There should be no difference because the video track hasn’t been touched. Some software will display the length of the longest track rather than the length of the main video track. It’s likely that the the audio track was originally longer than the video track and because of the offset it’s now shorter.
You can use tools like ffmpeg and mediainfo to count the actual frames in each to verify.
Koboldcpp should allow you to run much larger models with a little bit of ram offloading. There’s a fork that supports rocm for AMD cards: https://github.com/YellowRoseCx/koboldcpp-rocm
Make sure to use quantized models for the best performace, q4k_M being the standard.
I only have 60 down and 12 up so I cap about 80% of the time with a short uncapped window late at night.
Tun0 is the interface that most vpns are using so I assume proton is the same.
The drive is visible to the OS so if they have any kind of management software in place which looks for hardware changes it will be noticed.
In the context of a “war game” this makes sense. If you remain completely neutral it’s impossible to win. Any examples of similar scenarios the model saw during training would have high aggression rates.
Anywhere speculative investment is involved there are cult like patterns. If your investors don’t believe that your product is going to revolutionize its field you’re not going to get the kind of funding these startups want.
Nobody cares until someone rich is impacted. Revenge porn has been circulating on platforms uninhibited for many years, but the second it happens to a major celebrity suddenly there’s a rush to do something about it.
So the solution is to just not do that.
A language model can’t determine good from bad because it’s only trained to predict the next token based on what it has seen.
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.
The article says “already” like this is the result of something new and not the machine translation we’ve had for well over a decade.
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.
600 also scales to PAL standards
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.
What’s the deal with Alpine not using GNU? Is it a technical or ideological thing? Or is it another “because we can” type distro?