Microsoft AI Chief Says Conversational AI Will Replace Web Browsers

erek

erek

8-D
2FA
Joined
Dec 19, 2005
Messages
18,599
“The technological revolution produces a new interface, which completely shuffles the way that things are distributed. And some organizations adapt really fast and they jump on board and it kind of transforms their businesses and their organizations, and some don't. There will be an adjustment. We'll look back by 2030 and be like, "Oh, that really was the kind of moment when there was this true inflection point because these conversational AIs really are the primary way that we have these interactions." And so, you're absolutely right. A brand and a business are going to use that AI to talk to your personal companion AI because I don't really like doing that kind of shopping. And some people do, and they'll do that kind of direct-to-consumer browsing experience. Many people don't like it, and it's actually super frustrating, hard, and slow.

And so, increasingly you'll come to work with your personal AI companion to go and be that interface, to go and negotiate, find great opportunities, and adapt them to your specific context. That'll just be a much more efficient protocol because AIs can talk to AIs in super real-time. And by the way, let's not fool ourselves. We already have this on the open web today. We have behind-the-scenes, real-time negotiation between buyers and sellers of ad space, or between search ranking algorithms. So, there's already that kind of marketplace of AIs. It's just not explicitly manifested in language. It's operating in vector space.”
Source: https://slashdot.org/story/24/12/09...s-conversational-ai-will-replace-web-browsers
 
Google is very afraid of this, because most Google searches are people asking simple questions that a decent AI can easily scrape.
The biggest issue is making sure that the AI is actually delivering the content correctly and not fabricating incorrect answers.
 
This is pretty much the dystopia I want no part of.

I want to see the raw source data myself and make my own interpretation, 100% of the time. I don't want to rely on anyone or anything else for that (person or AI).

What I value about humans is our ability to create, and grasp complex and complicated topics, and describe them to others. If we are just going to outsource that to AI, then what remains of us? Undisciplined? Unable to know, do, create or invent anything on our own? The entirety of my concept of human value is tied to human creation, invention and doing. If we take that away, then there is no human value left. Then we have just turned ourselves into dumb sponges that soak up whatever nonsense we are spoon fed.

I don't want to live in that dystopian hellscape.

There are many reason I don't like AI, including not feeling like I can trust its output, but the above, that is the biggest one. Turning all people into useless bumps on a log that can't figure out anything themselves.

Its going to turn us into these pathetic excuses for life:

1733785278245.png


At which point we might as well become the batteries in the Matrix, because there is no value in that life.

If you cannot do. If you cannot create. Then why exist at all?
 
Lakados said:
Google is very afraid of this, because most Google searches are people asking simple questions that a decent AI can easily scrape.
The biggest issue is making sure that the AI is actually delivering the content correctly and not fabricating incorrect answers.
Click to expand...
This...i hate using google. co-pilot does a really good job. I barely use google for work searches anymore
 
it already did quite a bit for me and many, people that have the real life version (personal assistant) I do image have them do a lot of what we use the website for (hotel-plane ticket-banking, when app did not already do it) instead of doing it themselve, will be similar for peasant people now.
 
Zarathustra[H] said:
Of course it is. As humans we shape our own destiny.

If enough of us do not want this future, we have the power to stop it.

The only problem is, sufficient numbers of people rarely get fired up about things to have an impact.

Complacency is humanity's big killer.
Click to expand...
It's not up you or me individually, is what i am saying.
 
Zarathustra[H] said:
This is pretty much the dystopia I want no part of.

I want to see the raw source data myself and make my own interpretation, 100% of the time. I don't want to rely on anyone or anything else for that (person or AI).

What I value about humans is our ability to create, and grasp complex and complicated topics, and describe them to others. If we are just going to outsource that to AI, then what remains of us? Undisciplined? Unable to know, do, create or invent anything on our own? The entirety of my concept of human value is tied to human creation, invention and doing. If we take that away, then there is no human value left. Then we have just turned ourselves into dumb sponges that soak up whatever nonsense we are spoon fed.

I don't want to live in that dystopian hellscape.

There are many reason I don't like AI, including not feeling like I can trust its output, but the above, that is the biggest one. Turning all people into useless bumps on a log that can't figure out anything themselves.

Its going to turn us into these pathetic excuses for life:

View attachment 696929

At which point we might as well become the batteries in the Matrix, because there is no value in that life.

If you cannot do. If you cannot create. Then why exist at all?
Click to expand...
Interesting view, will some become so dependent upon it that they cannot function well if the lights go out? Most likely. Then again we are already at that stage I would say.

What one does with something, is choice, good or bad. I would love an accurate personal assistant if reliable and can find information so I can achieve something faster. I don’t have to become docile, but better.

If it allows me to do more of the things I want to do, being more capable over time, that would be a positive.

At my job, I’ve used AI, to rapidly find the key information to control a motor drive. Did not have to dig through literally a thousand page manual to find the parameters I want. It has made me way more efficient, productive. Many other types of uses it has worked out well.

It comes down to what one does in the end, choice.
 
People “have a constitutional right to be a dumb-ass.” Bonus points if AI (not a goiogle search) gets you it’s origin.
 
noko said:
I don’t have to become docile, but better.
Click to expand...
We are efficient at optimizing only the skills we need. Skills replaced by AI are likely to atrophy. That might be okay, but it is also likely to make us dependent on ai for those skills.
 
noko said:
At my job, I’ve used AI, to rapidly find the key information to control a motor drive. Did not have to dig through literally a thousand page manual to find the parameters I want. It has made me way more efficient, productive. Many other types of uses it has worked out well.

It comes down to what one does in the end, choice.
Click to expand...
Automation is the key to making one person more efficient at their job and the way to make thousands homeless because AI will take what it can do as work.

Regarding AI search - any script can be modified to return a filtered result that will hide information from the searcher and if we can currently get around it using other methods or search engines, we won't be able to if there is only AI search.
 
Firstly, asking a computer a question isn't new. Android and iOS users have been doing this for some time. It's also not web browsing. It may pull data from the web and show you a website, but it's not web browsing. Microsoft hasn't learned since the Kinect that nobody trusts it with that kind of information. I also still feel like a total asshat asking my phone a question. It's something I normally do when I'm busy working on something else or driving. There's also a level of privacy that is lost by just yelling things out loud.
 
I have no interest in this outcome unless I can see the results of multiple AI responses simultaneously (AI aggregator), at which point I'll be wasting orders of magnitude more computing resources ($$$) than traditional search for the same results.
 
I thought carbon footprints were important? All these AI datacenters seem like a lot of resources burning for results that apparently most people won't trust anyways. Can we put that back into protein folding or something please.
 
DukenukemX said:
It's also not web browsing.
Click to expand...
Exactly the point made by the word replace, asking a question and getting a direct answer replace opening a browser, googling and reading wikipedia or something else.

Same for saying out lout get me the cheapest plane ticket from x to y at date z, instead of googling and looking for a website that do those kinds of things.

And will have limits, specially depending how well the ai render things what browsing it can replace (it cannot replace youtube for video for example), for a lot of things you want more than a voice feedback and a rich graphical UI, for those things they need to get really good at it (all the things that rich people ask personnel assistant to do instead of them browsing the web to do it could be obviously made by AI agents with you just talking to them like you would if you were rich with PAs, but they need to get really good and not need your feedback)

Darunion said:
I thought carbon footprints were important?
Click to expand...
It is because cost is important, so cost (footprints) are going down fast.

Evolution in software and hardware, brought cost per token by 45,000 since Pascal gpus days, by about 240x in the last 18 months, there is a possibility that it will be way less power hungry than traditional compute (already is at predicting meteo, "rendering" in "4k" or evaluating cancer obviously, but that could become more and more the norm for inference to be cheaper in carbon footprints than actually calculating-simulating)
 
Last edited:
As long as AI is having bias programmed into, it will never replace anything. And boy oh boy are algorithm like OpenAI chock full of labotomizing bias. Is there anything in it that doesn't have a hardcoded response if the user asks the "wrong" question?
 
Armenius said:
As long as AI is having bias programmed into, it will never replace anything. And boy oh boy are algorithm like OpenAI chock full of labotomizing bias. Is there anything in it that doesn't have a hardcoded response if the user asks the "wrong" question?
Click to expand...
It can still replace browsing for an answer to simply ask your phone out lout which leg on a led is the negative one.... and it giving you the answer.

A giant amount of web browsing (almost all of it) as virtually no controversy-bias in the sense implied here.

Also think about what it is replacing for most people.... google search, wikipedia, etc..., it is not if that was not full of it.
 
Twisted Kidney said:
Generative AI still doesn't generate anything. Who is going to generate the content for it to steal though?
Click to expand...

That is exactly the issue with this.

Right now AI is trained on web data (with or without permission in many cases, which is shitty, but it is what it is)

If AI answers replace web pages, what will be used to update AI training over time so it can give new answers?

The web pages certainly won't be sticking around if they no longer gain the necessary ad revenue to keep them alive. If you though ad blockers were a problem, this is going to be a WAY bigger problem.

Unless someone finds a good answer to this one, AI will wind up being self terminating in this regard, by biting the hand that feeds it.
 
1.1.2.3.5... said:
Depends on what modalities you're considering specifically here, but i have a very very good idea. Probably shouldn't say, but it's kinda obvious if you're in this space.

Here's a hint, Jensen specifically calls it out near the end of his keynote speech at gtc 2024.
Click to expand...

I have no idea. I don't think I've ever watched a keynote speech from any of these events.

Please don't tell me you are thinking AGI (Artificial General Intelligence). That is a unrealistic pipe dream I don't think will ever happen. Not even in a thousand years.

There is only so much you can do with nested probability functions.

1.1.2.3.5... said:

View: https://www.youtube.com/live/Y2F8yisiS6E?si=kt1ZL4LzUOWBEUwH
Click to expand...


Thanks for the link, but I am not going to watch that. I know it is only going to make me angry. I hate this AI bullshit so much. I don't want to live in an AI world. I think it is time for Butlerian Jihad.
 
Last edited:
Zarathustra[H] said:
Please don't tell me you are thinking AGI (Artificial General Intelligence).
Click to expand...
I have a big ignore list and if I hadn't found other comments you made that I find to be intelligent I would have added you to my list for that!!! So yeah, no...i am not thinking that and no i don't think it's a big secret and no i dont feel comfortable saying it for reasons. Jensen basically says it near the end...just watch it.

Also, think about models combining modalities for training and inference. What percentage of human communication is verbal vs visual...etc...
 
1.1.2.3.5... said:
I have a big ignore list and if I hadn't found other comments you made that I find to be intelligent I would have added you to my list for that!!!
Click to expand...

Thank you, I guess? 😂

Though your comment is a little bit ambiguous. I can't tell if it was that I suggested you might be thinking of AGI that was the potential ignore list trigger, or if it was that I suggested that AGI is an unrealistic pipe dream.

Personally I have a very very short ignore list. Don't get me wrong. There are lots of people on here I disagree with, many of them very strongly, I just have strong feelings about the dangers of isolating ourselves in echo chambers, so I try to keep myself exposed to people I disagree with, even through it is frustrating at times.

I have blocked a very small number of users, but generally it was not because I disagreed with them, and more because they were just being dickheads.

Actually, I just looked at my ignore list, and it is empty, which is weird, as I distinctly remember adding two or three users to it. Maybe their dickheadedness got them banned? 😂


1.1.2.3.5... said:
So yeah, no...i am not thinking that and no i don't think it's a big secret and no i dont feel comfortable saying it for reasons. Jensen basically says it near the end...just watch it.
Click to expand...

Ugh, I really don't want to watch that as it is going to make me angry for the rest of the day, and I don't want to be angry for the rest of the day, but I have to admit, the fact that you don't want to say it out in the open has me a little perplexed and intrigued. Is it because it has a potential of spiraling the thread out of control?

In that case I am going to go out on a limb and guess it has to do with monitoring people against their will and training AI based on their conversations, movements caught on video, etc. etc. And yeah, that would be a "sparking civil war" level of "hell no" for me if it ever came to pass. Like blowing up Nvidia/Alphabet/Meta headquarters with everyone inside it kind of "hell no".

This is something that must be assured that it will never come to pass by any means necessary.
 
Zarathustra[H] said:
Though your comment is a little bit ambiguous. I can't tell if it was that I suggested you might be thinking of AGI that was the potential ignore list trigger, or if it was that I suggested that AGI is an unrealistic pipe dream.
Click to expand...
Okay, sorry. Let me be clear...we are not going to solve agi anytime soon and it's stupid to say so. In other words, it's so obvious that we are not going to have agi that it's insulting to even suggest it 😉.
 
Zarathustra[H] said:
In that case I am going to go out on a limb and guess it has to do with monitoring people against their will and training AI based on their conversations, movements caught on video, etc. etc. And yeah, that would be a "sparking civil war" level of "hell no" for me if it ever came to pass. Like blowing up Nvidia/Alphabet/Meta headquarters with everyone inside it kind of "hell no".
Click to expand...
Imma try to make this more clear...


https://resources.nvidia.com/en-us-ai-optimization-content/ai-digital-twins-industrial-automation

What they have made is a world where there is a lot of synthetically generated data in multiple modalities for both training and testing. They can make as much of that as they need to and they can and will add in voice (real and synthetic) as well as text and all the other bits. Taken together these can create much better responses than any one modality. And with synthetic data you're not limited on the input space (as you mentioned above).
 
1.1.2.3.5... said:
Okay, sorry. Let me be clear...we are not going to solve agi anytime soon and it's stupid to say so. In other words, it's so obvious that we are not going to have agi that it's insulting to even suggest it 😉.
Click to expand...

I tend to agree, but there are lots (like huge numbers) of people out there who seem to think otherwise, and it is sometimes difficult to predict who is who, so I don't think it is that out there to suspect someone might think it is around the corner.

In the grand scheme of things, other than screen names, arguing about tech, AI and privacy, and the occasional "picture of your rig/server/storage/office/mancave" and sometimes the odd errant political argument that almost gets us banned, we don't really know that much about each other :p
 
Zarathustra[H] said:
I tend to agree, but there are lots (like huge numbers) of people out there who seem to think otherwise, and it is sometimes difficult to predict who is who, so I don't think it is that out there to suspect someone might think it is around the corner.

In the grand scheme of things, other than screen names, arguing about tech, AI and privacy, and the occasional "picture of your rig/server/storage/office/mancave" and sometimes the odd errant political argument that almost gets us banned, we don't really know that much about each other :p
Click to expand...
okay totally fair
 
1.1.2.3.5... said:
https://resources.nvidia.com/en-us-ai-optimization-content/ai-digital-twins-industrial-automation

What they have made is a world where there is a lot of synthetically generated data in multiple modalities for both training and testing. They can make as much of that as they need to and they can and will add in voice (real and synthetic) as well as text and all the other bits. Taken together these can create much better responses than any one modality. And with synthetic data you're not limited on the input space (as you mentioned above).
Click to expand...

So wait a minute....

What is the source of this "synthetic" data?

AI generates data, and then AI trains on that data?

That just seems like a recipe for compounding errors to me.
 
Zarathustra[H] said:
So wait a minute....

What is the source of this "synthetic" data?

AI generates data, and then AI trains on that data?

That just seems like a recipe for compounding errors to me.
Click to expand...
Depends....

All models are wrong, some are useful. For the world your robot lives in, you can decide the level of physics fidelity you want to give. Same for image quality, etc....We have more than enough data already from written language etc. what is missing is the normal context of the rest of the world (which we humans all have as we learn). So now add in verbal and visual data with the written data and the corresponding response you might have from a fully dynamic physics plant model, and you are going to get massive lift.

Only a matter of time before these models are put into machines that will begin interacting with the world. The difference between how humans learn and how these machines will learn as that everyone has to generate their own datasets through their experience. The robots will have a collective dataset that will be much more robust. Again, this is NOT solving for AGI, this is taking and combining data to make a much larger n-dim surface to minimize the loss function on. There's just a lot more information that comes through visual means, spatial auditory means, an environment that is reactive to things...etc...The language responses, aided with with the other modalities will get massive model lift.

Just think about adding radar data, multispectral cameras, IR cameras, fish-eye cameras, smell sensors, multiple IMU sensor data....and combinging that into a contiguous training set.
 
Last edited:
  • Like
Reactions: noko
like this
You must log in or register to reply here.
Back
Top