Windows AI Studio simplifies generative AI app development by bringing together cutting-edge AI development tools and models from Azure AI Studio Catalog and other catalogs like Hugging Face. You will be able browse the AI models catalog powered by Azure ML and Hugging Face, download them locally, fine-tune, test and use them in your Windows application. As all of the computation happens locally, please make sure your device can handle the load.
↫ Windows AI Studio Preview on GitHub
Nothing particularly exciting here, until you get to the installation process, as noted by Venn Stone on Mastodon: you need to install Linux, in the form of Ubuntu 18.04 or higher on WSL, before you can use this Microsoft offering.
I don’t know, but that’s just funny.
Thom please, I have got to know, why do you call AI “AI”? What is it that makes it not real, and what would it take for it to become real to you?
It is just terminology. Just ast with “smart” things. smart implies cognition and predictive cababilities as well as logic. I fail to see how a “smart” fridge or power outlet fullfill the criteria.
TLDR: It sounds cool and invokes feelings of futurism in the customers.
The difference is that AI is actually smart, but not perfect, of course. I think that people who think otherwise tend to either be in some sort of denial, or simply have not tried using GPT-4 to do real work.
A chatbot on the commodore vic 20 can be considered “smart” if you push the boundaries of “what is what”, same can be said about the same chatbot being AI. All humans seems to have some pre-programmed code, likea breathing and vomiting from the time life on earth was just a small one cellular tube.
A smart outlet has less cpu power than C64 and can do a lot less. Where is the “smart part” as defined by my previous post?
Most of this current crop of rushed “AI” offerings are versions of large language models “LLM” – which is in actually, a hack. It’s just predictive text generation, and as such, is the cleanest example of “artificial intelligence” in recent memory. There are other things that often get the label “AI” that are more accurately described as “machine intelligence” (which often gets the label “machine learning” or “ML”), but these LLM products are not that. AI is accurate to describe these products – IMHO.
Artificial Intelligence and Machine Learning have very well defined and widely known meanings in the industry. At lest the part that are interested in them, along with a very long tradition in computer science / academia.
Artificial Intelligence is about reasoning and problem solving. It could be as simple as a few “if” statements, or could be backed by a very large deep neural network.
Basically what you interact with to find the optimal route for navigation, playing chess against a computer, or translating a text from English to French.
Machine Learning is for learning mathematical functions by machines. Yes it is just that. So it could again be as simple as learning the mean and the median, or optimizing a transformer network using Adam or gradient descent.
In case of LLMs, they have already surpassed humans in many tasks, including language understanding and translation. Do they make silly mistakes? Yes, they do, but apparently much less than an average human with an average IQ on those tasks.
Anyway, there are nice tutorials on this subject, anything from practical python ones:
https://www.tensorflow.org/text/tutorials/transformer
Or simple wikipedia articles:
https://en.wikipedia.org/wiki/Attention_(machine_learning)#A_language_translation_example
AI today is restatic and lacks logic and predictive capabilites.
If the AI had this, then all problems on earth could be solved rather fast, without wiping out humanity. Imagine inventions that could be done if logic and predictive capabilites was ran on distributed CPU’s.-
NaGERST,
Well, arguably humans are one of the biggest impediments to solving almost all of our world’s problems through mass ignorance and vilification driven by greed and identity politics. These keep us stuck not only prevent progress, but cause us to divert massive amounts of resources to repression and war. We are the problem. The irony is that any AI smart enough to figure that out would be a threat to us, or at least our existing social structures. For AI to remain beholden to the existing social order, it cannot be allowed to get smart enough to solve the problems that are keeping the social order intact.
I’m trying to understand what you meant by this.
Looked this up, but could not find the meaning.
AI and logic has been together for a very long time. For instance Prolog is over 50 years old:
https://en.wikipedia.org/wiki/Prolog
How do you think Google maps route planning for example, anticipates traffic patterns? Or IDS systems identifies harmful attacks? Or simple online marketing campaigns work?
Again, there seems to be many nebulous “definitions” of AI floating around. I am still trying to figure that out.
I believe the reason is simple:
The opens source tools that they use “just work” on Linux, and Windows already support WSL. No need to try to port everything to Windows unnecessarily.
Look at the instructions for Hugging Face:
https://github.com/microsoft/windows-ai-studio/blob/main/walkthrough-hf-dataset.md
Even though the backends run on Linux, the UI part is still controlled on the Windows side. And as a bonus, they can always sell Azure resources to do this once the process becomes more mature.
@sukru Agreed it seems the simplest solution is again the most likely.
I work for an AI company and I will say it isn’t just Linux but only Ubuntu that is supported in the AI world. Despite Redhat’s best efforts if anything it’s supper of the current most common AI tools and libraries has actually declined in the last year. . Despite Ubuntu’s utter dominance in the installed base they haven’t really figured out a way to monetize it. Microsoft has no option but to use WSL and Ubuntu.
Python is easier to use under Linux. That’s it.
The linux distro model is banana if you take a step back and observe the untold replication of effort on the same solved problems. Then they plugged systemd in and gave linux cancer.
Yes, only one banana… that edit feature was quite good.
I do not know if it is funny. It is not like microsft is any type of leader in the space at all. The comepetitors in this particular space is all running linux in one form or another (except perhaps AIX for ibm clients)
Microsoft is an interesting country in many cases, and is uniquely positioned to make a win32 and uwp/metro distro where it would not just work like in windows, but they would beat wine for modern stuff, and they can fall back on wine for the old stuff.