I’m trying the 20b weights model in LM studio now, and it’s not having any issues with providing summaries of plots of movies/shows/episodes. Do you know what kind of system prompt or any other details on what’s needed to keep it from responding?

The specific example the article uses is a guy requesting a new work computer, so he uses an LLM to create a lengthy request that lists out all the reasons he needs a new one.

It used to be that writing out a long request for something like that showed you were serious about it and willing to invest real time/effort into the request. A simple “I need a new computer because mine is old” type request would likely be dismissed. Using LLMs to write long requests for stuff like this is bypassing what was significant about the long-form request to begin with.

Here’s some first impressions from someone who’s gotten to try it out.

Special thanks to @GoogleDeepMind for inviting me to try out Genie 3. I’m excited to share my thoughts on this early research prototype and also some of my live recordings below:

I spent the whole day playing with the system and when it works, it is truly mind blowing🤯. It is the first neural game engine / world model I have tried that generalizes so well and has long term world consistency. Here’s a couple of examples from my live recording and some thoughts on what it means for the future of gaming, robotics, digital experiences and ASI.

Where it shines:

• Truly general-purpose and quick startup time. Works exceptionally well for gaming environments but also generalizes to other industrial and real-world scenarios.
• It learns physics. Although there are systematic failures even for rigid body physics, it was clear to me that it can learn game engine and non-rigid physics without an underlying engine (and in limit learn from game engines via training data).
• It works exceptionally well for stylized environments with characters walking around. This will have implications for concept artists, level designers and game devs.
• It is way more fun than video models, indicating that there are high retention consumer experiences waiting to be built with this in the future
• Photorealistic walk throughs and drone shots work exceptionally well
• Global illumination and lighting works surprisingly well
• Visual memory is quite powerful and the same objects approximately remain coherent under occlusion and longer time horizons

Open Problems:

• Physics is still hard and there are obvious failure cases when I tried the classical intuitive physics experiments from psychology (tower of blocks).
• Social and multi-agent interactions are tricky to handle. 1vs1 combat games do not work
• Long instruction following and simple combinatorial game logic fails (e.g. collect some points / keys etc, go to the door, unlock and so on)
• Action space is limited
• It is far from being a real game engines and has a long way to go but this is a clear glimpse into the future.

The Future:

• It is impressive enough for me to have strong conviction that this is going to disrupt the gaming industry. It is super early days and there are a lot of failures but the writing is on the wall. Lots of challenging scientific, engineering and scaling problems to be solved but it is going to happen in the next 5 years.
• This is the final piece before we get full AGI and now I think we are well on our way to truly solve it once something like this is scaled up. In many ways it is more ASI than AGI but this is a matter of definitions. The fidelity and generalizability will reach human-level and quickly surpass humans
• People are going to combine this with 3D AI and LLMs to build AAA games.

Having now spent some time with Newelle, it asks for permission before running commands as well. By default the commands are run inside the flatpak sandbox as well. It even gives the option to run the commands in an external window and see the results before deciding whether to let the LLM access the results of the command (in case you’re worried the results might have private info you don’t want shared.

Overall it handles it really well to be honest.

I’ve tried Gemini-CLI, and while it can run terminal commands, it asks permission before each one. Being able to read the commands and then click yes once I see what it’s doing can work pretty well in my experience, and should be somewhat safe.

They do recommend running gemini-cli inside a container though, for added safety. I’m not sure if that’s possible with Newelle yet.

The main issue is that SteamOS Beta updates frequently break Decky loader, but it’s rarely an issue if you’re on stable. The worst issues I know of with plugins were IsThereAnyDeal plugin was causing performance issues, and DeckyRecorder was causing some decks to be unable to switch to desktop mode.

Overall though it’s rare to have serious issues from Decky, especially on stable SteamOS. You may not want to install a bunch of unnecessary plugins that you won’t ever use, but you shouldn’t hesitate to use any that meet your needs. If you ever have steamOS issues (crashing/etc) you will want to disable decky and see if that fixes it before you contact Valve though.

I’d be happy to take over !gai@sopuli.xyz. I hadn’t realized there was a sopuli community for it, or I would have already been sharing some interesting articles there. I made a post here.

I’d also be happy to help as a moderator for !anarchychess@sopuli.xyz. I don’t have as much to contribute there, but I’ve always enjoyed following that community. It might be better to have a lead mod there that is more active within the community though, so if anyone else wants that position I’d be happy to concede it to them. I have a comment here.

A lot of the smaller communities here sound really interesting as well, but are completely devoid of posts. I agree with some of the other comments here that I think those communities should be retired, let someone else restart them later if they’re motivated to foster a new community.

It wraps around, you go low enough and you suddenly become a master.

At least CSS loader exists and works. Native theme support (probably through the point shop) would be nice though.

Additionally, desktop mode also got native support for changing the artwork from the same menu.

When you have your game library sorted alphabetically, some games aren’t grouped together like they really should be. For example, you might want all the Yakuza games grouped together, but the newer games are called like a dragon and are completely separate. You could now go and have the like a dragon games sorted as if they were also named Yakuza, so that all the games are grouped together.

Another example for me personally is that I have a bunch of Metroid games on my deck, and almost all of them are grouped together, except for Super Metroid and AM2R. Now I can move those to be listed alongside the other Metroid games.

I’ve been playing Nightreign with friends, mostly using the seamless coop mod to allow 2 players. The game just updated with to allow 2 players natively, so we’ll have to decide if we want to keep using the mod for the rare times when we have 4 people. Overall it’s been really fun, I’ve enjoyed it a lot.

Installing it isn’t hard, you download the latest release.zip file from the GitHub, enable developer mode in decky, go to the dev menu in decky, hit install from zip, and then pick the zip file you just downloaded.

I installed it to check the feature set but haven’t really used it besides that.

https://github.com/aarron-lee/SimpleDeckyTDP?tab=readme-ov-file

I believe this plugin allows you to set different power profiles for plugged in vs portable play. It’s not on the decky store for whatever reason (which is disappointingly common with new plugins), so you’ll have to install it manually.

Relevant xkcd

I recently helped work on a firetruck at a shop that works on existing trucks. They repair/rebuild their bodies, and do a lot of custom work. It’s common for firestations to buy decommissioned trucks from other stations, and have this shop rebuild them rather than buying a new truck.

The truck I was working on had a light tower on it, with 6 led lights. During the work, one of the employees there asked me how much I thought the individual lights on the tower cost. An equivalently bright LED light would normally cost around $100, but because it was going on a firetruck I guessed it would have a premium price and guessed $300-400. Instead he told me each light was about $1500, making the set of 6 lights worth $9000.

The whole job was interesting, basically everything they had to buy new was way overpriced compared to the non-firetruck versions of the parts.

Asking for money is fine, no one complains about having to buy Steam games. The dev is literally developing this full time, and there’s nothing wrong with him charging for it.

That said, he’s charging a lot. I feel like $40 is way too much for a subscription, especially considering how good the free alternatives are. I’m guessing this would have been more finacially successful at a lower price point, while also helping more people.

Pretty cool, I didn’t know about the moonlight fork.

Probably going to see a lot of memes about Wii u running the newest most demanding games.

KDE defaults to Wayland on most distros now, so it might be worth confirming.

You can run

echo $XDG_SESSION_TYPE

to find out.

Are you using X11 or Wayland on your PC? Steam streaming from X11 works fine, but streaming from Wayland is really finicky or doesn’t work in my experience.