jval43

It's a common requirement. Chinese models in general, regardless of where they're hosted. The fear is backdoors, time bombs, or other unexpected behavior.

And these aren't technical discussions really, more of a risk management / political requirement thing.

I want to see your 2-3 package managers :-) Unless you count apt, aptitude, apt-get etc separately.

One saving grace with Linux distros is that usually the maintainers package versions that are compatible with each other. So if you stay with the default repos things work.

But yeah once you install newer versions from non-distro repos it might be fine (e.g. git) or things will break in subtle ways.

I can't believe how this opinion got so popular, because I disagree completely with the Bash bashing.

I totally get it for people who only ever used .BAT files (our entire IT department at work). It must feel like a godsend to them. Bash of course never worked for these people; running it on Windows via MinGW/Cygwin is just a bad experience.

But the relentless bashing of Bash itself and the longstanding Linux (and Unix) ecosystem alternatives in general is unjustified in my opinion.

Bash itself is elegant, powerful and concise, and just works on Linux. You can write spaghetti code, but you don't have to. Readable scripts are just as easy. "Skill issue" if you will.

And yes, Bash-isms exist, and I got reminded of that fact when Ubuntu shipped Dash as the default /bin/sh shell for a while or when I'm on macOS. But it's really not a huge deal: just use the real !#/bin/bash and ignore the "um actually in Posix..." complainers on the internet.

I do appreciate that Windows finally has a capable scripting language, and while I don't like the lengthy syntax / cmdlets, I also see some benefits. But Bash is still great nonetheless.

I recently generated a "plain English" document out of a huge technical spec, and steering this process was the main time sink.

Opus 5.5 will make broad changes in meaning and scope, but subtly. A human copy editor would at least pause before making claims unsupported by evidence (unless they're marketing); an LLM will happily fill in the gaps. Adversarial checks are helpful but have limits.

Unfortunately, I believe most people's command of written language is simply limited (including their native tongue), so they genuinely don't notice or care.

Using a copy editor in the way you describe is sensible, but I suspect many people will just do one pass and blindly apply every suggestion.

"Height-adjustable gear lever", lol.

Zero info about the engine on the page - or is it electric? The page doesn't really say.

For a sports car, that sort of info is kinda important?

tzdata routinely issues historical corrections in newer versions.

So while the political time zones don't change retroactively, the technical rules used in computers can.

Oh they tried: https://news.ycombinator.com/item?id=26957215

Sometimes it won't even do a search if the query is too detailed or too long - in my case GTIN/EAN numbers.

No, I actually want to search for that whole string, it's not gibberish!

It was so much faster you could not believe it. It was a revelation and even though "degoogling" was already a thing, Chrome was so much better than the rest that you just had to use it. Tabs actually worked without crashing your PC.

The risks were clear from day one but Google and Chrome were great to both users and devs, and it stayed like that for a long time.

Google Chrome did this in 2008. Full-text search over all visited pages, stored offline. It was very useful and I miss it.

Nobody seems to remember it, even though it was a headline feature. Was removed in 2013, I think due to technical constraints.

Will definitely try this.

No. If you're the leader you should know what you're leading. It's literally the job.

But the problem is right there: "non-technical" leaders. Never work for one if you can avoid it.

Reading that, the execution is now deterministic but the soundness is still probabilistic and <100%, no? Unless I'm reading it wrong.

It's not the academics writing these LLM hype posts.

You wouldn't announce your LLM has solved the daily crossword-puzzle.

Academics actually studying that book would likely focus on the historic or otherwise interesting aspects of the works, instead of the small puzzle at the end. Nevermind publishing or announcing the result in this "major-breakthrough" way. Maybe it was solved already but buried as a footnote somewhere.

Book-ciphers are well-known: https://en.wikipedia.org/wiki/Book_cipher

No pictures from 1832, but I found a photo describing the phenomenon. So low-res one can barely see the effect: https://onehundredmountains.blogspot.com/2008/04/sunrise-on-...

It's standard cryptography, not some new-fangled tech! Wikipedia even has some easy examples:

https://en.wikipedia.org/wiki/Zero-knowledge_proof

Main caveat is that ZKPs are probabilistic. The protocol (number of rounds etc) determines how sure, e.g. 99.9%. But never 100%.

Second caveat: tech- and crypto-bros play fast and loose with the term "ZKP", either because they don't know any better (marketing) or they straight up lie. Whether any application you run actually uses ZKP (or any other cryptography scheme) is unknown unless you have the source code.

Site is down. This should be the same PDF I think: http://wexler.free.fr/library/files/necker%20(1832)%20observ...

I disagree. First we got everything as HTML/Electron apps, now the trend is everything as TUIs. Cross-platform, better, but still worse somehow.

What falls to the wayside are the actual and real productivity and usability improvements afforded by "real" GUI apps over the last 40-50 years.

Don't get me wrong I'm a huge terminal user and absolutely love the clarity of a terminal and flexibility of a shell. But if I have to "learn" to use a TUI app - and they're not as simple to use as their authors might believe - I might as well learn to use the full GUI app with all the niceties available because of it.

It's entirely possible future models will refuse to answer if they notice you're using an external harness.

Instead they'll point you to their respective vendor APIs for the specific use cases.

We're talking about trillions in value to be captured, they'll try everything.

Especially Apple. Buying a first gen Apple product hasn't been great since like 1999 at least. Very few exceptions.

Play the Gameboy version. I bought every single official Tetris version and they all suck in conparison.

Gameboy emulators are now available again in the iOS app store or even as a a web app. Or just play on real hardware.

It was so close! I could already see the end of all the predatory publishing and flourishing of open-access. Some even required by law. Knowledge was finally going to be free.

But no, it will presumably get much worse as LLMs are inserted into this equation as yet another and new gatekeeper.

(Disclaimer: I have publications with ACM, non open-access. And ACM wasn't even too bad, it's the others that give me pause.)

My ACM papers are published in full on my website. Not preprints. In fact thats the only reason I have a website, just in case someone actually wants to read them.

The copyright agreement (actually copyright assignment/transfer) with ACM has had a carve-out for this case in it for a while, but it has to be a personal website.

A paperclip always worked, even on the slot loading iMacs.

Impressive debugging skills, and thank you for the benchmarks. Now I'm wondering if mlx-engine / mlx-lm have these bugs too.

One minor thing: as you are concerned with honest numbers, the graphs should be logarithmic on the y-axis too (like they are on the x-axis). Otherwise it's hard to see whether the curve is sublinear or linear.

Very carefully - and I mean extremely carefully - word it so that it describes what it does, while implying what it should do without confirming that it actually does that.

Also helps if you fix the bug or change the behavior, the docs are still technically correct. I'm only partially kidding, I swear I've seen this a million times in documentation I read.

This is one of the reasons I currently use Gemini for daily use and research.

Google has lots of experience with search history, and presumably handles this better than new companies.

There was this article recently, which I personally found interesting:

https://news.ycombinator.com/item?id=47208573 Inside the M4 Apple Neural Engine, Part 1: Reverse Engineering (maderix.substack.com) 376 points | 3 months ago | 122 comments

Same for me, when riding the bus. Really wanted it to work and tried over the course of several weeks, unfortunately it didn't help.

Thank you, that's a good response and I agree with many of your points.