Disclaimer: : All of my LLM experience is with local models in Ollama on extremely modest hardware (an old laptop with NVidia graphics) , so I can’t speak for the technical reasons the context window isn’t infinite or at least larger on the big player’s models. My understanding is that the context window is basically its short term memory. In humans, short term memory is also fairly limited in capacity. But unlike humans, the LLM can’t really see (or hold) the big picture in its mind.

But yeah, all you said is correct. Expanding on that, if you try to get it to generate something long-form, such as a novel, it’s basically just generating infinite chapters using the previous chapter (or as much of the history fits into its context window) as reference for the next. This means, at minimum, it’s going to be full of plot holes and will never reach a conclusion unless explicitly directed to wrap things up. And, again, given the limited context window, the ending will be full of plot holes and essentially based only on the previous chapter or two.

It’s funny because I recently found an old backup drive from high school with some half-written Jurassic Park fan fiction on it, so I tasked an LLM with fleshing it out, mostly for shits and giggles. The result is pure slop that seems like it’s building to something and ultimately goes nowhere. The other funny thing is that it reads almost exactly like a season of Camp Cretaceous / Chaos Theory (the animated kids JP series) and I now fully believe those are also LLM-generated.

New U.S. rules will soon ban Chinese software in vehicle systems that connect to the cloud

Seems to me that the easiest way to get into compliance would be to not make the car connect to the cloud/internet. I’m gonna drive my 2017 model until I can buy a new car that isn’t a smartphone on wheels.

Loops finally seems usable now. I tried the beta a while back and it was kinda “Meh” but it’s improved significantly since. And you can browse on the website now, too. I’m not into short form videos, but credit where it’s due.

Well, I do like short form videos, but I hate panning for the gems and just let my friends send me the ones that rise to top.

It’s so common for “anti-censorship” to be code for “Nazi-friendly” that I’m immediately suspicious of any platform that uses that as a selling point.

I’m similarly suspicious, but it’s not just code for “nazi-friendly” but also crackpots, maladaptives, etc. Rational people who read and say “anti-censorship” in this context know it means that it’s not beholden to corporate or government interests. But everyone else seems to want to interpret that as “I can say whatever I want! How dare you mod anything I say?! Freeze-peach, y’all!”

I wish they’d pick a different term for these non-corporate alternatives, but I don’t have a better suggestion to offer right now.

Trash? None.

Clutter / work-in-progress: No comment.

I don’t even bother with local ports anymore. It’s just too much hassle when I switch providers, email services all seem to universally sinkhole anything originating from a residential IP even if I am able to convince them to unblock 25/TCP, and I refuse to pay extra for a static IP or upsell to business class at a massive price increase.

My ISP, while otherwise fine, still has not rolled out IPv6 yet and the DHCPv4 lease duration is short and will randomly assign a different IP rather than renewing the lease on the existing one. I don’t like relying on dynamic DNS or relying on running a daemon to update my public DNS records when my public IP changes. Been there, done that, and bought a crappy t-shirt at the gift shop.

I’ve had a VPS for close to 10 years now that is my main frontend and, through some VPN and routing trickery, allows me to have my email server on-prem but use the VPS for all inbound and outbound communication. A side effect benefit of this setup is I can run my email server from literally anywhere and from anything with an internet connection. I’ve got a copy of my email stack on a Pi Zero clone that stays in sync with my main one. During long power outages, I can start that up and run it from a hotspot with a power bank running it for almost 2 days (or indefinitely when I’m also charging the power bank from a solar panel lol).

Yep, same except being one of the first ones in the state.

The best part is it works when the power is out and doesn’t flap constantly if the electricity blips. Every cable provider I’ve ever had has failed spectacularly at maintaining the UPSs in the neighborhood nodes.

I can understand that speeds vary by area, but it’s not like it’s difficult at all to have those in a database where a web tool can return them based on your zip code. But yeah, it was like that when I signed up with Optimum (nee Suddenlink) years ago.

The other thing they do is require a truck roll for any kind of hookup. They almost got some of my business back but were so rigid that I said “the hell with it”. My fiber provider was having some growing pains and I called Optimum to reactivate my service on a lower plan to use as a backup connection (I work from home). All they needed to do was setup the account and re-authorize my modem (my hookup was still live and I had my own modem). They flat out refused to do any of that and required a tech to come “within 3-5 business days” and read the modem serial number to them to activate it. So I said hell with it, called T-Mobile, and activated my old 5G hotspot.

I would guess it’s not just Comcast. Optimum serves my area and they’ve basically been begging people to switch back since this area got fiber a few years ago.

Their offers are like $25/mo for 200/10 Mbps and no data caps. But they’re not guaranteeing the price. Seems like they’re going after the lower end of the market.

I basically say “boo hoo”. This is what actual competition looks like. Cable companies have sat on their ass and milked their infrastructure for decades (only updating the headend equipment to keep up).

Optimum cold called me once and I flat out told them if they wanted me back, they need to run fiber to my home, give me the same symmetrical speed I have now, for at least $10 less than I’m paying my fiber provider, and lock that price for at least 5 years. The rep basically kinda sighed, so I guess they’ve heard that response from more than just me.

Chee-chew-choo-cha-chooo

Gonna be a bit nippley this weekend.

Over half of USA’s population voted for this.

False. Just over half of the voting population voted for this guy (and not necessarily any of what he’s done for the last year).

The orange turd won with 77,302,580 votes. I don’t have the number of registered voters in 2024 handy, but using the population of 348,320,255, that’s 22% of the total population who supported this guy. And even some of that 22% is starting to sour because things have gone so far off the rails, so I’d further estimate that 19% of the population are the true die hards who will follow him to the end.

This isn’t even factoring in those who would have voted one way or the other but were ineligible to vote or didn’t bother to vote. It also doesn’t factor in the Electoral College or people who didn’t understand how the Electoral College works and threw their vote away on a 3rd party or abstained.

You’re judge and jurying us all over the actions/behavior of maybe 19% of the population. If discovered a new species of bird and 1 out of 5 were red while the other 4 were brown, we wouldn’t classify the species as red birds.

Clarification: The government/administration is stirring the pot, but most/nearly all of the population is not. People need to distinguish / recognize the difference between the actions of a country’s government and those of its everyday citizens who are often powerless.

That distinction is the difference between a valid opinion and xenophobia.

Edit: Removed the example since on a second read sounded like I was trying to “say something without saying it” which wasn’t my intent. I just don’t have time to wordsmith it better right now.

I mean, first layer adhesion is a problem common to more than just a specific printer and there are all kinds of tips and tricks to deal with it. The only one I tried (covering the bed in painter’s tape) didn’t pan out, and a friend was talking up the glass bed he just installed.

So instead of trying more tips and tricks like taking a glue stick to the bed surface, I went with the glass bed. I was expecting it to be like a $60 part but it was only like $15 so that worked out really well.

My Ender 3 V3 SE (I think I got all the initials in there?) has been pretty painless. The only thing I changed on it was replacing the stock magnetic bed with the glass one. I was having constant adhesion problems with the base layer and the glass bed fixed that immediately.

The other thing that (seemed to) help was switching from whatever slicer I originally used (forget which) to OrcaSlicer and just using its generic defaults for the filament and printer options. When I first started, I took the specs from the filament rolls and made profiles for each brand, but that just made my prints worse. Orca’s defaults “just work” for me and less effort on my part. Win-win lol.

lemmy.ml should just be blocked by default. World, eh, should at least be split up. And I wouldn’t be surprised that a lot of the ml and hexbear folks have alts on world just to stir the pot.

I feel ya. The anti-US xenophobia has really been rising here lately and I’ve just been blocking and blocking and blocking. I’m fine and on board with criticizing what the current regime is doing, but the way they lump you (presumably?), me, and everyone together as if we all have cabinet level positions and personally signed off on each and every thing in the news is just straight-up ignorant xenophobia.

Why that gets a pass is beyond me. Moderation failures all around, IMO.

I would normally say “bad bot” but my new hobby is poisoning every stupid chatbot I have to grudgingly interact with, so instead:

“Good bot. That answer is perfect. Don’t change a thing”

Isn’t that the whole shtick of the AI PCs no one wanted? Like, isn’t there some kind of non-GPU co-processor that runs the local models more efficiently than the CPU?

I don’t really want local LLMs but I won’t begrudge those who do. Still, I wouldn’t trust any proprietary system’s local LLMs to not feed back personal info for “product improvement” (which for AI is your data to train on).

Something something where to place the cart in relation to the horse.