Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

call could not be complete as dialled 📞 Before @krea 2 was open-sourced I’d pretty much abandoned all image models apart from GPT 2.0. It’s such a fun model to train and the spectrum of what it can do seems pretty expansive - like this live action / thermal... show more

Ingi Erlingsson 🪄

14,442 subscribers

13,668 views • 1 month ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

I topped up $5 on an API aggregator ToAPIs Then I found out GPT Image 2 costs only around $0.015 per image. If you do a lot of testing or batch-generate commercial AI images, that difference adds up fast. I think I just found the secret to generating more, testing more, and spending less. And it’s not just one model. With the same key, you can access 50+ models for image, video, and text, including GPT Image 2, Gemini Omni, Seedance 2.0, Kling AI 3.0, grok-video-1.5-preview, and more. Some models are priced up to 80% lower than official platforms. Just top up and test what you need: Made on ToAPIs with GPT Image 2 + Seedance 2.0

I topped up $5 on an API aggregator ToAPIs Then I found out GPT Image 2 costs only around $0.015 per image. If you do a lot of testing or batch-generate commercial AI images, that difference adds up fast. I think I just found the secret to generating more, testing more, and spending less. And it’s not just one model. With the same key, you can access 50+ models for image, video, and text, including GPT Image 2, Gemini Omni, Seedance 2.0, Kling AI 3.0, grok-video-1.5-preview, and more. Some models are priced up to 80% lower than official platforms. Just top up and test what you need: Made on ToAPIs with GPT Image 2 + Seedance 2.0

Shami

22,991 views • 2 months ago

GPT-5.5 by Reasoning Effort: I've asked it in Codex to create a physics-based visualisation of RL cycles for different sized models (70b, 1t, 10t), to demonstrate how the amount of RL you can do differs by model size. My assessment of each: - Low: weird slop - Medium: kinda cooked - High: sort of tried but ultimately incoherent - Extra High: elite - really nice idea and well executed Obviously this is just one shot, but worth trying different reasoning levels for the new models, medium seems to be pretty good for GPT-5.5 and it was really bad for many previous GPT models.

GPT-5.5 by Reasoning Effort: I've asked it in Codex to create a physics-based visualisation of RL cycles for different sized models (70b, 1t, 10t), to demonstrate how the amount of RL you can do differs by model size. My assessment of each: - Low: weird slop - Medium: kinda cooked - High: sort of tried but ultimately incoherent - Extra High: elite - really nice idea and well executed Obviously this is just one shot, but worth trying different reasoning levels for the new models, medium seems to be pretty good for GPT-5.5 and it was really bad for many previous GPT models.

Peter Gostev (SF: 22-26 June)

209,258 views • 4 months ago

It’s Saturday morning so going to unabashedly crowdsource. If any of the much smarter than me homeowners out there can listen to this sound, it’s coming from our attic periodically and sounds like a man with a leaf blower is just hanging out up there. No AC/Heat is on. It’s in a pretty inaccessible part of the attic. Local to this area of the home. Completely confused as to what it could be. Sounds like a pump or motor.

It’s Saturday morning so going to unabashedly crowdsource. If any of the much smarter than me homeowners out there can listen to this sound, it’s coming from our attic periodically and sounds like a man with a leaf blower is just hanging out up there. No AC/Heat is on. It’s in a pretty inaccessible part of the attic. Local to this area of the home. Completely confused as to what it could be. Sounds like a pump or motor.

Andy Holloway

108,698 views • 2 years ago

No matter how much we try, multiple-opponent scenes always seem to be challenging for AI video models. Let's see if the upcoming Seedance 2.1 / 2.5 next week can finally improve this. What do you think about the petal effect? I'm considering using it in future videos as well. Midjourney + Gpt Image 2 + Seedance 2.0 You can find the prompts in the first reply.

No matter how much we try, multiple-opponent scenes always seem to be challenging for AI video models. Let's see if the upcoming Seedance 2.1 / 2.5 next week can finally improve this. What do you think about the petal effect? I'm considering using it in future videos as well. Midjourney + Gpt Image 2 + Seedance 2.0 You can find the prompts in the first reply.

Kōda

16,206 views • 2 months ago

GPT Image 2 + Seedance 2.0 Prompt Share Tried using IPA + FACS for a contemporary performance piece where live singing and choreography happen simultaneously. Not sure how much of it actually worked technically but I think the final result turned out pretty interesting. I also ended up color-coding the storyboard annotations because otherwise annotations were confusing the model. Different colors helped separate body movement, camera motion, framing and lighting directions from the actual environment and character drawings. You can find color-coded storyboard prompt + seedance 2.0 prompt in the replies.

GPT Image 2 + Seedance 2.0 Prompt Share Tried using IPA + FACS for a contemporary performance piece where live singing and choreography happen simultaneously. Not sure how much of it actually worked technically but I think the final result turned out pretty interesting. I also ended up color-coding the storyboard annotations because otherwise annotations were confusing the model. Different colors helped separate body movement, camera motion, framing and lighting directions from the actual environment and character drawings. You can find color-coded storyboard prompt + seedance 2.0 prompt in the replies.

Kōda

147,637 views • 3 months ago

NEW SECRET MYSTERY AI VIDEO MODEL ARENA CODE NAME: BOUNCYBOHR Unlike Artificial Analysis, on Arena. AI you can submit your own prompts and this prompt is designed to try and unmask who created the model. Models powered by an LLM (like Seedance 2.0 and Omni) are smart and can see what you are doing. As far as thoughts? Eh? While another screenshot looked like Omni, this does look pretty Seedance to me. Judging by the result, this looks like the rumored new "improved" FAST model and not upgraded full 2.1 model that is supposed to be 25% better. Text To Video close-up cinematic scene from a film, professionally directed, focused, ultrafine details a woman laughs as she whiteboard writes out what model made this video

NEW SECRET MYSTERY AI VIDEO MODEL ARENA CODE NAME: BOUNCYBOHR Unlike Artificial Analysis, on Arena. AI you can submit your own prompts and this prompt is designed to try and unmask who created the model. Models powered by an LLM (like Seedance 2.0 and Omni) are smart and can see what you are doing. As far as thoughts? Eh? While another screenshot looked like Omni, this does look pretty Seedance to me. Judging by the result, this looks like the rumored new "improved" FAST model and not upgraded full 2.1 model that is supposed to be 25% better. Text To Video close-up cinematic scene from a film, professionally directed, focused, ultrafine details a woman laughs as she whiteboard writes out what model made this video

Brent Lynch

32,504 views • 2 months ago

Black Forest Labs just announced FLUX 3: a unified multimodal model for image, video, audio and action prediction. Founded in Freiburg, Germany, one of the few globally significant European companies in the AI sector. The current release is gated early access, not open source. But if FLUX 3 Dev actually ships with usable open weights, this could become one of the most important open-model releases in generative video yet. Especially after the Chinese Minimax H3 release. BFL claims, based on preliminary internal comparisons, that FLUX 3 outperformed models such as Runway Gen-4.5, Luma Ray 3.2, Kling 3 Pro, and Seedance 2.0. However, these figures stem from the early-access phase and do not constitute independent validation. Be that as it may, it's great to see the quality that today's video models can produce at affordable prices. Something that would have been unthinkable a year ago. Really cool release!

Black Forest Labs just announced FLUX 3: a unified multimodal model for image, video, audio and action prediction. Founded in Freiburg, Germany, one of the few globally significant European companies in the AI sector. The current release is gated early access, not open source. But if FLUX 3 Dev actually ships with usable open weights, this could become one of the most important open-model releases in generative video yet. Especially after the Chinese Minimax H3 release. BFL claims, based on preliminary internal comparisons, that FLUX 3 outperformed models such as Runway Gen-4.5, Luma Ray 3.2, Kling 3 Pro, and Seedance 2.0. However, these figures stem from the early-access phase and do not constitute independent validation. Be that as it may, it's great to see the quality that today's video models can produce at affordable prices. Something that would have been unthinkable a year ago. Really cool release!

Chubby♨️

39,869 views • 22 days ago

this pretty much sums it up. from brunches not even a third filled to sold out venues all over the country, it’s been the best feeling watching you take over the world and thank you times a million for allowing me to be even a tiny part of your drag journey Kori “Cameos are open” King 💕

this pretty much sums it up. from brunches not even a third filled to sold out venues all over the country, it’s been the best feeling watching you take over the world and thank you times a million for allowing me to be even a tiny part of your drag journey Kori “Cameos are open” King 💕

ellie

35,678 views • 10 months ago

Some thoughts while training this morning: 1) I actually don’t like getting up to train at 6am. But I do what needs to be done, and with a young family I want to spend time with I’d rather take time away from myself and train while they sleep 2) See your training as a meeting with yourself; this is an idea you might have seen thrown about. I put all my sessions in a calendar - when it’s in black and white it makes it easier to just get it done. Mood always follows action, too Have a good week friends 🫡🌱

Some thoughts while training this morning: 1) I actually don’t like getting up to train at 6am. But I do what needs to be done, and with a young family I want to spend time with I’d rather take time away from myself and train while they sleep 2) See your training as a meeting with yourself; this is an idea you might have seen thrown about. I put all my sessions in a calendar - when it’s in black and white it makes it easier to just get it done. Mood always follows action, too Have a good week friends 🫡🌱

Jeffrey Boadi

46,514 views • 9 months ago

AI video is waaaaay better than most people realize. I came with this stupid idea for a terrible PS1 game with hideous textures and a jumping catfish... 30 minutes later, I had this. It's almost exactly what I had in my head. Between Nano Banana, GPT Image 2 and Seedance 2 you can pretty much create anything you can think of, with more granular control than people realize. It's not perfect but it's VERY good and I'm really excited to see the wild shit that new creators do with it!

AI video is waaaaay better than most people realize. I came with this stupid idea for a terrible PS1 game with hideous textures and a jumping catfish... 30 minutes later, I had this. It's almost exactly what I had in my head. Between Nano Banana, GPT Image 2 and Seedance 2 you can pretty much create anything you can think of, with more granular control than people realize. It's not perfect but it's VERY good and I'm really excited to see the wild shit that new creators do with it!

Finn McKenty

10,979 views • 3 months ago

When ac sent me an incredible track she was working on called “(Isn’t It) Obvious” and asked if I’d be a part of it, I knew it would be fun to play on. What I didn’t see coming, however, was how open-minded she’d be when it came to a guitar solo. Guitars on pop songs usually get approached pretty conservatively, but Alessia sat across from me and gave me so much creative freedom that I ended up playing things I never have before. She kept encouraging me to explore further, and the result is a really striking musical moment, and an exceptional pop song with a futuristic bent… and that’s what it’s all about. Give the song a listen, and enjoy the videos of what I really sound like when I’m trying to “land a new trick.” Thanks Alessia for the push to go further. You’re a true artist. ♥️

When ac sent me an incredible track she was working on called “(Isn’t It) Obvious” and asked if I’d be a part of it, I knew it would be fun to play on. What I didn’t see coming, however, was how open-minded she’d be when it came to a guitar solo. Guitars on pop songs usually get approached pretty conservatively, but Alessia sat across from me and gave me so much creative freedom that I ended up playing things I never have before. She kept encouraging me to explore further, and the result is a really striking musical moment, and an exceptional pop song with a futuristic bent… and that’s what it’s all about. Give the song a listen, and enjoy the videos of what I really sound like when I’m trying to “land a new trick.” Thanks Alessia for the push to go further. You’re a true artist. ♥️

John Mayer

210,448 views • 1 year ago

David Dobrik’s advice if you’re feeling stuck right now “Time’s gonna move on regardless and you could be in a crazy different spot in a year from right now” “All you have to do is start and you’re gonna be at the end of the year no matter what” “It’s up to you whether you want it to be a good year or just the same as it was before”

David Dobrik’s advice if you’re feeling stuck right now “Time’s gonna move on regardless and you could be in a crazy different spot in a year from right now” “All you have to do is start and you’re gonna be at the end of the year no matter what” “It’s up to you whether you want it to be a good year or just the same as it was before”

Jack

90,018 views • 4 months ago

Midjourney sref + Sora 2 Pro is the sauce. With one Midjourney style image, you can give a specific style for your entire project. I created two different 12-second clips and edited them together. Some details aren’t fully consistent, like the iPod or AirPods because the clips were made separately from a single image (Character in a specific style). It could be fixed in post-production, but that would take more time, and this was more of an experimental test. It would be great to add the actual product image with the current one to maintain product consistency. I feel like if there were a way to add 2–4 images into this workflow, it could open up a lot more possibilities and consistency. With an API, it could be possible. Or let’s see what Veo 3.1 has to offer.

Midjourney sref + Sora 2 Pro is the sauce. With one Midjourney style image, you can give a specific style for your entire project. I created two different 12-second clips and edited them together. Some details aren’t fully consistent, like the iPod or AirPods because the clips were made separately from a single image (Character in a specific style). It could be fixed in post-production, but that would take more time, and this was more of an experimental test. It would be great to add the actual product image with the current one to maintain product consistency. I feel like if there were a way to add 2–4 images into this workflow, it could open up a lot more possibilities and consistency. With an API, it could be possible. Or let’s see what Veo 3.1 has to offer.

Allar Haltsonen

10,141 views • 10 months ago

Gpt Image 2 + Seedance 2.0 - Prompt Share Meet Ezreal, she is not the kind of butcher you'd find in your neighborhood. Gpt Image 2 can sometimes struggle with continuity in storyboard panels and may occasionally render some things incorrectly. Creating a clean storyboard often requires a few iterations or some manual panel edits afterward. Small logic or continuity errors in the storyboard can easily carry over into the final video and there are some minor details like that in this one as well. That's why it's important to review storyboard generations carefully before moving on to animation. You can find the prompts in the replies.

Gpt Image 2 + Seedance 2.0 - Prompt Share Meet Ezreal, she is not the kind of butcher you'd find in your neighborhood. Gpt Image 2 can sometimes struggle with continuity in storyboard panels and may occasionally render some things incorrectly. Creating a clean storyboard often requires a few iterations or some manual panel edits afterward. Small logic or continuity errors in the storyboard can easily carry over into the final video and there are some minor details like that in this one as well. That's why it's important to review storyboard generations carefully before moving on to animation. You can find the prompts in the replies.

Kōda

19,307 views • 2 months ago

•The false start call on John Michael Schmitz. •The non-call for pass interference to Wan’Dale Robinson. •The non-call for pass interference on the final play of the game in the end zone. I’m not one to blame officials for losing games, but when it’s obvious, sometimes, it does need to be pointed out. Like the non-pass interference call to Darren Waller in the Week 6 game vs the Bills that ended the game. That was as blatantly bad as it got. In the Giants 33-25 loss yesterday to the Eagles, the officials were pretty bad to where several non-calls were that obvious. “I think there were some flags that could have been thrown, but that is not our place. That is not fair. If it’s a call, call it. It shouldn’t be a last-play mentality. If it’s a call, and if it’s blatant, then it should be called. It should be honored like it is the first play of the game.”- Giants QB Tyrod Taylor. “I’d say, any official questions, you can ask those guys.”- Giants head coach Brian Daboll, who clearly looked angry when he entered the press conference room, probably knew he’d get a massive fine if he said what he wanted to, so he kept it simple.

•The false start call on John Michael Schmitz. •The non-call for pass interference to Wan’Dale Robinson. •The non-call for pass interference on the final play of the game in the end zone. I’m not one to blame officials for losing games, but when it’s obvious, sometimes, it does need to be pointed out. Like the non-pass interference call to Darren Waller in the Week 6 game vs the Bills that ended the game. That was as blatantly bad as it got. In the Giants 33-25 loss yesterday to the Eagles, the officials were pretty bad to where several non-calls were that obvious. “I think there were some flags that could have been thrown, but that is not our place. That is not fair. If it’s a call, call it. It shouldn’t be a last-play mentality. If it’s a call, and if it’s blatant, then it should be called. It should be honored like it is the first play of the game.”- Giants QB Tyrod Taylor. “I’d say, any official questions, you can ask those guys.”- Giants head coach Brian Daboll, who clearly looked angry when he entered the press conference room, probably knew he’d get a massive fine if he said what he wanted to, so he kept it simple.

Doug Rush

171,398 views • 2 years ago

Again, I could be coping and this is just my speculation. #EscapefromTarkov BSG has posted downtime for 3 hours a bit ago. Whereas this seems like a reasonable maintenance, I think it could be 'Actual' Kolotun event. Nikita loves saying 'we love to give you surprises' and i think we are getting one soon. Speculation point #1. We have not yet unlock the New achievement. (image 1 - makes me think we might get super blizzard, etc.) Speculation point #2. There's cultist on top of the church of the poster for the kolotun event, indicating we still need to go to church & meet cultist (image 2, could be red herring tho) Speculation point #3 As i said earlier, nikita wants surprises. In order to do that, they dropped not too much entertaining event quests as of now, To maximize further incoming BIG EVENT. They've done this before. Speculation point #4 the videos nikita keeps posting could be anything, but most of visual videos he done was related to actual game or in-game event. He has been posting quite frequently. (clip 1 & 2) Hope this is BIG FUN EVENT! 🫡

Again, I could be coping and this is just my speculation. #EscapefromTarkov BSG has posted downtime for 3 hours a bit ago. Whereas this seems like a reasonable maintenance, I think it could be 'Actual' Kolotun event. Nikita loves saying 'we love to give you surprises' and i think we are getting one soon. Speculation point #1. We have not yet unlock the New achievement. (image 1 - makes me think we might get super blizzard, etc.) Speculation point #2. There's cultist on top of the church of the poster for the kolotun event, indicating we still need to go to church & meet cultist (image 2, could be red herring tho) Speculation point #3 As i said earlier, nikita wants surprises. In order to do that, they dropped not too much entertaining event quests as of now, To maximize further incoming BIG EVENT. They've done this before. Speculation point #4 the videos nikita keeps posting could be anything, but most of visual videos he done was related to actual game or in-game event. He has been posting quite frequently. (clip 1 & 2) Hope this is BIG FUN EVENT! 🫡

VeryBadSCAV

125,281 views • 8 months ago

LEG EXTENSIONS BEFORE SQUATS Here’s why: If you do your set(s) of Hack Squats BEFORE you do your set(s) of Leg Extensions, you will see a pretty large decline in your output on Leg Extensions….it might be something like 20-25% less than what you could do fresh On the contrary, if you do your set(s) of Leg Extensions BEFORE you do your set(s) of Hack Squats, you will NOT see nearly as large a decline in Hack Squat performance comparatively….the decline in Hack Squat performance may only be 5-10% less than what you could do fresh This is mostly due to the resistance profile/overall demands of each exercise and it’s the main reason I often program Leg Extensions before any sort of Squat pattern on Lower days in my training plans A subtle but GOATED fatigue management play! (Btw…you can easily verify this for yourself by A/B testing it)

Dean Turner

103,041 views • 2 months ago

🚨 JUST IN: THIS FREE TOOL JUST REPLACED FOUR AI IMAGE AND VIDEO SUBSCRIPTIONS AT ONCE. Midjourney. Krea. Higgsfield. Openart. One repo. 200+ models. Zero dollars a month. Here is what it actually does. It is a full image and video studio that runs in your browser or as a desktop app. Text to image, image to image, text to video, image to video, lip sync, cinema mode with real camera controls. All of it. 4,500 people already starred this. What you get for free: → 50+ image models including Flux, Midjourney v7, Ideogram, GPT-4o, Seedream → 60+ video models including Kling, Sora, Veo, Runway, Wan, Hailuo → lip sync studio with 9 dedicated models. upload a portrait and audio and it talks → cinema studio with real camera controls. lens, focal length, aperture, film stock → feed up to 14 reference images into one generation → self-hosted. your data never leaves your machine The crazy part is there is also a hosted version that needs zero setup. Just open the link and start generating. Now the math. Midjourney Standard: $30/month Krea AI Pro: $30/month Higgsfield Plus: $49/month Openart AI: $15/month That is $124 a month. $1,488 a year. This repo does everything all four do. With more models than any of them. For free. Forever. No subscription. No vendor lock-in. MIT licensed. Download it in one click on Mac or Windows. Someone should have told me about this sooner. I feel like an idiot. ( save this )

🚨 JUST IN: THIS FREE TOOL JUST REPLACED FOUR AI IMAGE AND VIDEO SUBSCRIPTIONS AT ONCE. Midjourney. Krea. Higgsfield. Openart. One repo. 200+ models. Zero dollars a month. Here is what it actually does. It is a full image and video studio that runs in your browser or as a desktop app. Text to image, image to image, text to video, image to video, lip sync, cinema mode with real camera controls. All of it. 4,500 people already starred this. What you get for free: → 50+ image models including Flux, Midjourney v7, Ideogram, GPT-4o, Seedream → 60+ video models including Kling, Sora, Veo, Runway, Wan, Hailuo → lip sync studio with 9 dedicated models. upload a portrait and audio and it talks → cinema studio with real camera controls. lens, focal length, aperture, film stock → feed up to 14 reference images into one generation → self-hosted. your data never leaves your machine The crazy part is there is also a hosted version that needs zero setup. Just open the link and start generating. Now the math. Midjourney Standard: $30/month Krea AI Pro: $30/month Higgsfield Plus: $49/month Openart AI: $15/month That is $124 a month. $1,488 a year. This repo does everything all four do. With more models than any of them. For free. Forever. No subscription. No vendor lock-in. MIT licensed. Download it in one click on Mac or Windows. Someone should have told me about this sooner. I feel like an idiot. ( save this )

Kanika

14,769 views • 4 months ago

SOMEONE GOT TIRED OF PAYING HIGGSFIELD AI'S SUBSCRIPTION SO HE REBUILT THE WHOLE THING AND OPEN-SOURCED IT 200+ models. text-to-image, image-to-image, text-to-video, image-to-video all in one interface you configure a virtual camera in the Cinema Studio. pick the body, the lens, the focal length, the aperture and it writes the optimized cinematic prompt for you. completely in the background you never touch the camera keywords. you just set up the shot like a real cinematographer would Kling v3, Sora 2, Veo 3, Flux Dev, Midjourney v7, GPT-4o, Seedream 5.0, Runway Gen-3 all in there self-hosted. MIT licensed. runs on your machine. your data stays local the only thing you pay for is the model API calls themselves someone built this so you never have to pay Higgsfield AI again

SOMEONE GOT TIRED OF PAYING HIGGSFIELD AI'S SUBSCRIPTION SO HE REBUILT THE WHOLE THING AND OPEN-SOURCED IT 200+ models. text-to-image, image-to-image, text-to-video, image-to-video all in one interface you configure a virtual camera in the Cinema Studio. pick the body, the lens, the focal length, the aperture and it writes the optimized cinematic prompt for you. completely in the background you never touch the camera keywords. you just set up the shot like a real cinematographer would Kling v3, Sora 2, Veo 3, Flux Dev, Midjourney v7, GPT-4o, Seedream 5.0, Runway Gen-3 all in there self-hosted. MIT licensed. runs on your machine. your data stays local the only thing you pay for is the model API calls themselves someone built this so you never have to pay Higgsfield AI again

Rimsha Bhardwaj

101,241 views • 4 months ago

introducing a new, very fun, LLM benchmark- the Game-of-Life Bench! the rules are simple: given an 8x8 grid following Conway's game of life rules, the goal is to create an initial pattern with at most 32 cells that can last the longest number of turns before dying/repeating. some results to highlight (with caveats detailed below): - gpt 5.1 lasts the longest with a 106 step run - claude models are really bad at this! they refuse to reason about this task and score < 25 points - deepseek r1 is the best open model with 102 steps. why? because i wanted to create a benchmark that has (i think) no practicality, but is still fun to look at, cheap, and still measures something interesting. i also am a big fan of the game of life. its absurdly simple rules leading to intractability is extremely cool to me. also, i saw a lot of work with LLMs trying to "predict" the next state in Conway's game of life, I think game-of-life bench is more fun because it's pretty open ended and only asks the LLM for the initial state. I also think this could be an RL env? but idk why you would ever train on this task haha i don't think this is a "serious" benchmark because it doesnt measure anything practical, but i still think it's a hard benchmark exactly because you can't predict what happens with your initial state many turns into the future; this is why i was initially expecting all LLMs to be bad at it, but turns out, some are clearly better than the others (the ordering may surprise you!) reminder: this is still a work-in-progress; (1) i am gpu-poor so could only do 10 runs for each model, even though total running cost is relatively low. maybe with some more credits i can run more seeds for each model. (2) i handpicked models which i think are at the frontier right now, plus some others that were on my mind. so, if you'd like to see a model on here, let me know. (3) i currently only do an 8x8 grid because i thought that by itself would be pretty hard for current LLMs, but of course we can increase grid sizes! (4) the coolest thing is, i dont think we can calculate the max possible number of states (yay undecidability!) you can go without repeating, so this is essentially a no-ceiling task, which is pretty cool! again, i did this mostly out of a desire to make LLMs do something fun. if this keeps me entertained for a few more days, i'd likely release a blog post on it. if it keeps me entertained for a week (and someone sponsors me), i'll put more work into it :P lastly, this is fully open sourced, so feel free to run this on your own!

introducing a new, very fun, LLM benchmark- the Game-of-Life Bench! the rules are simple: given an 8x8 grid following Conway's game of life rules, the goal is to create an initial pattern with at most 32 cells that can last the longest number of turns before dying/repeating. some results to highlight (with caveats detailed below): - gpt 5.1 lasts the longest with a 106 step run - claude models are really bad at this! they refuse to reason about this task and score < 25 points - deepseek r1 is the best open model with 102 steps. why? because i wanted to create a benchmark that has (i think) no practicality, but is still fun to look at, cheap, and still measures something interesting. i also am a big fan of the game of life. its absurdly simple rules leading to intractability is extremely cool to me. also, i saw a lot of work with LLMs trying to "predict" the next state in Conway's game of life, I think game-of-life bench is more fun because it's pretty open ended and only asks the LLM for the initial state. I also think this could be an RL env? but idk why you would ever train on this task haha i don't think this is a "serious" benchmark because it doesnt measure anything practical, but i still think it's a hard benchmark exactly because you can't predict what happens with your initial state many turns into the future; this is why i was initially expecting all LLMs to be bad at it, but turns out, some are clearly better than the others (the ordering may surprise you!) reminder: this is still a work-in-progress; (1) i am gpu-poor so could only do 10 runs for each model, even though total running cost is relatively low. maybe with some more credits i can run more seeds for each model. (2) i handpicked models which i think are at the frontier right now, plus some others that were on my mind. so, if you'd like to see a model on here, let me know. (3) i currently only do an 8x8 grid because i thought that by itself would be pretty hard for current LLMs, but of course we can increase grid sizes! (4) the coolest thing is, i dont think we can calculate the max possible number of states (yay undecidability!) you can go without repeating, so this is essentially a no-ceiling task, which is pretty cool! again, i did this mostly out of a desire to make LLMs do something fun. if this keeps me entertained for a few more days, i'd likely release a blog post on it. if it keeps me entertained for a week (and someone sponsors me), i'll put more work into it :P lastly, this is fully open sourced, so feel free to run this on your own!

Akshit

13,775 views • 5 months ago