fp64

The house in Zurich I live in was initially built something like in the 12th century, but the floor I reside on was added in the 17th century. This is the case for most houses in the old town. So this is going on for a while.

31 cats from Ankara, Turkey, in this study. They point it out themselves in the "Limitation" section of the paper. Not sure what to think of it

Physical units help modeling and review, it's much easier to sanity check values from just reading, and comparing constants. I've heard that for rendering/raytracing, a lot of people move to model light intensities by actual physical values which makes it much easier to author and compare - such as for example "oh I want a light here that is as bright as this old lightbulb I like so much, but there's sunlight also shining through the window". That's much easier than estimating it in arbitrary units, in particular as here human perception is non-linear. I assume this would be similar for other simulations.

Meh, I have the impression the blog author really hates Zig's new Writer (fair, I disagree, but fair), but his criticism in this example is in my eyes slightly questionable -- it is a bug in the implementation and not a conceptual issue. He then uses quite some loaded phrasing like "I must be too dumb to understand this" and "I can't be really too dumb can I?" which I think ruin the discussion (as do the titles. He failed to convince me, for instance, that the new Writer was inherently unsafe by design). It feels like a "Look I told you!!! You run into bugs like this!!!" which is not helpful for a feature/refactor that was already advertised as complex and not fully implemented or verified.

Disclaimer: I'm a zig fanboy and do all my hobby stuff in it

Not to be confused with the Hokusai woodcut print titled "The dream of the fisherman's wife" (1814) (NSFW I guess?)

I strongly disagree. There are things far, far worse than JavaScript. I would even go so far as calling it "quite decent". I like to use it for prototyping and scripting quite a bit, it can be rather efficient and the "standard library" is very decent in my eyes. It has some footguns, and certainly used to have a couple more (that are discouraged now, but still people complain and call it bad because technically, you can still use the bad parts that any linter refuses). I even really like the idea of protoype-oriented programming and find it a bit sad we lost this in favor of classes, but I guess this actually makes the language a bit easier. Disclaimer: I am not a webdev, and if I do webdev, I use TypeScript. Personally, I consider e.g. Python far worse.

Not sure if this works out, from my understanding the $50k would not even remotely cover a hypothetical for-profit categorization (should only cover less than 500 annual users, while my understanding is that Hack Club is significantly larger, by some orders of magnitude).

I find it very difficult to inspect the email headers in Outlook, I think for the iOS app it's not even possible. It's almost like they want to make it less transparent and secure

So if it were indeed a for-profit account, it would also be okay to give them just a couple of days to "find the money" or otherwise lose 11y of history?

Well, there's legacy code and/or horrible git history that also needs fixing at some point. Also I have witnessed how the history can send you down a wrong path. I don't agree that this is a good argument.

I think you are right. At least they should document that it only works if you have netstat installed and using it on a device called "en0"

Cars primary purpose which the majority of people are using them for is not breaking the speed limit.

Remove defense from that list, they are the good guys now. I see you already omitted pharmaceutical industry so you're at least almost up to date

As I said in a different reply, I would not know how to address the engineering and management aspect, let alone the legal aspect, which are the biggest blockers and likely the reason this is not already done.

It's fine to develop "in the open" but this is a pitch for several million dollars, and handing that out without any credentials or track record is just not happening.

At least provide rough estimates for what is needed to get this done. What architecture? What training data? Where does the training happen? Who manages and administrates the cluster? Volunteers who try this the first time or paid experts? What solutions to failure recovery, to storage, to tracking and monitoring? Who has the last word on fundamental decisions? How will the legal component be handled? Do they already have a good law firm, how much would that cost? Will the first training be successful right away or how many iterations will be required? Can you even get access to the required GPUs at a reasonable price point? Train on older architecture? How much effort is required and planned to save cost by making training more efficient, ........

You should still not compromise on error reporting, for example. The user would not know if a failure occurs because it can't create the /tmp file, or the URL is wrong, or DNS failed, or the response was unexpected etc. These are things you can lose hours to troubleshooting and thus I would not fork it and make my own if I have to add all these things.

I also disagree that it's small but mighty, you popen curl that does the core task. I am not sure, but a bash script might come out even smaller (in particular if you compress it and make it self expanding)

I can only speak for myself and here it hits a nerve. Yes, the idea is fine, but the idea is only "crowdfund training", and not how this can be actually and practically implemented. It shows this exact ignorance of people who have no idea about anything but are sure that just with enough funding they can change the world, people just need to see how smart they are. I am not saying that this fully applies to OP, more that this is a regular occurance and can get people rather annoyed, at least it does for me, and thus lead to such harsh and blunt responses.

He "ships" a website with a gmail address and nothing substantial. I could do the same, likely better, in 20 minutes. How could I even tell this from a scam, there is nothing of substance. And the great idea is just obvious and all the painful details to make it work are completely ignored

What are you shipping here?

Training a model is not only expensive, but also technically challenging on a pure engineering level. Cluster management, storage, backups, access, fault recovery, and so forth. While crowdfunding training of a LLM is a nice idea, personally I would not invest in something this "uncooked". Why do you believe you are able to properly manage $5M+ and the infrastructure necessary? I've met several people that believe if you just get the funding the rest is super easy, and it's slightly infuriating.

Apart from that, I agree with the other comments that the website looks unprofessional and llama is a bad name to use for this.

If I would want to give this a shot, I would first get engineers committed with a plan to start as soon as there's funding, set up a non-profit to handle the operation, and make sure that potential investors get the impression I knew what I was doing by providing a full plan and timeline, including addressing the legal challenges (among those, make it clear that the resulting model will be commercially usable and not sued to death. Are you planning on guaranteeing indemnification or do you want to release the model as-is? Etc)

Why do you compress the executable? I mean this is a fun part for size limit competitions and malicious activities (upx often gets flagged as suspicious by a lot of anti virus, or at least it used to), but otherwise I do not see any advantage other than added complexity.

Also interesting that "ultra lightweight" here means no error reporting, barely checking, hardcoding, and magic values. At least using tty color escape codes, but checking if the terminalm supports them probably would have added too much complexity......

This is hilarious, can totally understand the fork and will take a look

I think in the article it was even stated that the port is even accessible without drilling!

Pretend that I have to pay to watch TV at home? I don't know, sounded fun to me in my head

Plus I meant for the manufacturer to "repair" it. Maybe tamperproof gets triggered if you accidentally drop the device in an unfortunate way

I was consistently tempted to give emacs a fair try but after now more than 20 years of vim it got more and more difficult to even imagine betraying it. Thankfully, now with neovim I can spend hours in my config files, and with fennel I can even lisp.

Was my guess as well, maybe it's even possible to use it to flash new keys so the device can be used again?

Now I am curious if I can find a terminal myself, if they are actually getting phased out it might not be too difficult to find a used one...

I understand what you are saying, but how am I supposed to know the reasoning behind it if for certain reasons I am not supposed to know it?

In this case, I do not care why he had changed his name, I read the beginning of the article and it did not explain why this is something that needs discussion, and stopped reading when it just stated his birth name.

I thought it was not proper to discuss names of people they do not like to use anymore?

What’s another example of different rules depending on size? “Content Moderation” I think? Anything outside the DMA? I think small companies get some exceptions for documentation requirements?

Yes, exactly! Thank you!

A long time ago there was this “web of trust”, I don’t think it exists anymore. Was one of the big CA and you could get different certificates through some form of vouching, I think it even went as far as meeting people to show your ID and then they sign you or something. As it was run by a big CA, not really distributed but IIRC they kept their involvement minimal. It’s been a long time but if you’re curious maybe look into that

I am not sure what your argument is supposed to be. That I am wrong because I haven’t gotten the hang of it after all these years? I do understand generators. For a long time. I still believe they are usually an anti-pattern, same like massive abuse of inheritance where you have 20 classes but only a single instance but you made your code “future proof”.

The point I wanted to make is that using generator, in particular like here, is something that I consider ugly and difficult to maintain and it will probably have to be rewritten when trying to export to TorchScript. I really do not see how “just get a hang for it” can help me reevaluate my perspective

Edit: Maybe you were hung up on my “standardised” - I have to admit I do not know how thoroughly the PEP defines generators and if really all edge cases are defined without needing to check the python source code. From past experiences, my trust in python language standards is a bit shaky, as it had been difficult to reproduce the very exact behaviour using a different language, or python features - without requiring digging through the sources.