There’s a lot to cover here but I’ll try to touch on each point:

The key requirement is fast memory that can be addressed by your GPU, and ideally a lot of it - hence the insane cost of this hardware right now.

Remember that you need space for the model’s weights (think of this as its ‘knowledge base’) and the context window, which is basically the data needed for the LLM to keep track of your current conversation with it (effectively its short term memory).

With smaller pools of VRAM (8-16gb) you will have to compromise and either have a more capable model that will lose context quickly and start hallucinating, or a less capable model that can maintain a session for a bit longer but overall less ‘smart’.

For software - there are a couple of options for running the LLM itself, Llama.cpp is one of the more popular tools and is the one that I use. It has a web UI with the usual chat interface, and also exposes an API that you can plug other tools (e.g. opencode) into, depending on your use case.

In terms of hardware recommendations, at 20GB+ of VRAM you do have a bit more headroom compared to more consumer grade GPUs, but to be honest the most cost effective way to get a shitload of VRAM is likely not with a dedicated GPU but actually using a system based around a recent APU.

I got a Minisforum MS-S1 last year for exactly this purpose. It is based on AMD’s Strix Halo platform which it has in common with the Framework Desktop and a couple of other similar devices.

It has 128gb of unified RAM which can be divided between the GPU and CPU however you like, so plenty of capacity for even fairly chunky models. It also uses a tiny amount of power compared to a more traditional system with a dedicated GPU, while also giving really reasonable performance for most AI workloads, more than enough for use in a homelab.

For cloud rental - doable, but pricing is a factor, and of course this will not actually be running locally.

Usability - manage your expectations, but overall for a lot of use cases and of course depending on the model that you are running and the resources you throw at it, it can be comparable with especially older iterations of ChatGPT, Gemini etc.

But remember, you are not a Google or an Anthropic and do not have an infinite pool of compute to throw at your model, nor do you have access to the specific models they are using.

I’m not sure if this is triggering either thalassophobia or submechanophobia but either way thanks I hate it

Same, I set it up a few years ago and both me and my partner have been using it since then with no issues at all, it’s completely replaced Google Photos for us.

We’ve also set up immich-frame and repurposed an old Google Nest hub to use as a digital photo frame.

This might not actually be the first one, but one of the earliest games I definitely remember actually buying with my own money was Geoff Crammond’s Grand Prix 2. I would have been around 7 or so.

Definitely worth it, great game and the demos on the CD introduced me to Transport Tycoon, and the XCom and Worms franchises - and things kind of snowballed from there!

Ferrari F1: I sleep

Ferrari WEC: real shit

He is the manifestation of the ‘I made this’ meme. Among other things.

You are assuming that they have a personality to begin with.

I have witnessed companies make this exact mistake before - they have a legacy system written in $LanguageA that they either cannot find developers to maintain, believe is badly written, or does not support some new feature they want to implement (or some combination of the three) - and decide to solve this by taking the existing codebase and porting/transpiling it to $LanguageB (which is more modern, performant, is easy to hire developers for, etc) - without actually rewriting or rearchitecting anything.

What they are actually doing is substituting one kind of tech debt for another. The existing code that was poorly written and/or not well understood is now just bad code written in a different language. Fixing bugs or implementing new features now takes just as long, if not longer to account for the idiosyncrasies of how the code was ported.

And now this is being done by AI with even less oversight than usual? Recipe for a maintenance disaster.

As others have said, 100% a leak.

I would advise to stand on a chair or stepladder underneath the ceiling and check to see if it is still level. If you see an obvious deformation around the stain, this will be being caused by water pooling on top of the ceiling plasterboard. In which case, once the leak is sorted, you will likely need to drain the pooled water, cut out the damaged section, replace it, then replaster and repaint.

We had exactly the same issue in our last house. It was in a difficult to see spot hidden behind our kitchen cabinets. We only realised the severity of the issue when the ceiling boards gave way and fell on my head.

There are only two industries that call their customers ‘users’…

I think you may have Europe confused with Japan.

I went from a manual to an EV. For an everyday use point of view there is just no comparison. Acceleration is effortless, start/stop traffic is no longer a nightmare, it’s quiet and refined. It is the ideal daily driver. Even on longer trips I no longer feel fatigued after driving for 4-5 hours (the enforced charging stop helps with that).

I personally would not go back to an ICE car in general, manual or not, for everyday use.

From an enthusiasts perspective, however, this is a different question. I wouldn’t rule out getting an ICE manual for fun/weekend use in the future - the kind of driving where you can actually enjoy the level of fine control and feedback that a manual gives you, rather than just wasting it in traffic. But it would have to be something pretty special.

If your employer expects you to access corporate resources or be available to respond / on-call out of hours, then they should issue you a corporate device to do so.

I can’t speak from experience as I don’t own any Amazon devices, but I have read reports that it seems to work fine with the FireTV variant of Android.

The dev has only tested it against Chromecast with Google TV, with that said I’m using it on a Shield TV and a Shield Pro and it runs fine on both.

Google is already doing this with their default Android TV launcher. I tolerated their home screen ‘recommendations’ for a while as they occasionally highlighted something interesting to watch, but one day I switched on the TV and was greeted with a huge advert banner for a fucking watch on the home screen.

At that point I spent a few hours setting up FLauncher on all my ATV devices.

Same. Coming up to 4 years owning my Model 3 with no major issues and no work needed other than normal serviceable items common to all cars (tyres, wiper blades, cabin filters, etc).

On the flip side, one of my old coworkers who got his Model 3 at the same time as me had a litany of problems from day one. We used to joke that his car had been built by an intern on a Friday night before a major holiday.

I don’t do enough miles these days to justify getting rid of a perfectly good, functional, almost brand new car and buying a new one - I plan to just run it into the ground instead.

I don’t think I’d buy another Tesla in the future, though. Not necessarily because I care what people think of the car I drive, but because Tesla has made some astonishingly stupid decisions with their new/refreshed cars. No physical drive selector? No TURN SIGNAL STALK? Yes, because I love having critical vehicle controls on a movable surface. Come on now.

I use both. Pi-hole running in a docker container on one of my home servers which my gateway is configured to assign as the default DNS for all clients, and uBlock Origin on all my browsers to catch everything else.

Pihole is pretty good at catching ads on platforms that are not suited to browser based blockers (IoT devices, streaming boxes etc) but it isn’t perfect and is best used in conjunction with another solution.

That is exactly what I did with my dumb washing machine (and dishwasher).

Each has a ZigBee energy monitoring smart plug which is connected to a local Home Assistant instance. Spent an hour or two writing automations based on the power draw reported by the plugs and now I get push notifications that report whenever either machine finishes its cycle (including how long it took).

Worf is in the opening scene because he’s trying to sue Picard over that barrel.

This seems like a brilliant solution to a problem that shouldn’t exist.