It just so happens that many video codecs are based on image formats, so ffmpeg already has a lot of the complex machinery to do so available to also implement these image formats - internally it can just handle it as a single frame of video with specialized formats for that.

Imagemagick (and other tools) also work, but why use multiple pieces of software if you already is adequate? ImageMagick is also software, and can also have vurnabilities.

You would be inventing some style of UUID. Include a timestamp in front, so that it is sortable and you have Snowflake.

That’s fair. I actually don’t think we disagree that much - I just think I have trouble conveying what I am trying to say. Whenever someone talks about ‘shallow statistical predictions’, I think about older techniques like Statistical Machine Translation which even had trouble with things like word order, LLMs handle text on a higher level of abstraction (which I described as a form of textual understanding) - and hence handle things like word order better - but are still inherently statistical predictors. The model stores info about how words interact and relate to one another, but it does not ‘understand’ what the words actually (physically?) represent beyond these interactions nor does it ‘understand’ what it is doing. Albeit, those interactions are modeled well enough to give a convincing replica of doing so.

That is no more understanding the problem than my spelling or grammar checking understands the comment I’m writing.

My general point would be that even a grammar checking can have some form of understanding of the text, no matter how shallow. The checker probably has a rule for when a is used versus an, if this rule generalizes across new words that rule that were previously unseen is a form of ‘understanding’ of the language being used in my view, despite being overly simplistic, while rote memorization - having a list of words that are followed by an - may not be.

LLMs are a weird case, because their internal representations for many concepts generalize even across new settings / inputs - in that sense the model has a form of understanding of what is being given, while for many other concepts the patterns break down even in the simplest of cases. It may ‘know’ that the preceding text is the writing of an essay - and it should autocomplete accounting for that -as is defined in its weights, but this understanding is shallow - it does not know why it knows, or how it knows - it cannot self reflect as it does not see and understand its own internal workings, or account for that. Yet the internal representations represent a form of text understanding that can be useful nonetheless - it is a language model after all.

My comment was intended to show this duality, hence the duality between the two paragraphs.

And this is the crux of my beef, I think, because stochastic pattern matching is not understanding, it’s a mathematical representation of how the model processes your input tokens. The fact that it has to start over every time you provide it input, and uses the previous input/output tokens as context is why this is not ‘understanding’, it’s just fancy token prediction that gives a middling-to-passable facsimile to intelligence and understanding things.

The problems you note in your second paragraph fundamentally undermine the argument that there is any form of understanding to the AI, because those are basic mistakes that a trivial understanding of the problem would prevent.

I am not entirely grasping the point you are trying to make here. I am certainly not arguing that it is conscious, self-aware, or in any way not a mechanical procedure that is being performed (I would not argue for that!). My key point is that it is not a simple black or white it understands / it does not understand - it may have internal representations that relate many concepts together, allowing it to draw upon these links when generating text, giving it a certain semantic understanding of the language and text it is using, while simultaneously not having a bit of self-awareness.

These models do have some form of understanding though. There are features for bugs and typos, and general features that map descriptions and pieces of code. It understands the code in so far it helps with next token prediction.

The bigger problem is that these language models are inherently unreliable and stochastic in how they generate. You request a feature - and it destroys something else in the process - because a single incorrect prediction caused it to diverge and skip a portion of the original code. You request a small modification - and it decides to restart from scratch - because random sampling made a different way of doing something more likely rather than what was there already. Errors compound and the model has no way of fixing or correcting them.

It is not just him, but also the people he is trying to reach. Even if he gets assistance & the resources to set up an independent platform for this, he still has to get people to use it - which is a much higher bar to meet.

20 minutes for 4km would be 12km/h, which would really be taking your time. 15km/h is the slowest I would usually go when taking my time, and that would take about 16 minutes to cover a 4km distance.

A single plane emits a lot, but if you are not using a private jet - it enjoys the same benefit as other forms of ‘public transport’: emissions are shared among multiple passengers. Especially if you are in economy, in a packed plane there are a lot of people to correct for that.

Furthermore, emissions for planes are far from uniform for the distance travelled: a disproportionate amount of emissions occurs as take-off.

A short or long haul flight is more efficient than an ICE car [source]. This concerns efficiency though - not total emissions - so if you use a car, but travel a shorter distance, emissions will still be less. Also, this source is from 2023, things will probably have shifted around a bit.

The BSODs I’ve had on Windows in recent years have all been due to faulty hardware. Once that was resolved I’ve had no crashes. Can’t really blame the OS for that.

Or better: abolish the concept of districts entirely. Can’t gerrymander what doesn’t exist.

Good ol’ Goodhart’s law.

Going off the ycombinator discussion too, it certainly looks like it… I was hoping for some announcement for increased investment into an open source ecosystem, got an ad for a shady whitelabeled nextcloud host instead.

Here in the Netherlands it is 2.08€/L (9.10 U.S. dollars / US gallon) at its cheapest and 2.48€/L (10.84 U.S. dollars / US gallon) at its most expensive. While a lot of that is taxes, we are well past that mark.

Most LLM chatbots don’t push back when they should. When combined with situations like these, at a large scale, even 5 percent is abysmal, let alone 55 percent.

To be similarly pedantic: Ctrl+C is a hotkey that sends the corresponding ASCII code / codepoint to signal something, it is not an ASCII code itself.

You could have the same character be sent by using Ctrl+Q (if you were to remap it), and not break compatibility with other processes while doing so: the codepoint being sent would be the same. From a technological perspective there is nothing special about the key combination Ctrl+C specifically, but altering this behavior in a terminal absolutely wreak havoc on the muscle memory of terminal users, and altering it’s behavior in a text editor on everyone else’s.

The key issue is that the request is to change behavior in one place (browser) to match that of a rare case (terminal), causing a mismatch with the frequent case (office suites, mail programs, …). The terminal is the odd one out, not the browser, and ought be the one to change the default for the reason you provide.

In practice, a terminal is a special case and not just a text input window, and current convention is that Ctrl + C aborts / cancels.

(You could of course have a duplicate hotkey, but now you are inconsistent w.r.t. other browsers, and there will be someone else who will be annoyed by the difference)