Indeed. In several remote ethnographic studies I did for organisations in the past year, few people are overly impressed with LLM tools and the common sentiment was "Too early. Maybe later." There is an increasing weariness with tech hype. These AI companies may be hurting themselves in the long run.
You have a very sobering and realistic outlook on things. Not just in this article but the handful of others I've read so far which is why I subscribed. Looking forward to seeing more of your analysis / takes on things which help me see reality more clearly. Thank you for your insights on matters that matter more than most.
I am involved in creative industries, with a strong background in visual arts. I am often nonplussed by how generic these images are, and how difficult it still remains to obtain something really original put of these tools. So I would agree with mr Merchant's analysis.
I mostly agree, but for many companies, they can pay $$$ and get something truly original or they can pay almost nothing and get something derivative. Most will choose the latter, I’m afraid.
I also work in these industries professionally. I don't know about "most" but definitely a decent chunk will at least take a couple of throws at it before they realize it doesn't actually pay off at scale.
If you have a budget there's basically no good reason to use these generators if you even care about the quality of your product a little bit.
It's important to interrogate AI evangelists exactly what they mean by "better". To me, it's gotten better at looking more like stock photography and video. Most jobs dont benefit from this. This tech is not developed by people who understand the industry they are ostensibly attempting to disrupt
I'm not really worried about the long term but in the short term things are probably going to be a little rough until company leadership learns that you can only cut so many corners. And if they don't, the people who still want to make things will find out a way to make things without them.
I don't create as my job but do a lot as a hobbyist and I agree with this sentiment. It's all just so "amazing" in an incredibly.boring way. It's just all so, soulless and weird looking in a way that is hard to put into words.
But, especially with the marketing, it's incredibly impressive to bean counter types who are not creatives.
Once past a very impressive demo in a localized context, I'm lost on where extending this becomes useful ... let alone a sustainable business. An AI-generated Giphy is hardly a compelling.
I might start willingly lending my attention when you extend the video length and add synchronized sound, dialog, and layer in storytelling and direction. But that's visual storytelling, and GAIs ability at even just written storytelling has been boringly average. By design, really.
Thus this feels like another photocopied page from the shock-and-awe-but-useless emerging tech playbook. Watching these videos, I feel like a robotics company just shared video of a robot doing parkour or dancing to The Contours. And for some odd reason I cannot come up with any valid reason why I or any business would want that... let alone pay them $400,000 for the pleasure.
Great perspective, as always! I really hope more people reporting on this kind of news learn a bit more about it before republishing these companies’ press releases.
It’s making me feel more and more like a pedant every day but it drives me crazy that we’re all just going along with calling this shit “AI” like that isn’t the biggest marketing tool they have. The double whammy of Open AI being neither “open” nor “AI” is maybe the worst of all.
Here’s hoping a few of us have learned a bit from the last few years! Thanks for the insight!
And tech is also managing to both be in full hype mode and treat its workers worse (layoffs, pay cuts, unreasonable targets).
Something very typical is happening: finance bros and MBA-educated executives and VC bankers are pushing the boom despite barely understanding the technology. They want to see “numbers go boom” without understanding much else. They have the money and the power so no one says anything, except under their breath.
The more I look at this hype cycle driven tech culture the more I wonder what it is that California has actually given humanity in terms of innovation besides a rapacious form of VC-funded marketing. It’s almost like they learnt from Hollywood how to create reality.
Agreed. It’s a shame so much of the rest of the world looks to California as a vision of the future. I see in it a desolate dystopia that isn’t even particularly interesting.
In defense of the rest of the world California all by its lonesome is something like the world's fifth largest economy. American media is still dominant, and filtered mostly through California, and until the past 15 years or so a lot of innovation did come from there.
The rot didn't set in fully until after 2008, though COVID and remote work becoming the standard has accelerated the decline.
My prediction is that Texas and various Southwestern and Southeastern states will suffer a similar rot and decline over the next decade as the worst actors in California are moving to those states dumb enough to think attracting racist billionaire scammers with stagnant ideas who are opposed to paying taxes is a positive thing.
I was warned away from moving to Florida by a veteran IT worker in 2008 when my employer offered a fat stack of cash to migrate. He called it 'the place where IT careers go to die.' His theory was the state was flooded with near retired or semi-retired IT pros and relatively few jobs, so the employers got IT staff on the cheap, and it drove down demand for younger workers.
Very good reference point with these examples. Thank you. The 'pervasive sense of inevitability' also is a very good way to characterise an important aspect of 'GPT-fever'.
The final paragraph's "if we accept" sadly probably should start with "we're going to accept (in the short term)" as there is little chance well-reasoned analysis like this is going to make a dent in our imagination. As a species, imagination is our forte (and the evolutionary advantage of our brains), reason far, far less so.
On the topic of stalling. ChatGPT usage may be stalling. But (a) it now has a competitor and the free version of BARD is at the level of the paid version of GPT (4). I don't know if that has been calculated in yet. More importantly, (b), it's not just 'use' that is stalling, performance of LLMs themselves has largely stalled and even more size hasn't done much. A lot of effort has been spent in 'programming' in/around these platforms, as in all the fine-tuning developed by OpenAI between 2019 and 2022, and all the 'engineering the hell around limitations' that is going on behind the scenes as well. Two explanations of the diminishing effect of sizing and the 'engineering the hell out of it are these: https://ea.rna.nl/2024/02/07/the-department-of-engineering-the-hell-out-of-ai/ and https://ea.rna.nl/2024/02/13/will-sam-altmans-7-trillion-ai-plan-rescue-ai/
I would like to see AI mimic the minds behind this post and Brian's (Merchant) combine them if you will, to evaluate any such technological trend or new roll out. Train the AI based on your 2 brains, THEN instruct it to perform and write an analysis of whatever the situation might call for. " It's IMPOSSIBLE for 2 like minds to come together without creating a THIRD. Earl Nightengale If we can do this somehow, sign me up for THAT Substack newsletter. I'll be very eager to hear all what it knows. We'll call the new publication "Deep Insights on Matters" that matter most.
As somewhat technical audience, I wasn't bothered too much.
It is technically incorrect (if you take 'stitching together' literally, which probably wasn't meant like that), but from a functional level there is really some of this. It is like GPT that doesn't have access to Shakespeare's Sonnet 18, but is perfectly capable of re-generating it to the letter. Not by copy-pasting, but because of the likelihood of token patterns. See https://ea.rna.nl/2023/12/26/memorisation-the-deep-problem-of-midjourney-chatgpt-and-friends/ No difference for other transformer-based systems, for a large part they work the same, just replace 'token embedding' bij 'patch embedding' (though patches are different in several other ways — I admit I am a bit hazy on the exact details here)
I'm not even sure why companies like Disney are pushing this.
They want to save money, I get that. But they're killing their own business model with this stuff.
If everyone can produce a Disney quality movie in their basement, then Disney has no business. No one will subscribe to their streaming service or go see their bland movies. They'll watch sora ai movies they make themselves.
I definitely agree with the socio-economic argument about the enormous powers these platform companies are amassing.However, “OpenAI’s CHATGPT is shedding users.. and have peaked” or “Some of the videos.. definitely aren’t good” doesn’t seem to be a strong argument for not getting dragged into a technology that is yet to reach a fraction of its potential.
“…this tool will unleash creativity and make the impossible possible…” that describes a human artist!
Indeed. In several remote ethnographic studies I did for organisations in the past year, few people are overly impressed with LLM tools and the common sentiment was "Too early. Maybe later." There is an increasing weariness with tech hype. These AI companies may be hurting themselves in the long run.
You have a very sobering and realistic outlook on things. Not just in this article but the handful of others I've read so far which is why I subscribed. Looking forward to seeing more of your analysis / takes on things which help me see reality more clearly. Thank you for your insights on matters that matter more than most.
I am involved in creative industries, with a strong background in visual arts. I am often nonplussed by how generic these images are, and how difficult it still remains to obtain something really original put of these tools. So I would agree with mr Merchant's analysis.
I mostly agree, but for many companies, they can pay $$$ and get something truly original or they can pay almost nothing and get something derivative. Most will choose the latter, I’m afraid.
I also work in these industries professionally. I don't know about "most" but definitely a decent chunk will at least take a couple of throws at it before they realize it doesn't actually pay off at scale.
If you have a budget there's basically no good reason to use these generators if you even care about the quality of your product a little bit.
It's important to interrogate AI evangelists exactly what they mean by "better". To me, it's gotten better at looking more like stock photography and video. Most jobs dont benefit from this. This tech is not developed by people who understand the industry they are ostensibly attempting to disrupt
I'm not really worried about the long term but in the short term things are probably going to be a little rough until company leadership learns that you can only cut so many corners. And if they don't, the people who still want to make things will find out a way to make things without them.
I don't create as my job but do a lot as a hobbyist and I agree with this sentiment. It's all just so "amazing" in an incredibly.boring way. It's just all so, soulless and weird looking in a way that is hard to put into words.
But, especially with the marketing, it's incredibly impressive to bean counter types who are not creatives.
Once past a very impressive demo in a localized context, I'm lost on where extending this becomes useful ... let alone a sustainable business. An AI-generated Giphy is hardly a compelling.
I might start willingly lending my attention when you extend the video length and add synchronized sound, dialog, and layer in storytelling and direction. But that's visual storytelling, and GAIs ability at even just written storytelling has been boringly average. By design, really.
Thus this feels like another photocopied page from the shock-and-awe-but-useless emerging tech playbook. Watching these videos, I feel like a robotics company just shared video of a robot doing parkour or dancing to The Contours. And for some odd reason I cannot come up with any valid reason why I or any business would want that... let alone pay them $400,000 for the pleasure.
Great perspective, as always! I really hope more people reporting on this kind of news learn a bit more about it before republishing these companies’ press releases.
It’s making me feel more and more like a pedant every day but it drives me crazy that we’re all just going along with calling this shit “AI” like that isn’t the biggest marketing tool they have. The double whammy of Open AI being neither “open” nor “AI” is maybe the worst of all.
Here’s hoping a few of us have learned a bit from the last few years! Thanks for the insight!
And tech is also managing to both be in full hype mode and treat its workers worse (layoffs, pay cuts, unreasonable targets).
Something very typical is happening: finance bros and MBA-educated executives and VC bankers are pushing the boom despite barely understanding the technology. They want to see “numbers go boom” without understanding much else. They have the money and the power so no one says anything, except under their breath.
The more I look at this hype cycle driven tech culture the more I wonder what it is that California has actually given humanity in terms of innovation besides a rapacious form of VC-funded marketing. It’s almost like they learnt from Hollywood how to create reality.
I think the VC tech bros know exactly what they're doing. They're running the same scam they've run several times over the past two decades or so.
Hype up some "groundbreaking" new tech which is actually a rehash of an old idea.
Create some flashy demos and prime the pump with VC money.
Get credulous suckers (fund managers and FOMO types) to pour money into the new thing.
Use that money to keep pumping up the hype and push the car uphill for as long as they can while taking a nice skim off the top for themselves.
When the business model for the big new things starts failing jump out of the way as the suckers get run over by the car running back downhill.
Repeat.
Agreed. It’s a shame so much of the rest of the world looks to California as a vision of the future. I see in it a desolate dystopia that isn’t even particularly interesting.
In defense of the rest of the world California all by its lonesome is something like the world's fifth largest economy. American media is still dominant, and filtered mostly through California, and until the past 15 years or so a lot of innovation did come from there.
The rot didn't set in fully until after 2008, though COVID and remote work becoming the standard has accelerated the decline.
My prediction is that Texas and various Southwestern and Southeastern states will suffer a similar rot and decline over the next decade as the worst actors in California are moving to those states dumb enough to think attracting racist billionaire scammers with stagnant ideas who are opposed to paying taxes is a positive thing.
Tesla just let go of 10% of it's work force. It was the largest employer in a town in Texas.
So it's already begun.
The ones who moved to Florida are even worse off.
I was warned away from moving to Florida by a veteran IT worker in 2008 when my employer offered a fat stack of cash to migrate. He called it 'the place where IT careers go to die.' His theory was the state was flooded with near retired or semi-retired IT pros and relatively few jobs, so the employers got IT staff on the cheap, and it drove down demand for younger workers.
Very good reference point with these examples. Thank you. The 'pervasive sense of inevitability' also is a very good way to characterise an important aspect of 'GPT-fever'.
The final paragraph's "if we accept" sadly probably should start with "we're going to accept (in the short term)" as there is little chance well-reasoned analysis like this is going to make a dent in our imagination. As a species, imagination is our forte (and the evolutionary advantage of our brains), reason far, far less so.
On the topic of stalling. ChatGPT usage may be stalling. But (a) it now has a competitor and the free version of BARD is at the level of the paid version of GPT (4). I don't know if that has been calculated in yet. More importantly, (b), it's not just 'use' that is stalling, performance of LLMs themselves has largely stalled and even more size hasn't done much. A lot of effort has been spent in 'programming' in/around these platforms, as in all the fine-tuning developed by OpenAI between 2019 and 2022, and all the 'engineering the hell around limitations' that is going on behind the scenes as well. Two explanations of the diminishing effect of sizing and the 'engineering the hell out of it are these: https://ea.rna.nl/2024/02/07/the-department-of-engineering-the-hell-out-of-ai/ and https://ea.rna.nl/2024/02/13/will-sam-altmans-7-trillion-ai-plan-rescue-ai/
I would like to see AI mimic the minds behind this post and Brian's (Merchant) combine them if you will, to evaluate any such technological trend or new roll out. Train the AI based on your 2 brains, THEN instruct it to perform and write an analysis of whatever the situation might call for. " It's IMPOSSIBLE for 2 like minds to come together without creating a THIRD. Earl Nightengale If we can do this somehow, sign me up for THAT Substack newsletter. I'll be very eager to hear all what it knows. We'll call the new publication "Deep Insights on Matters" that matter most.
Thank you for this, these are great points — and some good additional reading. Cheers
Can you, for once in ur life, stop spreading misinformation
Opinion is hardly misinformation.
“it’s automating the act of stitching together renderings of extant video and images.”
This is not true. These kind of statements may ring true to your non-technical audience but they erode the base of your other arguments
As somewhat technical audience, I wasn't bothered too much.
It is technically incorrect (if you take 'stitching together' literally, which probably wasn't meant like that), but from a functional level there is really some of this. It is like GPT that doesn't have access to Shakespeare's Sonnet 18, but is perfectly capable of re-generating it to the letter. Not by copy-pasting, but because of the likelihood of token patterns. See https://ea.rna.nl/2023/12/26/memorisation-the-deep-problem-of-midjourney-chatgpt-and-friends/ No difference for other transformer-based systems, for a large part they work the same, just replace 'token embedding' bij 'patch embedding' (though patches are different in several other ways — I admit I am a bit hazy on the exact details here)
I'm not even sure why companies like Disney are pushing this.
They want to save money, I get that. But they're killing their own business model with this stuff.
If everyone can produce a Disney quality movie in their basement, then Disney has no business. No one will subscribe to their streaming service or go see their bland movies. They'll watch sora ai movies they make themselves.
I definitely agree with the socio-economic argument about the enormous powers these platform companies are amassing.However, “OpenAI’s CHATGPT is shedding users.. and have peaked” or “Some of the videos.. definitely aren’t good” doesn’t seem to be a strong argument for not getting dragged into a technology that is yet to reach a fraction of its potential.