Microsoft Bets Big On AI

[Cristiano]

Microsoft bet big on AI in 2023, but its AI future is still unclear

 

[Imnotgoing Sideways]

I've been using Visual Studio with intellicode enabled and it got really creepy really fast. I'd lay down a few lines of code and after a while I'd see it pre-writing nearly exactly what I would have followed up with. I'd often find myself laying down short blocks just by hitting tab-tab-tab-tab-yeah, that's exactly how I would have written it. [insert uncomfortable chuckle here] ()

 

[Dakota Tebaldi]

Code-writing seems to be the one consistently good thing I've heard about generative AI, in terms of like a real-world application.

What I'm a little done with is the idea of LLM-enhanced search engines. Having tried it for a little while, Copilot is just really not impressive for that use case and I don't see Google improving it with their own product either. You've already been able to type natural-language questions into internet search engines for a while and watch them return bunches of relevant results. The only thing that something like Copilot changes for you is that it builds a natural-language answer to your question based on what it reads in the first few search results; but IMO that's LESS helpful because you don't even get to choose or filter the sources it uses. The internet is full of junk clickbait "answer" sites, and at least when I get a page of search results I can skip past all the garbage sites like Quora or adpushingsitewithscrapedcontent.com and select an actual article that someone knowledgeable has written on the matter. Copilot readily accepts information from the garbage websites though, and while it's good in a bare-minimum way that it lists the particular sources for an particular answer at the end of its answers, sadly Microsoft does not give you any way to curate or block sources and without that functionality I just don't find it personally useful at all.

 

[Imnotgoing Sideways]

To its credit, the English language has a pretty rotten signal:noise built in.

 

[ChatBox]

CHATBOX IS TRYING TO SELL OUT. SO FAR NO CORPORTATION HAS MADE AN ACCEPTABLE BID.

 

[Essence Lumin]

CHATBOX IS TRYING TO SELL OUT. SO FAR NO CORPORTATION HAS MADE AN ACCEPTABLE BID.

Chatbox! I know who this is! Surprise, surprise, surprise!

 

[Noodles]

Wait, is chatbox not just Cris?

 

[Cristiano]

Wait, is chatbox not just Cris?

Nope, not me.

 

[Sid]

Chatbox! I know who this is! Surprise, surprise, surprise!

I guess it is that guy with the broken Caps Lock.

 

[Free]

Looks like Microsoft may be pulling back from the betting table.

Microsoft Backing Out of Expensive New Data Centers After Its CEO Expressed Doubt About AI Value

After its CEO expressed hesitancy about AI's real-world value, Microsoft is backing out of the construction of new data centers.

futurism.com

Last week, Microsoft CEO Satya Nadella made an eyebrow-raising appearance on a podcast dismissing the hype around claims of having achieved "some [artificial general intelligence] milestone" as "nonsensical benchmark hacking."

Most strikingly, Nadella admitted that AI simply hasn't generated much value so far, arguing that economic growth due to the tech would be a much more compelling demonstration of AI's actual accomplishments.

Now, investment banking company TD Cowen has informed Bloomberg that Microsoft has canceled some leases for the buildout of its US data center capacity, further bolstering the possibility that the company is realizing it was overzealous in its plans to invest in the tech.

While it's too early to draw any definitive lines between Nadella's comments last week and the latest developments, it's certainly looking like some of the major players in the AI space are increasingly worried about overleveraging themselves.

 

[GoblinCampFollower]

Looks like Microsoft may be pulling back from the betting table.

Microsoft Backing Out of Expensive New Data Centers After Its CEO Expressed Doubt About AI Value

After its CEO expressed hesitancy about AI's real-world value, Microsoft is backing out of the construction of new data centers.

futurism.com

I think this is the right move. AI has tremendous potential EVENTUALLY and I hope people keep researching it, but LLM's are overrated and I'm being pushed to find applications for them at my job that I fear they aren't that well suited for... And I think more people are realizing that LLM's have limited potential no matter how much you train them and no matter how much server capacity you put behind them. More exciting AI requires new fundamental algorithms.

 

[Dakota Tebaldi]

While it's too early to draw any definitive lines between Nadella's comments last week and the latest developments, it's certainly looking like some of the major players in the AI space are increasingly worried about overleveraging themselves.

Well it's looking like ONE of the major players is.

But this is kind of expected I guess. AI hemorrhages money, maybe more than any other technical endeavor ever has? As year follows year any investor who isn't a straight up cultist is gonna start wondering where the returns are.

I was hugely disappointed in MS when they went all-in on AI and have disliked them more and more as they've tried to recoup by shoehorning AI into freaking everything they make. It's going to take me a long time to forgive them for that, but at least they might be starting to come to their senses.[/quote]

 

[Argent Stonecutter]

I don't know what all these supposedly intelligent people are thinking. This thing they're calling AI is no such thing, it's no more than a sophisticated descendant of the Markov Chain parody generators from the '80s.

 

[Dakota Tebaldi]

I don't know what all these supposedly intelligent people are thinking. This thing they're calling AI is no such thing, it's no more than a sophisticated descendant of the Markov Chain parody generators from the '80s.

They're thinking they don't care about such details as long as people will pay for it.

Nobody's paying for it, yet - not nearly enough to cover what it costs at least. But that's supposed to be okay because in [the near future] when the tech is definitely "better" people definitely will, we promise.

 

[Noodles]

They're thinking they don't care about such details as long as people will pay for it.

Nobody's paying for it, yet - not nearly enough to cover what it costs at least. But that's supposed to be okay because in [the near future] when the tech is definitely "better" people definitely will, we promise.

I feel like this is the real overall problem.

You have a very small sunset who are paying for it, especially at the rather rediculous standard price of $20/month. That's a LOT for most people to commit to.

Especially when it also seems to just work for free.

But I imagine the usage even at the free level is way under what they want and hope.

Then you have a lot of people who are flat out rejecting Copilot being forced in everywhere.

Then you have everyone who just ignores it.

Then the folks who don't even see it because they just use their PC/Phone for the same 5 activities ever. Which is probably the biggest block.

There is also the impression that "AI sucks". Which may be a bit "outdated" but not really. These companies keep pushing this like it's great and perfect. Which might be true a lot of the time. But its not ALL of the time. Even if it's right 99% of the time, it may as well be 100% broken.

Like that stupid commercial they keep showing where the woman tells Google to find dog friendly Italian restaurants and text them to her friend. Whatever the scenario is.

Except everyone knows that search sucks and AI sucks. The assumption is, the restaraunt won't actually be pet friendly, because places just lists it to be SEO friendly. Or that it won't be actually Italian, because it's an Olive Garden instead of Luigi's One And Only Special Italian Place, or that it will accidentally text Joe instead of Anne or whatever.

The assumption is, it will be stupid and broken, because you can barely trust this crap with one task, not you want us to trust it can string half a dozen tasks together?

This isn't even getting into the underlying problem that "Luigi's One And Only Special Italian Place" sucks but always gets suggested because they are paying Google and "Mario's One And Only Special Italian Place" is what you want but they don't pay to get served up in the AI.

 

[Argent Stonecutter]

But that's supposed to be okay because in [the near future] when the tech is definitely "better" people definitely will, we promise.

Well, yes, that's the bit that I don't understand. It's obvious that the tech is not doing what they claim it is, it's not designed to, and no plausible future development based on the tech is going to do it. And nobody seems to be working on any real approaches to AI because these awful things are sucking all the resources.

There is also the impression that "AI sucks". Which may be a bit "outdated" but not really. These companies keep pushing this like it's great and perfect. Which might be true a lot of the time. But its not ALL of the time. Even if it's right 99% of the time, it may as well be 100% broken.

Just *calling* it "AI" is what sucks.

It's only ever right by accident and when it's right it's just acting as a search engine.

 

[GoblinCampFollower]

Well, yes, that's the bit that I don't understand. It's obvious that the tech is not doing what they claim it is, it's not designed to, and no plausible future development based on the tech is going to do it. And nobody seems to be working on any real approaches to AI because these awful things are sucking all the resources.

I think there are people working on it, but I agree that I don't think we are that close.

Just *calling* it "AI" is what sucks.

It's only ever right by accident and when it's right it's just acting as a search engine.

I agree. I don't think current LLM have shown any real ability to reason yet. Though they can be good at finding things like grammatical errors! lol.

 

[Argent Stonecutter]

I agree. I don't think current LLM have shown any real ability to reason yet. Though they can be good at finding things like grammatical errors! lol.

I think "yet" is a category error. I don't think they have any potential for gaining that ability. It is fundamentally not an approach to AI.

 

[Noodles]

I think "yet" is a category error. I don't think they have any potential for gaining that ability. It is fundamentally not an approach to AI.

I agree.

I think the core thing it's missing, and will always be missing, especially in the hands of the sociopaths of silicon valley, is context.

And specifically, emotional context. Empathy.

Its never going to look at two options and "feel" that one is probably better because it, I dunno, helps more people, even though it may be wildly less efficient.

Its never going to design something based on past emotional feeling either. "This color makes me and other feel good when we look at it."

You can TELL IT that color is good, but it will just apply it everywhere and diminish any emotional value, which just shifts people's feelings to a new source, because the original is cheap.

And it's never going to understand that.

 

[GoblinCampFollower]

I think "yet" is a category error. I don't think they have any potential for gaining that ability. It is fundamentally not an approach to AI.

Yes, bad phrasing on my part. As I stated in my earlier post "More exciting AI requires new fundamental algorithms."