One of the first things they teach you in Experimental Physics is that you can’t derive a curve from just 2 data points.

You can just as easilly fit an exponential growth curve to 2 points like that one 20% above the other, as you can a a sinusoidal curve, a linear one, an inverse square curve (that actually grows to a peak and then eventually goes down again) and any of the many curves were growth has ever diminishing returns and can’t go beyond a certain point (literally “with a limit”)

I think the point that many are making is that LLM growth in precision is the latter kind of curve: growing but ever slower and tending to a limit which is much less than 100%. It might even be like more like the inverse square one (in that it might actually go down) if the output of LLM models ends up poluting the training sets of the models, which is a real risk.

You showing that there was some growth between two versions of GPT (so, 2 data points, a before and an after) doesn’t disprove this hypotesis. I doesn’t prove it either: as I said, 2 data points aren’t enough to derive a curve.

If you do look at the past growth of precision for LLMs, whilst improvement is still happening, the rate of improvement has been going down, which does support the idea that there is a limit to how good they can get.

At some point in my career I’ve actually designed mission critical high performance distributed server systems for a living, so I’m well aware of that.

You can still pack thousands of users per server and have very low latency as long as you use the right architecture for it (it’s mainly done with in-memory caching and load balancing) when you’re accessing gigantic datasets which far exceed the data space of a game where the actual shared data space is miniscule since all clients share a local copy of most of the dataspace - i.e. the game level they’re playing in - and even with the most insane anti-cheat logic that checks every piece of data coming in from the user side against a server-side copy of the “game level data space” it’s still but a fraction of the shared data space in equivalent situations in the corporate world, plus it tends to be easilly partitionable data (i.e. even in MMORG with a single fully open massive playing space, players only affect limited areas of the entire game space so you don’t really need to check the actions of a player against the data of all other players).

Also keep in mind that all the static (never changing or slow changing stuff) like achievements or immutable level configuration can still be served with “normal” latencies.

Further the kind LVL1 ISP that provides network access for companies like Sony servicing millions of users already has more than good enough latency in their normal service and hence Sony needs not pay extra for “low latency”.

Anyways, you do make a good and valid point, it’s just that IMHO that’s the kind of thing that pushes the running costs per-player-month from one dollar cents or less to, at most (and this is likely quite a large overestimation), a dollar per-player-month unless they only have tens of players per-server (which would be insane and they should fire their systems designers if that’s the case).

After over 3 decades as a gamer and tech user this is maybe the single most consistent important benefit for any open platform were you can just install Linux.

The rest is nice but this one means that 10 or 20 years from now your hardware might have been repurposed for something else and still be useful and in use whilst a closed platform will just be more junk in a junkyard or sitting in a box of those things you’ve kept just because you don’t like to throw expensive stuff away but will in practice never use again.

They’re stupidly cheap to operate per user when you have millions of them, which is how companies like Facebook manage to make a profit from merely showing adverts to users and with no subscription fees.

Remember that Sony gets a cut from games being distributed to their platform, so online fees are just them double dipping for extra profits.

They’re stupidly cheap to operate per user when you have millions of them, which is how companies like Facebook manage to make a profit from merely showing adverts to users and with no subscription fees.

I have an Orange PI Pro 5 16GB on a box that smoothly runs a full blown Ubuntu Desktop version and would fit in a pocket though it’s maybe a little too thick (from memory the box it’s about 3x5x2 cm).

Total cost was about $170.

The board itself would fit a thinner box, but you might have to 3D print one.

Mind you, a N100 Mini-PC that costs the same is even more capable as a Linux Desktop, but it’s significantly larger and will definitely not fit a pocket.

You can find cheaper SBCs capable of running a Desktop Ubuntu but in my experience (with a $35 Banana Pi P2-Zero) if you go too far down the price scale Desktop Linux performance stops being smooth, even if the board is a tiny thing.

It was actually quite surprising for me recently when I found out some of these things are perfectly capable Linux Desktops.

Playing Accounting Tycoon on the Steam Deck.

What’s there not to like?!

“Family friendly UI” is “ultra-advanced” stuff for me: remember, before Kodi on a Mini-PC in my living room (and, by the way, I got a remote control for it too) I had been using first generation Media Players with file-browser interfaces to chose files from remote shares on a NAS, so merelly having something with the concept of a media library, tracking of watched status and pretty pictures automatically fetched from the Internet is a giant leap forward ;)

There are downsides to being an old Techie using all sorts of (what was then) non-mainstream tech since back in the 90s. I’m just happy Kodi solved my problem of having an old Media Player hanging together with duct-tape, spit and prayers.

That said I can see how Kodi having all status (such as watched/not-watched tracking) be per-media rather than per (user + media) isn’t really good for families. More broadly the thing doesn’t even seem to have the concept of a user.

I haven’t tried Jellyfin but people’s talk of it doing transcoding (which Kodi doesn’t need to do as it simply decodes the video stream and shows it on the video output) leaves me with the idea that it’s not quite the same and does things I don’t really need.

That would be Kodi which I now use on a Mini-PC with Lubunto which has replaced my TV Box and my Media Player (plus that Mini-PC also replaces a bunch of other things and even added some new things).

Before I went down a rabbit whole of trying to replace my really old Asus Media Player (which was so old that its remote was broken and I replaced it with my own custom electronics + software solution so that I could remote control that Media Player from an Android app I made running on my tablet) which eventually ended up with Kodi on a Linux Mini-PC also replacing my TV box, I had no idea Kodi even existed and was just using the old Media Player to browse directories with video files in a remote share (hosted on a hacked NAS on my router, a functionality which is now on that Mini-PC which even supports a newer and much faster SMB protocol) using a file browser user interface to play those files.

It was quite the leap from that early 00s file browser interface to chose files to play on TV to a modern “media library” interface covering all sorts of media including live TV (why it ended up also replacing my TV box).

How much would that be in Libraries Of Congress if written down?

I disagree: as a Bachus of Programming I have successfully managed to at the same time both be a God of Programming AND having no clue what’s going on.

(The real joke behind the joke is that today I’m doing Shader programming so that’s quite close to reality and I could definitelly do with large amounts of wine or at least beer).

Here too the Schrödinger’s equations apply: a programmer’s state during coding is a superposition of both of these states until actually trying to run the code, at which point it collapses into one of the two states.

Death by being SWATEed after a Github commit.

The amount of effort I do to try and avoid using double parentesis is trully herculean.

I think that stuff is the product of a completionist/perfectionist mindset - as one is writting, important details/context related to the main train of thought pop-up in one’s mind and as one is writting those, important details/context related to the other details/context pop-up in one’s mind (and the tendency is to keep going down the rabbit hole of details/context on details/context).

You get this very noticeably with people who during a conversation go out on a tangent and often even end up losing the train of thought of the main conversation (a tendecy I definitelly have) since one doesn’t get a chance to go back and re-read, reorganise and correct during a spoken conversation.

Personally I don’t think it’s an actual quality (sorry to all upvoters) as it indicates a disorganised mind. It is however the kind of thing one overcomes with experience and I bet Mr Torvalds himself is mostly beyond it by now.

It was about as prescient as “640k is enough for everybody”, but in a good way.

In a highly simplified way:

• Think of Windows as an electricity provider with their own specially shaped wall socket.
• Linux is also an electricity provider with a differently shaped wall socket.
• In this metaphor Wine is just some guys providing an adaptor that makes the electricity of the Linux electricity provider available in a wall socket that has the same shape as the Windows provider’s.

Wine isn’t breaking Windows copyright because it doesn’t copy any of the Windows internals: instead it provides the contact points with the right “shape” for programs which were made to work in Windows to connect to to get their needs fullfilled, and then internally Wine does its own thing which is mainly using the Linux under it to do the heavy lifting.

Mind you, this simplification seriously understates just how complicate it is to implement what was implemented in Wine because the Windows interface is a lot more that just the shape of a wall socket.

I got an Orange Pi 5 Plus to play with smallish AIs (because it has an NPU) and I normally access it remotely, so I have to know its IP address to do it.

In order to easilly know the IP address of it, I’ve wired a little 128x64 monochrome OLED screen to it (Orange PIs, like Raspberry PIs have a pin connector giving access to GPIO and interfaces like I2C, Serial and SPI) which talks via I2C.

Turns out those interfactes aren’t active in Linux by default (I.e. no /dev/i2c-x), so I figured out that I had to add a kernel overlay to activate that specific interface (unlike with the Raspberry PI whose Linux version has a neat program for doing it, in the Orange Pi you have to know how the low level details of activating those things), which I did.

To actually render characters on that screen I went with an ARM Linux port of a graphics library for those screens I used before with Arduino, called u8g2)

Then I made a program in C that just scans all network interfaces and prints their names and IP addresses on that screen, and installed it as a Cron job running once a minute.

Now, as it turns out when you shutdown your Linux on that board, if you don’t disconnect it from power there is actually still power flowing through the pin connector to any devices you wire there, so after shutdown my screen would remain ON and showing the last thing I had put there, but because the OS was down it would naturally not get updated.

So the last thing I did was another small C program which just sends to that screen the command for it to go into power saving mode, shutting it down. This program was then installed as a Systemd Service to run when Linux is shutting down.

The result is now that there is a little screen hanging from the box were I put this board with Linux which lists its IP addresses and the info is updated if it connects other interfaces or reconnects and gets a new IP address. Curiously I’ve actually been using that feature because it’s genuinely useful, not just a funny little project.

In my country we have a saying: You can’t please both Greeks and Troyans. (Which, by the way, should be Athenians and Troyans to be Historically correct).

The point being that it’s impossible to please everybody all the time, so either there is no point in even trying or if you really care that much about pleasing people you have to pick which ones you want to please.

Further, for me it helps that I put a lot of value in Honesty, so I have almost no tendency to be fake or bullshit to try and please people, and dislike it when others do it to try and please me (and nowadays I am pretty good at detecting fakery) - I would much rather have people give it to me straight than try to bullshit me to “please me” (they’re not even doing it because of me: it’s generally done either as conflict avoidance strategy or trying to get people’s goodwill or sympathy to later extract some personal gains out of it)

“Will horrors never cease?!”