The definition of understanding they use is very shallow compared to how most would define it. Failure to complete a task consistently when numbers are changed, even when they don't effect the answer shows a lack of real understanding to most. Asking a model the sheet drying question for example will give different results depending on what numbers you use. Better models are better at generalizing but are still far from demonstrating what most consider to be real understanding.
This game gives me weird vibes. The studio already had 2 other games in early access before this one, both of which are seemingly abandoned or at a standstill development wise. The gameplay itself just kind of looks like a generic base building game but with Pokémon and guns. Most of the steam reviews are just making jokes about the knockoff elements, guns, animal cruelty, etc.
I honestly can't tell if this game is actually good or if it's just a brief trend.
Tech companies say they are putting in place systems to prevent AI being used for criminal or other malign purposes, and insist the new technology will create more jobs than it destroys
Automation doesn't create more jobs than it replaces, if it did there would be no point. In some cases it removes bottlenecks which allows for greater scale which in turn creates jobs, but these new jobs often require different skillsets from the ones displaced.
You're not really allowed to criticize them for this when your economic system encourages this exact behavior.
This isn't shocking at all. The markets for obscure language content are incredibly small so there's no incentive for most to spend resources on it. I'd argue mediocre machine translation is better than nothing at all in many cases, but for unsupervised training it does pose a challenge.
I'm not sure why it would be any different from how this is treated with search engines. Both scrape massive amounts of openly available data and make it available in some form. Any training data or information that a model could potentially spit out is already available through a search engine's index.
I think part of it has to do with how we cope with death. Almost all religions are centered around what happens when we die. Whether it's reincarnation or an afterlife, most believe that there's something beyond. I think that to a certain extent we're predisposed to have this mindset.
It won't matter until we hit 600. 600 integer scales to every common media framerate so frametimings are always perfect. Really they should be focusing on better and cheaper variable refresh rate but that's harder to market.
Flatpak is good for diversity. Users don't need to worry about whether the obscure distro they want to use has the software they want in its repos. If a distro supports flatpak it will work with most popular software out of the box.
Very strange article. It lists several front ends, some of which are not open source, as well as some raw models without clearly distinguishing between them. RWKV was mentioned which is cool I guess.
The first option listed should have been huggingchat, followed by the various local UIs, with a separate section discussing the models themselves.
I'm normally all for making fun of pedophilia in the church, but this is just gross. Using the term "sex toys" in reference to children is disgusting and negates any possible humor that this meme could have had. This post goes beyond dark humor and is just distasteful.
Fragmenration is not preferred by most users and many of the services and systems these two run require large international teams. That's not to say that they don't need to be checked by antitrust.
This has been known for a long time. The main point of contention now will be who is liable for infringing outputs. The convenient answer would be to put the responsibility on the users, who would then have to avoid sharing/profiting from infringing images. In my opinion this solution can only apply in cases where the model is being run by the end user.
When a model is served online, locked behind a subscription or api fee, the service provider is potentially selling infringing works straight to the user. Section 230 will likely play a role, but even then there will be issues in the cases where a model outputs protected characters without an explicit request.
We're not breaking ground on AI innovation (in fact, we're using an old, "deprecated" file format from a whole six months ago)
The ggml format isn't "deprecated" it's completely dead. In those 6 months we've also seen 2-4x speedups on some systems, not to mention improved accuracy via kquants. I don't know why they would build out a new extension with such an ancient dependency.
So the solution is to just not do that.