as much code as you possibly can. Reduce your surface area. Do the same thing you've always been doing.

Security has always been a cat and mouse game. The usual precautions about software updates, least privileged access, separation/sandboxing, and precautions like not pasting/installing random software (especially something on Show HN) continue to apply; but at this point, I'd suggest thinking more towards a model of "How do I best detect, and contain a compromise" model.

One thing I do is a wallet.dat honeypot with (now) ~$1700 of bitcoin. Any movement essentially shuts down my home network from external access. At worst, it's a justifiable tax deduction, not capital gains :)

There’s always going to be consumer products, even if the value proposition is worse.

If there’s a silver lining, I hope it leads to more optimisation, less Electron, and less bloat.

Typed from my 8GB MacBook Neo; which is honestly plenty fast enough. It “feels” like a 16GB windows laptop, even tho of course I know it’s not, but there’s something about optimisation.

GLP-1s, while being pretty miraculous as a whole, do come with risks like every other drug. There’s plenty of 1 in 10k or 100k side effects.

Kurtzerkag has a good video about it: https://youtu.be/TYhNHX372ek?si=5ESz3ykXykyUcmYg

Love this concept, simple but brilliant idea with lots of utility. I think this can fit in a decent niche. I’ll give it a try!

Both labs are spying? Employees hang out at the same bars, have overlapping social circles, etc. Alcohol does what alcohol does.

Adaptive thinking was a good idea though. The old method of manually specifying how many thinking tokens you wanted as budget was just silly. The rollout might not have been great, but the change is good.

And, is it really even an oligopoly anymore? Open weight models are incredibly competitive in every way; whether you want to use US providers, Chinese official providers, self host, etc.

Official tweet from Tibo from OpenAI confirming they ran experiments tweaking “juice” (reasoning effort mapping values) and have reverted them: https://x.com/thsottiaux/status/2076495156757577895

As he confirmed, for at least a period of time, and for some users, “Sol xhigh” was actually “Sol high”, etc.

I’m also often not affected by Claude outages on the API; but Claude.ai is down.

Super long chats are slow and sluggish too at least on chatgpt/claude.ai; even on a semi beefy machine.

I’m sure it’s lowkey intentional, probably encourages users to spin up new chats; hence less context.

Serving LLMs at Anthropic scale is very, very different. It’s not SGLang or vLLM.

If you’ve tried setting either of these up, you’ll know how various tricky settings can impact throughout and model output quality; and those are much simpler stacks.

Even homelabbers are getting into disaggregated compute; e.g. one GPU for prefill, another for decode.

Obviously Anthropic and co are using a mixture of GPUs and hardware and clusters; not everything is just GB300 or whatever; so you then get into hardware quirks, kernel optimisations that may deliver huge speedups at the cost of a tiny bit of KL divergence, etc.

And I believe they’ve publicly said they use TPUs for inference too, but I doubt exclusively; and I’m sure that’s well optimised too.

Finally, Google has publicly stated they intentionally and silently degrade/poison models in response to distillation attacks; who knows what the other companies do.

Inference isn’t flat 24x7, peak hours have more usage, but you buy/rent servers; not servers only for peak hours.

At their scale, you’d have to be setting money on fire if you’re not doing dynamic inference optimisations based on load.

API and consumer subscriptions are treated differently; all trackers measuring via API won’t notice this.

I think ADHD is a good parallel. I absolutely have it, I’m diagnosed and have had symptoms since childhood.

But even so, I often realise I self-justify not doing something and attribute it to my ADHD, excessively. It becomes a clutch, and I gravitate away from the willpower I do have.

That reminds me, it’s time to reach out to some friends and make weekend plans, even if I’d prefer having a quiet one. I’ve had too many recently :)

I think the examples you listed (doomscrolling, being sedentary, junk food) are related but distinct to ‘avoiding friction’. My perspective is it’s about quality, not the form; even if they correlate.

I don’t enjoy going to the gym, but I love swimming and hiking, and love that. I haven’t worked out in a gym for years, I don’t want to, and I’m quite healthy and reasonably active.

Some people are just introverts, and I think the ‘optimal’ amount of IRL time is different for every individual. Much like my gym vs swimming example; quality social interaction can take many forms; some of my closest and most meaningful friendships over decades have been based on games (voice calls; discord chats; hanging out in person sometimes and going on trips together sometimes), and that works for me; and I don’t think I’m missing anything.

Prioritise genuine human and social interaction, and for introverts, get out of the comfort zone once in a while (being a hermit is not good). But whatever form that takes is secondary to the quality and how it fits your personality and quirks.

For me, commits are the atomic block; and a good commit should be self-contained anyway; where no additional context is necessary. If an agent can't figure out what a commit is supposed to do, with only the commit title/description and diff, it's a bad commit, and this has always been true in software engineering. I do not pass prompts between the two ordinarily.

After claude or codex finishes a commit, I switch console tabs and ask the other to review it; with the commit ID. I often find it helpful to inject a bit of human knowledge, and callout any areas of attention I see from a quick skim. (But that could just be me wanting to not abstract myself away from software engineering that much :)

Sometimes, I do ask Codex to read my ~/.claude/; and vice-versa. But, generally, I try to keep as much knowledge (e.g. investigations, reports, deep dives) inside the git tree as possible; so that is not necessary.

I don't use skills, but I do have ~/.codex/AGENTS.md and ~/.claude/CLAUDE.md. These are high-level instructions for what (1) I consider readable, maintainable code and patterns, and (2) workarounds for empirically observed model behavior IOdon't like, such as Astra being a bit of a "over-correct over-validation nit-picker". I keep these human-authored, and update regularly based on what I find annoying.

I do all of this before I submit a PR; but of course, for trivial stuff (e.g. CSS changes, copy/string changes, etc), I don't bother.

My practices and workflows do change over time. Back in the ~Opus 4.5 days I'd often define a rubric/criteria in a markdown file, iterate with AI to improve it, and that's the "spec". I've stopped doing that since GPT 5.6; partly because models have gotten a lot better at understanding high level intent from the context; and partly because nearly all models these days feel 'gradermaxxed' when working like that.

Finally, consumer $100/$200mo subs get you _so_ far, I get a lot of value from both. I used to have multiple Claude subs for a while, but trying to 'get full value' made me work on projects just for the sake of it; so 2x$200/mo is my cap :)

There's far more than zero evidence, see https://arxiv.org/html/2402.08806v1 ; or the industry-standard practice of using multiple model families as LLM judges; or even 1P implementations https://code.claude.com/docs/en/advisor (which misses most of the benefit; since you want different model families, with different pretrains and posttrains).

The BBC of today is far from what it was a decade ago. So much trust and credibility has been lost.

Claude models don't go out globally either. Bedrock is fine. Azure is fine.