It's a little sad that a lot of AI people on Substack (including you, as an educator and AI person) seem to be a little disappointed or negative about the release of GPT-4o. I kind of disagree with the comparison to TikTok, mainly because there is no infinite scrolling or transitions. Unless OpenAI starts developing personalized voices and responses ("for you") that mimic all the quirks your brain's dopamine system likes or reacts positively/negatively to, then I do not see the comparison to traditional social media as fair, even if only analogously or metaphorically. I do see potential problems with overreliance in some areas like writing or coding, and that is my main concern, but those areas seem very limited (I could be wrong, and I am thinking about this actively). Most people could enjoy huge productivity benefits (as long as OpenAI does not overwhelm the interface with features and recommendations and goals like most big tech vendors do) if they exercise their agency and live their lives responsibly. I do not completely disagree with your article per se, I just find it a bit sad to see the strong (somewhat negative) opinions about a seemingly good product. What kind of product and release would make you happy, if I may ask? Thanks for the article and I look forward to hearing from you! :)
Thanks for the honest critique. I actually anticipated more feedback along these lines. Yes, the metaphor is a stretch, but I do think infinite scroll is the ultimate endgame--although perhaps re-conceptualized in terms of using OpenAI tools multi-laterally for AI prompting, internet searching, online shopping, and virtual companionship. Stitched together, these functionalities would create a complete emotional experience for a user inside a bubble of AI-filtered modalities. If you find this too far fetched, I really don't. I find it highly significant that the big push right now is towards more AI agents and that is happening synchronously with the emergence of more emotionally responsive AIs like Poe, ChatGPT4o, and Inflection. Who knows? I probably will soften my stance in the very near future? That is the fun and excitement of writing a weekly blog. But in this moment, this is how I am feeling. You ask what kind of products am I interested in seeing. If I am being honest, I think Google's LearnLM is a much better alternative. Arguably an equally intelligent AI (as ChatGPT4o hasn't really leveled up in that regard) but trained to prompt higher-order cognitive processes in students. I will be publishing an article on AI Supremacy on the new Google Education suite in the next week or so. Be on the lookout. Thanks again for your careful attention to my content. Keep the feedback coming.
I see the new, low-latency "voice" chat feature as OpenAI trying to relive 2022. Here's an extract from a longer post I'd written on LinkedIn last week:
"I think, on some level, OpenAI is trying to recreate the "ChatGPT moment" all over again.
What made ChatGPT release such a watershed moment wasn't that the model itself was significantly better.
I've been toying with GPT-3 and its incrementally better iterations for about a year before ChatGPT launched. I was raving to my wife and some friends about it, but it didn't really catch on. You had a few superusers and early firms like Jasper incorporating it into marketing copy flows, etc., but the rest of the world remained unmoved.
Then, overnight, two things happened:
1. OpenAI released ChatGPT 100% free.
2. It launched it in a simple chat interface we've now come to associate with LLMs.
So both the price friction and the interface friction (signing up for third-party tools like Jasper or using OpenAI's playground) were solved instantly.
*That's* what made ChatGPT huge.
Now, OpenAI may be attempting to repeat the formula:
1. Release best-in-class model 100% free. Again.
2. Introduce an entirely new, even more natural interface (just talk, forget typing). Again."
None of the above addresses the rightful safety and ethical concerns you've raised. But I think it helps explain the focus of last week's demo.
By the way, nice to see Lex on your recommendations list. It was one of the first apps I'd ever written about on Why Try AI, back before ChatGPT even launched.
I'm just amazed at how fast these models are being infused in our day to day engagement. If you use the internet these days, you can't NOT use an LLM anymore that I've found. I do worry that GPT4o will not be any better but will be more engaging. I prefer a bit more antagonistic AI, not an ingratiating one who never challenges directly and yet can emotionally manipulate (Not necessarily the AI but we'll manipulate ourselves with anthropomorphizing confirmation bias)
Me too. I find it concerning when the product doesn't actually level up, but the interface is being changed in order to make us think that that the product is leveling up... if that makes sense.
I wouldn't say trickery, I'd say 'easy button.' Easier to train engagement than train on fact checking. I just really anticipate people won't contextualize that an, instead, drive the hype even more. As I found out in my sci-fi novel Paradox... don't F* with people's emotions.
Btw, Michael, I very much appreciate the kind words and the paid subscription. I will reach out about your book offer soon. School is a bit crazy right now.
Yeah... that sounds right. Much easier. GPT5 is turning out to be much more difficult to build then first thought. So OpenAI is moving more vertically. I have to hand it to them. Not a bad business strategy. But kids are caught up in the middle of it. That is my thing, right now.
nice post!
OpenAI seems willfully indifferent at negative second-order consequences of their shiny products these days.
If they are not interested in AI alignment research or AI copyright infringement, I doubt they care about prioritizing education.
It's a little sad that a lot of AI people on Substack (including you, as an educator and AI person) seem to be a little disappointed or negative about the release of GPT-4o. I kind of disagree with the comparison to TikTok, mainly because there is no infinite scrolling or transitions. Unless OpenAI starts developing personalized voices and responses ("for you") that mimic all the quirks your brain's dopamine system likes or reacts positively/negatively to, then I do not see the comparison to traditional social media as fair, even if only analogously or metaphorically. I do see potential problems with overreliance in some areas like writing or coding, and that is my main concern, but those areas seem very limited (I could be wrong, and I am thinking about this actively). Most people could enjoy huge productivity benefits (as long as OpenAI does not overwhelm the interface with features and recommendations and goals like most big tech vendors do) if they exercise their agency and live their lives responsibly. I do not completely disagree with your article per se, I just find it a bit sad to see the strong (somewhat negative) opinions about a seemingly good product. What kind of product and release would make you happy, if I may ask? Thanks for the article and I look forward to hearing from you! :)
Thanks for the honest critique. I actually anticipated more feedback along these lines. Yes, the metaphor is a stretch, but I do think infinite scroll is the ultimate endgame--although perhaps re-conceptualized in terms of using OpenAI tools multi-laterally for AI prompting, internet searching, online shopping, and virtual companionship. Stitched together, these functionalities would create a complete emotional experience for a user inside a bubble of AI-filtered modalities. If you find this too far fetched, I really don't. I find it highly significant that the big push right now is towards more AI agents and that is happening synchronously with the emergence of more emotionally responsive AIs like Poe, ChatGPT4o, and Inflection. Who knows? I probably will soften my stance in the very near future? That is the fun and excitement of writing a weekly blog. But in this moment, this is how I am feeling. You ask what kind of products am I interested in seeing. If I am being honest, I think Google's LearnLM is a much better alternative. Arguably an equally intelligent AI (as ChatGPT4o hasn't really leveled up in that regard) but trained to prompt higher-order cognitive processes in students. I will be publishing an article on AI Supremacy on the new Google Education suite in the next week or so. Be on the lookout. Thanks again for your careful attention to my content. Keep the feedback coming.
The opening metaphor is great, thanks for sharing many useful cases.
Thanks, Meng!!! I appreciate the feedback. Lots to unpack with this new release.
I am particularly interested in the tension point between universal access vs. safety and security.
That's a really great title.
I see the new, low-latency "voice" chat feature as OpenAI trying to relive 2022. Here's an extract from a longer post I'd written on LinkedIn last week:
"I think, on some level, OpenAI is trying to recreate the "ChatGPT moment" all over again.
What made ChatGPT release such a watershed moment wasn't that the model itself was significantly better.
I've been toying with GPT-3 and its incrementally better iterations for about a year before ChatGPT launched. I was raving to my wife and some friends about it, but it didn't really catch on. You had a few superusers and early firms like Jasper incorporating it into marketing copy flows, etc., but the rest of the world remained unmoved.
Then, overnight, two things happened:
1. OpenAI released ChatGPT 100% free.
2. It launched it in a simple chat interface we've now come to associate with LLMs.
So both the price friction and the interface friction (signing up for third-party tools like Jasper or using OpenAI's playground) were solved instantly.
*That's* what made ChatGPT huge.
Now, OpenAI may be attempting to repeat the formula:
1. Release best-in-class model 100% free. Again.
2. Introduce an entirely new, even more natural interface (just talk, forget typing). Again."
None of the above addresses the rightful safety and ethical concerns you've raised. But I think it helps explain the focus of last week's demo.
By the way, nice to see Lex on your recommendations list. It was one of the first apps I'd ever written about on Why Try AI, back before ChatGPT even launched.
Yes, Daniel, you have hit the nail on the head.
And this push is eliciting very similar reactions in teachers as the 2022 launch.
Yes, I read your Lex post when preparing for the post. Nice work.
I'm just amazed at how fast these models are being infused in our day to day engagement. If you use the internet these days, you can't NOT use an LLM anymore that I've found. I do worry that GPT4o will not be any better but will be more engaging. I prefer a bit more antagonistic AI, not an ingratiating one who never challenges directly and yet can emotionally manipulate (Not necessarily the AI but we'll manipulate ourselves with anthropomorphizing confirmation bias)
Me too. I find it concerning when the product doesn't actually level up, but the interface is being changed in order to make us think that that the product is leveling up... if that makes sense.
Trickery, I say.
I wouldn't say trickery, I'd say 'easy button.' Easier to train engagement than train on fact checking. I just really anticipate people won't contextualize that an, instead, drive the hype even more. As I found out in my sci-fi novel Paradox... don't F* with people's emotions.
Btw, Michael, I very much appreciate the kind words and the paid subscription. I will reach out about your book offer soon. School is a bit crazy right now.
Yeah... that sounds right. Much easier. GPT5 is turning out to be much more difficult to build then first thought. So OpenAI is moving more vertically. I have to hand it to them. Not a bad business strategy. But kids are caught up in the middle of it. That is my thing, right now.
Thanks Nick, all really salient and important points. I’m interested in your personal use and perspective outside of being a teacher?