Microsoft Azure OpenAI lets enterprises feed corporate secrets to ChatGPT Microsoft wants to make it easier for enterprises to feed their proprietary data, along with user queries, into OpenAI's GPT-4 or ChatGPT within Azure and see the results. This functionality, available as a public preview via the Azure OpenAI Service, eliminates the need for training or fine-tuning your own generative AI …
But how do you turn this stuff off? Any geek tools out there yet that can do this for us?
Why is there no off switch on this ChatGPT stuff? Surly this is breaking all kinds of data protection laws?
I keep finding AI.exe running on client machines. Which comically is then flagged by AMD Gaming Tools as "Alien Isolation has been running for 8 hours...."
I just want to turn it off as I will not use it. Seems a simple request.
Please... call it what it really is...
A monthly TITHE to the Lords of the Manor (aka Microsoft, Google, Amazon, Apple etc)
It can be paid in cash or in the huge amounts of data that they will willingly suck from our computers.
Oh wait...
They are not our computer even if we paid good money for them especially, if they run Windows. MS can f'k with them to their hearts contents and NOT get sued into oblivion for doing so.
Linux used to be a safe haven but even parts of that are falling to the corporate megalords (SITH). SNAP packages anyone?
Do these systems respect internal corporate security boundaries?
Pile in your in corporate data - ask interesting questions - get "interesting" answers that you're not supposed to be privy too? Or are even right?
Anyone else looking at AI as a Nutrimatic Drinks Despenser - outputs almost, but not quite, entirely unlike tea.?
The answer will be a resounding NO!
Everything to do with this sort of information gathering is for one purpose only, to make money for those who gather it.
That so many people willingly just let it be gathered is part of the problem. It is now accepted that the "normal" is pretty much everything you don on any sort of device connected the internet it harvested.
Most people will not turn off all the typing monitoring "to improve stuff".
Even on corporate devices it is very hard to stop the data gathering entirely as everything has Internet access now because so many tools are "as a service".
"how much does my boss earn?" Oh, as a computer operator I knew that, because we used to print the payslips. They were multipart and the outer later got inked, and the inner got carboned, then the outer was split off. I happened to stop the printer to check the alignment (it was tight, had to get all the text aligned in the boxes and we used to stop and check the run in case it needed correction) and I stopped it on the bosses slip. He was doing well, back in the early 90s.
Their unimpressive demo had a limited data subset that you replicate in 45 min with any programming language of your choice.
The most used phrase by Microsoft staff was "additional cost."
They didn't spend 10 billion as a gift to users. This is a business investment they want to make a profit on. You'll eventually be paying for each submission and response. Very limited business application.
https://azure.microsoft.com/en-gb/pricing/details/cognitive-services/openai-service/
That option of storing your (siloed) data for OpenAI use is also staggeringly expensive compared to their usual AI offerings.
If you want to keep your data safer (who knows how secure the data silo approach is) but still use OpenAI could generate embeddings values for your text items (embedding and text values stored locally in some DB system, just use OpenAI* to generate embeddings) and then use embeddings best matches for scenarios such as "AI" driven help bot... Results wont be as good as the "chat" driven solution on your siloed data but will keep your data safe (text you request an embedding for is not saved / used in model training).
*.Or go one step further and just use some other locally running software to calculate the embeddings
OpenAI embeddings stuff here
https://openai.com/blog/introducing-text-and-code-embeddings
So we've just had one of biggest TITSIPs(tm) in yonk with the MoveIT fiasco releasing tons of into the wild, and now MS and their ilk are prepared to take on even more data all in the name of "offering freedom and choice" ( meanwhile making a bomb of their shareholders ).
Yeah, I have a very bad feeling we're long overdue for some sort of apocalyptic event....3....2...
Re title - why??? What the hell do they get out of it?
All the examples seem to relate to finding stuff. If you can’t remember where you put something then maybe invest in a notepad? If it’s stuff generated by your colleagues then either ask them or invest in organising your company’s stuff better
I've yet to be impressed by any demonstration of an LLM that I've seen, and I've seen quite a few (particularly in papers and articles; I haven't wasted my own time with the things). Sure, they can produce banal natural text, and pulp-style illustrations, and other not-especially-good output, but the only actual utility I've seen is LLMs doing mediocre work for people too lazy to do better.
I've studied NLP as part of one of my degree programs and done some other ML work, and I've read quite a few of the LLM-related papers. I simply don't find unidirectional-transformer architectures or their results exciting. It's an inelegant, resource-intensive approach that's producing relatively little in the way of actually interesting new results, particularly given the massive hype and drooling excitement from the fans.
I'm not a big fan of deep-learning approaches in general – again, awesomely resource-intensive for the result, and not producing enough in the way of surprising results or new insights – but at least models like EfficientZero were trying to do something more novel than "throw a fuckton of hardware and data at the thing, optimize for shiny, and who cares what precision or recall or other metrics look like".
Training data:
- are NOT available to other customers.
- are NOT available to OpenAI.
- are NOT used to improve OpenAI models.
- are NOT used to improve any Microsoft or 3rd party products or services.
- are NOT used for automatically improving Azure OpenAI models for your use in your resource
...if you believe that, I've got a bridge to sell you. This is hosted on Azure after all.
And shouldn't are be is? Although data is a plural, that just looks so wrong.
While I'm happy to bash Microsoft, and currently I'd like to bash a lot of their "developers" with big, blunt or even pointy things (gah... ) I do have to admit that the statement made above is, for once, very clear and concise.
It doesn't address siloing of the data within an organisation, or even if that's possible, but it's a pretty strong statement. I'd rather that they were a little more explicit though, for example "none of the data is accessible in any way outside of the controlling organisation and the controlling organisation can delete and manage this data in any way they need to".
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
