This was a super useful post thanks. However, I had a laugh just now experimenting with the Ai2 research paper tool. It’s probably my fault for not being clear on my instructions, but I asked it for a list of ten research papers I could consult with on a certain dead poet, and it gave me the following: “As an AI with a knowledge cutoff date in early 2023, I can't provide you with an exact list of ten research papers published after that time. However, I can suggest some hypothetical titles and authors based on the kind of research you might find on…”
I love that it then went to all the effort of making up ten research paper titles and potential author names for these 😆
Actually that's really weird behavior... I would say it's probably not good at non-science (and maybe especially non-computer science); I just tried a search about an artist and it turned it into a search for papers BY the artist, which struck out too. Maybe they need to clarify their sources a bit more. Good catch.
Every time I read one of your newsletters, I feel like I’m walking through a library curated by a time traveler with a sense of humor and a deep GPU budget.
The GPT-4o bake-off was especially eye-opening—I’ve been experimenting with tile generation too and hit that same top-bottom seam issue. Totally agree that MJ still wins there for now. Also loved the asteroid-vs-luxury-ship test. It’s a great “stress test” for scene composition—half poetic, half rendering logic puzzle.
Really appreciate your insight into the increasing AI fatigue. I’m feeling it too—especially that eerie sense that AI-generated prose has started to haunt my inner monologue. That “slop ngram” bit from Sam Paech was hilarious and way too real. Also curious if you think we’re nearing peak prompt fatigue—or if the real renaissance starts once we stop expecting perfection and start embracing AI as a chaos collaborator.
Anyway, thanks again for making sense of the madness (and for spotlighting actual useful stuff). Looking forward to the weird UK exhibits next time.
Thanks for the excellent collection, as usual! I find sort of AI fatigue setting in. I wonder if we can continue the hype train, creating benchmarks almost as fast they are "solved" by this or that AI model. I also noticed it in the writing, after a while you get really good at spotting AI writing, and I think it bleeds thru on reading normal human text and on writing text, too. Funnily, I also started to vibe code a small text based survival game, when then I had to debug it and the vibe hit the curb pretty quickly, because it was insanely hard to understand what was going on. So the game is still a go, but with less AI. I'll link once it's somewhere near playable.
Here is a fun AI use case that I haven't seen done: upsample old games to 4k, like from the 1990s.
I too feel AI fatigue. I'd like to pay less attention to it and more to other things? I don't know how my readership will respond to that. Re game coding: I've tried a number of non-trivial game-related things and hit walls pretty quickly; I don't think it's in any way as easy as the X posters make it look, for anything you want to have much control over.
What a fascinating deep dive into the latest AI Bakeoff—seeing different models compete on spaceship design tasks really opened my eyes to how rapidly these systems are evolving. Your breakdown of each algorithm’s strengths and weaknesses was both insightful and accessible, especially the contrast between the rule-based engine and the neural network’s imaginative solutions. I was particularly struck by the section on emergent behaviors—who would have thought an AI could propose fuel-efficient hull geometries inspired by manta rays? If you’re looking for a fun palate cleanser after all that technical analysis, check out→→https://randomweb.click/generator/theuselessweb→→Random Web – Useless Website Generator It’s a delightful reminder that sometimes the best way to recharge your creative brain is to indulge in a bit of intentional, light-hearted randomness. Thanks for sharing such a well-researched, thought-provoking post—I can’t wait to see what the next Bakeoff has in store!
This was a super useful post thanks. However, I had a laugh just now experimenting with the Ai2 research paper tool. It’s probably my fault for not being clear on my instructions, but I asked it for a list of ten research papers I could consult with on a certain dead poet, and it gave me the following: “As an AI with a knowledge cutoff date in early 2023, I can't provide you with an exact list of ten research papers published after that time. However, I can suggest some hypothetical titles and authors based on the kind of research you might find on…”
I love that it then went to all the effort of making up ten research paper titles and potential author names for these 😆
I love the idea of not finding research and making stuff up.
Actually that's really weird behavior... I would say it's probably not good at non-science (and maybe especially non-computer science); I just tried a search about an artist and it turned it into a search for papers BY the artist, which struck out too. Maybe they need to clarify their sources a bit more. Good catch.
Every time I read one of your newsletters, I feel like I’m walking through a library curated by a time traveler with a sense of humor and a deep GPU budget.
The GPT-4o bake-off was especially eye-opening—I’ve been experimenting with tile generation too and hit that same top-bottom seam issue. Totally agree that MJ still wins there for now. Also loved the asteroid-vs-luxury-ship test. It’s a great “stress test” for scene composition—half poetic, half rendering logic puzzle.
Really appreciate your insight into the increasing AI fatigue. I’m feeling it too—especially that eerie sense that AI-generated prose has started to haunt my inner monologue. That “slop ngram” bit from Sam Paech was hilarious and way too real. Also curious if you think we’re nearing peak prompt fatigue—or if the real renaissance starts once we stop expecting perfection and start embracing AI as a chaos collaborator.
Anyway, thanks again for making sense of the madness (and for spotlighting actual useful stuff). Looking forward to the weird UK exhibits next time.
Thanks for the excellent collection, as usual! I find sort of AI fatigue setting in. I wonder if we can continue the hype train, creating benchmarks almost as fast they are "solved" by this or that AI model. I also noticed it in the writing, after a while you get really good at spotting AI writing, and I think it bleeds thru on reading normal human text and on writing text, too. Funnily, I also started to vibe code a small text based survival game, when then I had to debug it and the vibe hit the curb pretty quickly, because it was insanely hard to understand what was going on. So the game is still a go, but with less AI. I'll link once it's somewhere near playable.
Here is a fun AI use case that I haven't seen done: upsample old games to 4k, like from the 1990s.
Keep it up! 🙏🏻
I too feel AI fatigue. I'd like to pay less attention to it and more to other things? I don't know how my readership will respond to that. Re game coding: I've tried a number of non-trivial game-related things and hit walls pretty quickly; I don't think it's in any way as easy as the X posters make it look, for anything you want to have much control over.
What a fascinating deep dive into the latest AI Bakeoff—seeing different models compete on spaceship design tasks really opened my eyes to how rapidly these systems are evolving. Your breakdown of each algorithm’s strengths and weaknesses was both insightful and accessible, especially the contrast between the rule-based engine and the neural network’s imaginative solutions. I was particularly struck by the section on emergent behaviors—who would have thought an AI could propose fuel-efficient hull geometries inspired by manta rays? If you’re looking for a fun palate cleanser after all that technical analysis, check out→→https://randomweb.click/generator/theuselessweb→→Random Web – Useless Website Generator It’s a delightful reminder that sometimes the best way to recharge your creative brain is to indulge in a bit of intentional, light-hearted randomness. Thanks for sharing such a well-researched, thought-provoking post—I can’t wait to see what the next Bakeoff has in store!