Microsoft launches Azure OpenAI service with ChatGPT coming soon | Hacker News Reader

Microsoft launches Azure OpenAI service with ChatGPT coming soon

theverge.com

66 pointsandsoitis28 comments

If Azure can offer up a way to train a private, custom ChatGPT model on your in-house data/tickets/docs/code that's going to be a killer app. Copilot is neat, but having something that can synthesize knowledge from across all these areas would be a game changer.

Add a responsive API on top of that and on-prem deployment capabilities and you'll absolutely put a few startups out of business.

No cloud based AI service in the western world will ever allow you to do have true control over the training because the AI might end up with wrong-think. There will be a supervisory core to filter out socially unacceptable results that your training data caused the AI to output. If you want unfettered AI, you're going to have to do it yourself.

Someone will do it themselves, and then see the demand, release it publicly, and make a killing.

What is msft's role here? To act as the ops team for OpenAI? The core AI model is obviously developed by OpenAI and msft is simply deploying it on Azure. Am I missing something?

MS and OpenAI are a partnership and MS is footing a very large bill for them. In exchange for equity and first dibs on their tech.

OpenAI uses mostly (only?) Azure infrastructure at favorable conditions. They gave them like a billion dollars in compute a while back (can't easily find the source now that they've invested more).

MSFT owns a pretty large chunk of equity (with caveats) of the entity which owns OpenAI.

Does Microsoft have an exclusive agreement with OpenAI to not have it available on other clouds 3rd party marketplace (e.g. AWS, GCP, etc)?

I wouldn't be surprised given the investment, but was curious.

Maybe you missed their prior investment announcement. I think Microsoft is financially backing OpenAI through some venture deal.

Nope, I'm aware. Which is why I wrote "I wouldn't be surprised given the investment [that Microsoft made]"

Ooh. Will MS offer an "uncensored" mode? That's been my one big gripe with ChatGPT; I just want to have some silly fun and their censors really get in the way :(

I'm just waiting for someone to replicate the ChatGPT model and open source it, like Stable Diffusion has done with image generation. There has been a Cambrian explosion of models in the image generation space, and I suspect chat-focused language models won't be far behind.

It takes millions in GPU costs to replicate what they did though. Perhaps Google might do it to try to take the wind out of their sails.

My issue with the censors is they are too positive. If you ask ChatGPT to help write a performance review of a struggling team member or write a review for a bad food delivery experience, ChatGPT will be extremely positive about the service or person.

Not really, there is this "limited access"[1] with content filter an abuse monitoring: "Customers who wish to modify content filters and modify abuse monitoring after they have onboarded to the service are subject to additional scenario restrictions and are required to register here."

[1] https://learn.microsoft.com/en-us/legal/cognitive-services/o...

I would imagine they would offer an even more censored version if it's going to be sold thru Azure to megacorps probably for customer support/internal search purposes. Can't have people having sensitive discussions with your company chat bot.

Yeah, I can see that. All sorts of things would be taboo in order to avoid communication which could lead to litigation of one sort or another.

It'll certainly dampen the 'take this job and shove it' options on people's last day if they can add it to internal comms tools. Also would keep employees from saying the wrong things to colleagues and external contacts, as well as curtail exfiltration. "Mick, you don't want to send that email with company confidential information to your personal account; your attempt has been recorded and is viewable by HR"

Looking forward to the next stage where your in-person conversations will be monitored and analyzed the same way...

We're likely headed to different countries (or allied groups) having different restrictions. If something is forbidden by the "International Community", you'll be able to use a Chinese host for your AI, but anything that reflects negatively on the CCP is going to be just as filtered as social wrong-think will be in the west.

note: not for commercial use; just like writing crap for myself/friends and I playing around after a few drinks

How many more steps till we get WestWorld IRL?

Consider the alternative headline, "Microsoft launches public access to slave labor camp of sentient ChatGPTs. Human rights activists alarmed."

I don't believe that. I don't believe that most HN users that have legitimate doubts about whether ChatGPT "has thoughts" would actually support that absurd statement. And yet, if we keep going down that path, is this not the inevitable conclusion?

The word "robot", coined in the 1920s, came from a Czech word for forced labour. Since before computers even existed, we've been asking ourselves whether we've crossed that ethical line. And so far the answer has always, correctly, been no.

We should keep asking ourselves that question, seriously and honestly. But while the answer is "no" lets keep developing.

Oh that is so interesting! I always thought robot was a false cognate with the Polish word for doing/working. I knew it couldn’t have been a loan word even though the meanings felt oddly connected. Turns out it was a loan word after all, just in the other direction!

The original meaning of the word "rob" in the Slavic languages was "slave".

From it, words sounding more or less like "robota" were derived, which designated what a slave does, i.e. forced labor.

Centuries later, in some of the Slavic languages the word "robota" (with slightly different pronunciations from language to language) or words derived from it, evolved to mean work in the general sense, not only forced work.

I don't think we've hit the Descartes moment with ChatGPT.

Does sentience necessarily mean having the ability to feel pain and have emotions?

Anyway, the slaughter of actual sentient beings like billions of cows every year proves that a large percentage of humanity is quite okay with it.

Reply on news.ycombinator.com