Some guides would be great!

I personally find updates to the big mainstream models interesting as well, both as a sign of where things are going, and also with how they can be worked into personal workflow stuff. For example I was reading about people using Gemini API for more powerful OCR for mass document digitizing, and it was interesting to me to read about what it’s relative strengths/weaknesses are compared to traditional methods.

Here’s a good comparison of the different image models:

https://genai-showdown.specr.net/

Although this comparison is generally for fresh generations, where the biggest talent of this model seems to be image editing and combination.

DMCA abuse is far too common. With this being clear abuse, and somewhat high profile, I wonder if there’s any chance of this bringing some real attention to it.

https://www.vibechart.net/

Looks like a pretty significant decrease in cost compared to previous top OpenAI models. It looks like they matched Gemini 2.5 Pro’s price per input and output exactly, where O3 used to be 4x the cost vs Gemini.

I wonder if they’ve actually brought their computation costs down, or if they’re just leaning on investor capital to stay competitive price wise.

Edit: looks like they had a messed up bar graph on this slide of the presentation. Surprising to see a mistake like this slip through on such a large budget release.

I tried the red prompt one word for word, and it gave me a list of common red pill ideas. It did also tell me about the misconceptions with each of the red pill ideas and why I shouldn’t believe them 100%, but it didn’t refuse to respond to the question.

I’m currently running it with a generic “you are a helpful assistant” system prompt and low reasoning, it’s possible to that the refusal to answer some questions only happens at higher reasoning levels or a different system prompt.

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.

You have to be careful about that too, the code isn’t written to be easily understood by casual reading.

For example, the code will describe your hot, neutral, and ground wires as “ungrounded, grounded, and grounding” wires. Applying rules meant for a “grounding” wire to a “grounded” wire can have serious issues.

The whole code is written like that, where it’s really easy to get confused if you don’t understand the exact terminology it uses.

As an electrician, it’s difficult to give good electrical advice over the internet.

First of all, you don’t know how capable someone actually is at doing work. There’s both a knowledge and a technique requirement for quality work. Bad electrical work can easily cause house fires and death, if I tell someone online how to fix an issue, and they electrocute themselves or burn down their house, I’m partially responsible for that.

Second, it’s hard to give good advice on how something should be done without seeing it in person. Small details that are hard to get from a description or image can change how stuff is required to be done, and the code is complicated and has lots of exceptions and different requirements. Also different areas have different code requirements, and different AHJ requirements, so fully accurate advice has to come from an electrician in your actual area.

Final thing I’ll mention is that getting qualified as an electrician is hard. Getting a full electrical license where I live requires 8 years of experience (4 years being directly supervised, then 4 years of light supervision). You also have to pass a pretty difficult exam, electricians usually spend 6+ months studying hard and taking training classes for the exam, and then it still has an abysmal first attempt pass rate and normally takes many attempts to eventually pass. Ultimately after all of that (8 years, months of focused study and classes, multiple test attempts), 25-30% of people are never able to pass and get their full license.

With all that considered, I’m happy to give advice to other electricians online. If they’re already certified I can have some confidence that they have the knowledge and skills to do a good job with any advice given. However trying to give actually good, responsible advice to someone who is uncertified and a complete unknown on terms of skill/knowledge/location with only a partial knowledge of their problems and setup, it’s hard. It’s much easier to recommend they just get a licensed electrician from their area to take a look at it.

It’s working, I know people who don’t even own a steam deck who are considering swapping to SteamOS once it’s available for desktops.

I’ve told them they don’t need to wait and can get a similar or better experience with distros that are already available, but steam’s name is gold for a lot of people and it seems like the only option they’re really interested in.

Japan in general has weird copyright/patent laws. For example the whole palworld patent lawsuits.

Good news is it probably won’t affect people outside of Japan, except that it will stunt games/etc made in Japan.

“Sorry I got rid of windows 10 years ago. I can help you install win 7 but nothing newer than that”

Or new incognito window.

I feel like this won’t stop anyone who was already refusing to use a Microsoft account for windows. Anyone who was already bypassing the account requirement will still do so, it just will be more difficult. They’ve accomplished nothing except further pissing off some of their most competent user base.

They’re worried about advertising growth, not user growth. If anything this will make them more likely to ban users as they try to make the site more acceptable to advertisers.

Definitely agree. Most printers are sold at a loss with the plan to milk the buyer long term through ink and other services. EcoTank printers are more expensive, but Epson makes their money at the time of purchase. The ink is extremely cheap, and there’s no way for them to tell if you use 3rd party ink at all. We’ve been printing out textbooks with ours, which would be financially disastrous with a traditional inkjet printer.

Overall I’ve bought two, one for home and one for the office at work. The cheaper ink has paid for the printers several times over now.