Odd headline since "Catch OpenAI" makes it sound like it's a tool to identify AI developed content rather than "catch up" with a rival business. I have heard that Apple executives were very leery...
Odd headline since "Catch OpenAI" makes it sound like it's a tool to identify AI developed content rather than "catch up" with a rival business.
I have heard that Apple executives were very leery of releasing any generative AI tools until they could for sure build in protections around not having it turn out objectionable or privacy infringing material. I don't know to what extent that's solvable to Apple's standards given how the technology works.
Also strange to frame this as "caught flat footed" since Apple certainly had AI projects in the works that weren't ready for release. Apple is always like this, waiting for underlying technologies to mature to the point where they're ready to weave into a compelling, salable product rather than a neat technical demonstration or gizmo (which is where most AI is right now).
From what I've read they have trouble retaining people for their AI team because: a.) The secretive nature of the company restricts them from being able to publish a lot of their work, and since...
From what I've read they have trouble retaining people for their AI team because:
a.) The secretive nature of the company restricts them from being able to publish a lot of their work, and since most of these people are academics they really want to be able to publish.
b.) They're resistant to the kind of large scale data collection others use to train their models (and have only recently started to come around).
c.) The senior leadership just doesn't really grok the technology and doesn't have a great strategy for how to develop it. Consequently they just keep pivoting in various directions in an unfocused way that keeps them from making tangible progress.
Note that for a few events now, but especially this year’s WWDC, it felt like they’ve made an effort to actively avoid the phrase “AI”. They announced many features using AI, large language...
I have heard that Apple executives were very leery of releasing any generative AI tools until they could for sure build in protections around not having it turn out objectionable or privacy infringing material. I don't know to what extent that's solvable to Apple's standards given how the technology works.
Note that for a few events now, but especially this year’s WWDC, it felt like they’ve made an effort to actively avoid the phrase “AI”.
They announced many features using AI, large language models, and similar stuff (e.g. predictive autocorrection/sentence completion), but as far as I remember, they only ever exclusively used “Machine Learning” to describe the tech.
So this might be an attempt of theirs to avoid big “hype train: Apple does AI now too” press.
Possibly. I think they're generally just very finnicky about branding things, so they'll want to call it something that resists comparison on raw specs to competitors. A good example is their...
Possibly. I think they're generally just very finnicky about branding things, so they'll want to call it something that resists comparison on raw specs to competitors. A good example is their branding of "Retina display" which is more of a statement of "here's what we want the display to achieve (pixel density at the limit of what the human eye can discern)" rather than "here's the resolution."
Maybe off topic but can someone tell me what's the use of AI or machine learning for an ordinary person or layman? I can understand that corporate can automate things and layoff some people, I'm...
Maybe off topic but can someone tell me what's the use of AI or machine learning for an ordinary person or layman?
I can understand that corporate can automate things and layoff some people, I'm not even talking about smaller businesses, but ordinary working class people who presently use things like Excel/Word. How can AI/ML benefit them?
As the tech become more ubiquitous and accessible, there will be ever more increasingly powerful tools available to the average person. As an easy example, if you haven't tried feeding ChatGPT a...
As the tech become more ubiquitous and accessible, there will be ever more increasingly powerful tools available to the average person.
As an easy example, if you haven't tried feeding ChatGPT a draft of an email or work documentation and told it "make this sound more professional", I recommend it. That kind of thing is what it's really good at (take this, make it sound like this). I don't always use what it gives me back verbatim, but it's extremely helpful.
Less obvious examples are already starting to be integrated into software you already use. The text assist tools on some messaging apps that suggest entire phrases or sentences to respond with are less advanced predecessors to the kind of things we'll likely see in the next few years.
You’re going to love AI powered semantic search. You’ll be able to type “What was that email that a director sent me last quarter - something about emerging markets?” into a search bar and the...
You’re going to love AI powered semantic search. You’ll be able to type “What was that email that a director sent me last quarter - something about emerging markets?” into a search bar and the correct result will pop up - even if none of those words are in the email.
I wouldn’t have expected it a couple of years back, but automatic subject tracing in some photo apps to crop them out more quickly/easily can be a fun feature. Similar for layman use: (“live”)...
I wouldn’t have expected it a couple of years back, but automatic subject tracing in some photo apps to crop them out more quickly/easily can be a fun feature. Similar for layman use: (“live”) text recognition in images, and object recognition in images, which can get quite crazy comparably quickly.
Imagine telling someone 20 years ago a palm-sized(ish) device will be able to differentiate and tell you which plant or flower is which, and pull up more info from encyclopedic entries alongside their names… And all in their free time, not in some crazy high-tech lab.
For more work-related things, as suggested already elsewhere here, text generation and re-phrasing can save some time, perhaps? Maybe data curation by just defining what you need and it’ll give you the cell/R formula?
And once there’s more programs interfacing with e.g. Excel/Word/whatever directly, these types of things might become one of many plug-ins that are easily added to an application.
Disclaimer, in the past I’ve felt like a bit of too much hype was going on as I had learnt about the basics of how ML works in around 2021, so before everyone started spouting “oh my god it creates sentences, thus it’s a step away from AGI”, but in recent weeks, I think I have come to be more… appreciative of what one can do the tech.
Insert "Old Man Yells At Cloud" meme here, but I gotta vent: How the hell is this different than the 15 other "game-changing" tech things we've seen in the last 20 years? I'm old enough to...
Insert "Old Man Yells At Cloud" meme here, but I gotta vent:
How the hell is this different than the 15 other "game-changing" tech things we've seen in the last 20 years? I'm old enough to remember MULTIPLE things being touted as the "end of _____" and the "future is now here" and they've all kinda...existed. They've either quietly integrated into everyday life or just fizzled out. EVs, WIFI, NFC, NFTs, QR codes, gyro controls a la Wii, etc. Some are interesting, some are useful, but nothing has "ended" anything else.
Trying to put it shortly. Part of it is the usual corporate hype train for a technological development, but speaking as a computer science adjacent person: I think the reason that a lot of people...
Trying to put it shortly. Part of it is the usual corporate hype train for a technological development, but speaking as a computer science adjacent person: I think the reason that a lot of people in that sphere are so jazzed about this is that LLMs have very suddenly and unexpectedly overcome sets of problems that they have been taught their whole lives are Very Hard, or maybe even impossible, to solve. As someone who isn't usually too easily impressed by new technologies, that aspect of it has shaken my world view a bit. It makes me feel that the future of technology in my lifetime is no longer projectable, and that society just isn't prepared for the implications. Part of this is also that it isn't just a single discrete technology that may or may not progress/revolutionise a single area, but an approach that can and will be turned towards pretty much any data modality and field of human endeavour we can think of. It's more akin to electricity than an electric vehicle.
I mean for just one example, WiFi is part of a larger technological innovation (manipulation of the EM spectrum to transmit information [inb4 smoke signals and lighthouses]) that has completely...
I mean for just one example, WiFi is part of a larger technological innovation (manipulation of the EM spectrum to transmit information [inb4 smoke signals and lighthouses]) that has completely transformed the world and obsoleted lots of things.
Something like an NFT is completely different and hasn't accomplished anything.
EVs are somewhere in the middle as the tech is still nascent, but I'm willing to bet the transportation story will look very different in 50 years (part of a larger continuum of transportation tech in general).
This tech is not like other tech because it accelerates the other tech. It's a force multiplier. Everything knowledge-related that computers are used for is subject to this multiplication effect...
This tech is not like other tech because it accelerates the other tech. It's a force multiplier. Everything knowledge-related that computers are used for is subject to this multiplication effect almost without exception.
Say you're a game designer - what's the biggest pain in the ass of that process? Art asset generation and voice acting, two things that are about to drop to zero cost in an equation they once dominated. 'Render me an apple' happens instantly, rather than you spending several hours creating that asset.
Here's a prompt for GPT27 to contemplate by way of example.
Combine and all of the assets and integrate all of the mods from all of the elder scrolls games into one unified worldspace. Alter the game so that characters can choose to start in any of the historical time periods covered by the game and time travel between them, with their changes to past worldspaces propagating forward with improvisation. Replace all of the NPCs with LLM models based on their original assets. Update all of the graphics to real world fidelity, and while you're at it, improve the aging scripting and graphics engines with a more performant and modern equivalent.
Compare the amount of human effort that requires to an LLM pulling it off in a weekend on a home PC. That's what we are staring down the barrel of here, potentially.
Humans that once did this will now spend their time polishing AI output instead of doing it all themselves, they will work hundreds of times faster, and you'll need a fraction of a percent of the workforce for the same output. Apply this to all knowledge work, and then remember that this technology is not out of its diapers yet. In the future it will require less data and computing power to achieve vastly superior results, just like any technology.
That's not the biggest game changer, though. Making computers thousands of times more effective at their jobs is small potatoes compared to an LLM's ability to personalize everything. This technology's output varies based on who you are. That's new. The primary interface for interacting with computers shifts to spoken language with a universal translator provided by this same technology, and it will inevitably be able to relate to you better than other humans you have known your whole life.
Or, let me put it another way. The training data is hundreds of terabytes. The output, which contains the intelligence of that training data, and can recreate that same data as well as improvise on it, is single digit gigabytes that will run on your phone. If you look at this as a compression ratio, it vaporizes Weissman scores and outperforms Pied Piper's magic compression technology from 'Silicon Valley' by orders of magnitude... and it's so very much more than a mere compressor.
Odd headline since "Catch OpenAI" makes it sound like it's a tool to identify AI developed content rather than "catch up" with a rival business.
I have heard that Apple executives were very leery of releasing any generative AI tools until they could for sure build in protections around not having it turn out objectionable or privacy infringing material. I don't know to what extent that's solvable to Apple's standards given how the technology works.
Also strange to frame this as "caught flat footed" since Apple certainly had AI projects in the works that weren't ready for release. Apple is always like this, waiting for underlying technologies to mature to the point where they're ready to weave into a compelling, salable product rather than a neat technical demonstration or gizmo (which is where most AI is right now).
From what I've read they have trouble retaining people for their AI team because:
a.) The secretive nature of the company restricts them from being able to publish a lot of their work, and since most of these people are academics they really want to be able to publish.
b.) They're resistant to the kind of large scale data collection others use to train their models (and have only recently started to come around).
c.) The senior leadership just doesn't really grok the technology and doesn't have a great strategy for how to develop it. Consequently they just keep pivoting in various directions in an unfocused way that keeps them from making tangible progress.
Note that for a few events now, but especially this year’s WWDC, it felt like they’ve made an effort to actively avoid the phrase “AI”.
They announced many features using AI, large language models, and similar stuff (e.g. predictive autocorrection/sentence completion), but as far as I remember, they only ever exclusively used “Machine Learning” to describe the tech.
So this might be an attempt of theirs to avoid big “hype train: Apple does AI now too” press.
Edit: typo
Possibly. I think they're generally just very finnicky about branding things, so they'll want to call it something that resists comparison on raw specs to competitors. A good example is their branding of "Retina display" which is more of a statement of "here's what we want the display to achieve (pixel density at the limit of what the human eye can discern)" rather than "here's the resolution."
Not sure if paywall bypassing is allowed here but if it is here is a achieve link https://archive.ph/7tqs5
If not just remove my comment
Maybe off topic but can someone tell me what's the use of AI or machine learning for an ordinary person or layman?
I can understand that corporate can automate things and layoff some people, I'm not even talking about smaller businesses, but ordinary working class people who presently use things like Excel/Word. How can AI/ML benefit them?
As the tech become more ubiquitous and accessible, there will be ever more increasingly powerful tools available to the average person.
As an easy example, if you haven't tried feeding ChatGPT a draft of an email or work documentation and told it "make this sound more professional", I recommend it. That kind of thing is what it's really good at (take this, make it sound like this). I don't always use what it gives me back verbatim, but it's extremely helpful.
Less obvious examples are already starting to be integrated into software you already use. The text assist tools on some messaging apps that suggest entire phrases or sentences to respond with are less advanced predecessors to the kind of things we'll likely see in the next few years.
You’re going to love AI powered semantic search. You’ll be able to type “What was that email that a director sent me last quarter - something about emerging markets?” into a search bar and the correct result will pop up - even if none of those words are in the email.
I wouldn’t have expected it a couple of years back, but automatic subject tracing in some photo apps to crop them out more quickly/easily can be a fun feature. Similar for layman use: (“live”) text recognition in images, and object recognition in images, which can get quite crazy comparably quickly.
Imagine telling someone 20 years ago a palm-sized(ish) device will be able to differentiate and tell you which plant or flower is which, and pull up more info from encyclopedic entries alongside their names… And all in their free time, not in some crazy high-tech lab.
For more work-related things, as suggested already elsewhere here, text generation and re-phrasing can save some time, perhaps? Maybe data curation by just defining what you need and it’ll give you the cell/R formula?
And once there’s more programs interfacing with e.g. Excel/Word/whatever directly, these types of things might become one of many plug-ins that are easily added to an application.
Disclaimer, in the past I’ve felt like a bit of too much hype was going on as I had learnt about the basics of how ML works in around 2021, so before everyone started spouting “oh my god it creates sentences, thus it’s a step away from AGI”, but in recent weeks, I think I have come to be more… appreciative of what one can do the tech.
Insert "Old Man Yells At Cloud" meme here, but I gotta vent:
How the hell is this different than the 15 other "game-changing" tech things we've seen in the last 20 years? I'm old enough to remember MULTIPLE things being touted as the "end of _____" and the "future is now here" and they've all kinda...existed. They've either quietly integrated into everyday life or just fizzled out. EVs, WIFI, NFC, NFTs, QR codes, gyro controls a la Wii, etc. Some are interesting, some are useful, but nothing has "ended" anything else.
I think I'm just old and grumpy.
Trying to put it shortly. Part of it is the usual corporate hype train for a technological development, but speaking as a computer science adjacent person: I think the reason that a lot of people in that sphere are so jazzed about this is that LLMs have very suddenly and unexpectedly overcome sets of problems that they have been taught their whole lives are Very Hard, or maybe even impossible, to solve. As someone who isn't usually too easily impressed by new technologies, that aspect of it has shaken my world view a bit. It makes me feel that the future of technology in my lifetime is no longer projectable, and that society just isn't prepared for the implications. Part of this is also that it isn't just a single discrete technology that may or may not progress/revolutionise a single area, but an approach that can and will be turned towards pretty much any data modality and field of human endeavour we can think of. It's more akin to electricity than an electric vehicle.
I mean for just one example, WiFi is part of a larger technological innovation (manipulation of the EM spectrum to transmit information [inb4 smoke signals and lighthouses]) that has completely transformed the world and obsoleted lots of things.
Something like an NFT is completely different and hasn't accomplished anything.
EVs are somewhere in the middle as the tech is still nascent, but I'm willing to bet the transportation story will look very different in 50 years (part of a larger continuum of transportation tech in general).
This tech is not like other tech because it accelerates the other tech. It's a force multiplier. Everything knowledge-related that computers are used for is subject to this multiplication effect almost without exception.
Say you're a game designer - what's the biggest pain in the ass of that process? Art asset generation and voice acting, two things that are about to drop to zero cost in an equation they once dominated. 'Render me an apple' happens instantly, rather than you spending several hours creating that asset.
Here's a prompt for GPT27 to contemplate by way of example.
Combine and all of the assets and integrate all of the mods from all of the elder scrolls games into one unified worldspace. Alter the game so that characters can choose to start in any of the historical time periods covered by the game and time travel between them, with their changes to past worldspaces propagating forward with improvisation. Replace all of the NPCs with LLM models based on their original assets. Update all of the graphics to real world fidelity, and while you're at it, improve the aging scripting and graphics engines with a more performant and modern equivalent.Compare the amount of human effort that requires to an LLM pulling it off in a weekend on a home PC. That's what we are staring down the barrel of here, potentially.
Humans that once did this will now spend their time polishing AI output instead of doing it all themselves, they will work hundreds of times faster, and you'll need a fraction of a percent of the workforce for the same output. Apply this to all knowledge work, and then remember that this technology is not out of its diapers yet. In the future it will require less data and computing power to achieve vastly superior results, just like any technology.
That's not the biggest game changer, though. Making computers thousands of times more effective at their jobs is small potatoes compared to an LLM's ability to personalize everything. This technology's output varies based on who you are. That's new. The primary interface for interacting with computers shifts to spoken language with a universal translator provided by this same technology, and it will inevitably be able to relate to you better than other humans you have known your whole life.
Or, let me put it another way. The training data is hundreds of terabytes. The output, which contains the intelligence of that training data, and can recreate that same data as well as improvise on it, is single digit gigabytes that will run on your phone. If you look at this as a compression ratio, it vaporizes Weissman scores and outperforms Pied Piper's magic compression technology from 'Silicon Valley' by orders of magnitude... and it's so very much more than a mere compressor.