正在加载视频...

视频加载失败

返回首页

Start to explore 3rd person shooting game simulation by using Minimax H3. Here is what I have found: 01. A lot of fun. 02. Comparing to other models, H3 has a better understanding of the game mechanic. - Great 03. Be careful with prompt "QTE". Many models would treat...

Jinghui_Dong's profile picture

Jim HuiHui

1,617 subscribers

13,150 次观看 • 12 天前 •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Mia Khalifa & Michael Scott A job interview from The Office, but with… Sorry, guys—I just had to do this 🤣🤣🤣 I created this entire video on MiniMax Design (H3) using MiniMax H3—with just a single prompt. This model is absolutely insane. I’ve been running a lot of tests, and I honestly haven’t had this much fun with a model in a very long time.
0:14

Mia Khalifa & Michael Scott A job interview from The Office, but with… Sorry, guys—I just had to do this 🤣🤣🤣 I created this entire video on MiniMax Design (H3) using MiniMax H3—with just a single prompt. This model is absolutely insane. I’ve been running a lot of tests, and I honestly haven’t had this much fun with a model in a very long time.

Riccardo Wolf

21,218 次观看 • 19 天前

MiniMax H3 is a serious upgrade for AI video creation. Instead of relying on just text prompts, you can combine images, videos, and audio references to control the character, motion, camera, and sound. What stands out: → Up to 9 image + 3 video + 3 audio references → Native synced voice, music & sound effects → 2K video generation up to 15 seconds → Instruction-based video editing → Motion transfer → Better character, camera & voice control The interesting part is that you can build a full scene from references instead of endlessly regenerating until the result feels right. 🔗 Now is a great time to try MiniMax H3 on Magnific, especially with the current 2K offer.
0:12

MiniMax H3 is a serious upgrade for AI video creation. Instead of relying on just text prompts, you can combine images, videos, and audio references to control the character, motion, camera, and sound. What stands out: → Up to 9 image + 3 video + 3 audio references → Native synced voice, music & sound effects → 2K video generation up to 15 seconds → Instruction-based video editing → Motion transfer → Better character, camera & voice control The interesting part is that you can build a full scene from references instead of endlessly regenerating until the result feels right. 🔗 Now is a great time to try MiniMax H3 on Magnific, especially with the current 2K offer.

Tanvir Anjum

31,373 次观看 • 7 天前

🧪 My GEN-3 Prompting Process I get a lot of questions on how I find my prompts when using Image-2-Video in Runway. Here is a quick breakdown of the general thought process. If you have any further questions, let's chat in the comments below. 1️⃣ I always start by doing 3x generations without any prompts and additional settings. 2️⃣ I analyze those 3x generations and identify patterns. What did the model always do well, where did it fail. 3️⃣ I then use prompting and the different controllability features to eliminate where the model struggled on its own. --- General Tips --- *️⃣ There are some tokens which work universally. "Muted colors, low contrast" are great to preserve the colors of the original input image. "Static camera, natural movement" works fantastically to get cinematic shots. *️⃣ My I2V prompts are on the shorter side. It's usually a sentence describing the scene and then individual modifiers like the ones mentioned above to fix certain camera/lighting/movement artifacts. *️⃣ Start small and prompt engineer in steps. This is very much an iterative process which rewards you for understanding model behavior and knowing how to craft a visual architecture with words. --- Disclaimer --- Please note that this approach is more suitable for a professional workflow. Therefore, I recommend it for users on the unlimited plan who don't need to worry about credits.
0:10

🧪 My GEN-3 Prompting Process I get a lot of questions on how I find my prompts when using Image-2-Video in Runway. Here is a quick breakdown of the general thought process. If you have any further questions, let's chat in the comments below. 1️⃣ I always start by doing 3x generations without any prompts and additional settings. 2️⃣ I analyze those 3x generations and identify patterns. What did the model always do well, where did it fail. 3️⃣ I then use prompting and the different controllability features to eliminate where the model struggled on its own. --- General Tips --- *️⃣ There are some tokens which work universally. "Muted colors, low contrast" are great to preserve the colors of the original input image. "Static camera, natural movement" works fantastically to get cinematic shots. *️⃣ My I2V prompts are on the shorter side. It's usually a sentence describing the scene and then individual modifiers like the ones mentioned above to fix certain camera/lighting/movement artifacts. *️⃣ Start small and prompt engineer in steps. This is very much an iterative process which rewards you for understanding model behavior and knowing how to craft a visual architecture with words. --- Disclaimer --- Please note that this approach is more suitable for a professional workflow. Therefore, I recommend it for users on the unlimited plan who don't need to worry about credits.

Nicolas Neubert

46,240 次观看 • 2 年前

The more I sit with it, the more I think Michael Penix Jr.'s performance Sunday vs. Washington was special. Ceiling is so high. Final piece is consistency. This play in particular could be a sign of things to come for MPJ and Kyle Pitts. A lot of trust here & a great win by KP.
0:16

The more I sit with it, the more I think Michael Penix Jr.'s performance Sunday vs. Washington was special. Ceiling is so high. Final piece is consistency. This play in particular could be a sign of things to come for MPJ and Kyle Pitts. A lot of trust here & a great win by KP.

Will McFadden

25,077 次观看 • 10 个月前

After a few more hours, I think I've figured out Opus 5. Opus 5 is trained to be more agentic than anything I've used. All Claude 5 models are like that. So what changes? The way to interact with Opus 5 or contextualize it won't work the same way as with other models. It loves exploring, so it doesn't need much guidance for it. Unique preferences, artifacts, and references compliment it well and enable cleaner and more effective exploration and execution. Now that it can explore more effectively on its own and understand intent better, the best thing to do is to get out of its way (e.g., it doesn't need examples of your preferences; a clear high-level description of it works best). It's truly agentic in that sense. A good first step to provide better context for Opus 5 is to distinguish between what's situational and what needs persistence. Regardless, persistent system prompts and CLAUDE.MD needs to stay lightweight. Remove memories and tool descriptions from these. CLAUDE.MD is also a great place to tap into progressive disclosure by linking command/skills to it. On the situational side, agent skills and auto-memory can leverage progressive disclosure and the improved ability of the model to use its external context/knowledge. Conflicting and unnecessary instructions, which are common at this layer (mainly to ensure reliability), are going to throw off this model easily. That's the biggest change I had to make. Simple, clean, and clear prompts and skills work best. I had to clean a lot of my skills and system prompts. The way I prompt remains the same (usually clear and well-scoped). MCP tool descriptions are also more descriptive and have been deduped from the system prompt. Anthropic released a guide on the new rules for context engineering, which was helpful here. I started to test the recommendations and created a little artifact with the things that worked along the way. This might feel like a lot of work. Believe me, it has been frustrating. But I think we can expect future frontier models to become more agentic and smarter at figuring out the right context/gaps. The best thing to do is to prepare for that now. Boris Cherny mentioned that Opus 5 is their least prompt-injectable model yet. I am not sure if that was something they intentionally trained for or if it emerged based on how it was trained, which is to be extremely agentic in nature and more direct in execution.
0:15

After a few more hours, I think I've figured out Opus 5. Opus 5 is trained to be more agentic than anything I've used. All Claude 5 models are like that. So what changes? The way to interact with Opus 5 or contextualize it won't work the same way as with other models. It loves exploring, so it doesn't need much guidance for it. Unique preferences, artifacts, and references compliment it well and enable cleaner and more effective exploration and execution. Now that it can explore more effectively on its own and understand intent better, the best thing to do is to get out of its way (e.g., it doesn't need examples of your preferences; a clear high-level description of it works best). It's truly agentic in that sense. A good first step to provide better context for Opus 5 is to distinguish between what's situational and what needs persistence. Regardless, persistent system prompts and CLAUDE.MD needs to stay lightweight. Remove memories and tool descriptions from these. CLAUDE.MD is also a great place to tap into progressive disclosure by linking command/skills to it. On the situational side, agent skills and auto-memory can leverage progressive disclosure and the improved ability of the model to use its external context/knowledge. Conflicting and unnecessary instructions, which are common at this layer (mainly to ensure reliability), are going to throw off this model easily. That's the biggest change I had to make. Simple, clean, and clear prompts and skills work best. I had to clean a lot of my skills and system prompts. The way I prompt remains the same (usually clear and well-scoped). MCP tool descriptions are also more descriptive and have been deduped from the system prompt. Anthropic released a guide on the new rules for context engineering, which was helpful here. I started to test the recommendations and created a little artifact with the things that worked along the way. This might feel like a lot of work. Believe me, it has been frustrating. But I think we can expect future frontier models to become more agentic and smarter at figuring out the right context/gaps. The best thing to do is to prepare for that now. Boris Cherny mentioned that Opus 5 is their least prompt-injectable model yet. I am not sure if that was something they intentionally trained for or if it emerged based on how it was trained, which is to be extremely agentic in nature and more direct in execution.

elvis

37,647 次观看 • 24 天前

From product image to video with just one tool - Dzine As you may have noticed, this is one of my favorite tools. It is also very underrated, as probably 50% of my tutorials include some workflow. I was testing the new image-to-video option today, and I love it. Step - by step guide in comments 🔽 I can do 95% of a workflow without switching between apps. Image generation, Image to image with style reference, background removal, background generation, and 2 frames image to video. The only other app I have been using for this video is CapCut so that I can stitch it together. Step by step 🔽
0:11

From product image to video with just one tool - Dzine As you may have noticed, this is one of my favorite tools. It is also very underrated, as probably 50% of my tutorials include some workflow. I was testing the new image-to-video option today, and I love it. Step - by step guide in comments 🔽 I can do 95% of a workflow without switching between apps. Image generation, Image to image with style reference, background removal, background generation, and 2 frames image to video. The only other app I have been using for this video is CapCut so that I can stitch it together. Step by step 🔽

Teodora P L

28,523 次观看 • 1 年前

🎥 Comparing AI video models: Image to video • Gen-3 • Kling AI 1.5 • Hailuo MiniMax • Luma Dream Machine I used a Midjourney image in each model 4 times with no text prompt. This type of image is difficult for the AI to separate the subject in the front from the people behind her - they tend to move the group as if they are a single unit. But the results were interesting, as you can see! I chose my favorite results, below.
0:10

🎥 Comparing AI video models: Image to video • Gen-3 • Kling AI 1.5 • Hailuo MiniMax • Luma Dream Machine I used a Midjourney image in each model 4 times with no text prompt. This type of image is difficult for the AI to separate the subject in the front from the people behind her - they tend to move the group as if they are a single unit. But the results were interesting, as you can see! I chose my favorite results, below.

Heather Cooper

41,354 次观看 • 1 年前

Realizing people might not have realized that by "flower game" I mean this end screen is a DIRECT reference to the book of unveiling in Destiny 2, which itself refers to a modified version of John Horton Conway's Game of Life; the book refers to this as The Flower Game, a zero-player game where the only move is the initial set up of the pieces and their behaviors. I think if this Durandal is the one from the end of Marathon Infinity, then perhaps he's found a way to influence the Flower Game, and is using that influence to re-write the universe to create a reality that allows his escape from reality itself.
0:12

Realizing people might not have realized that by "flower game" I mean this end screen is a DIRECT reference to the book of unveiling in Destiny 2, which itself refers to a modified version of John Horton Conway's Game of Life; the book refers to this as The Flower Game, a zero-player game where the only move is the initial set up of the pieces and their behaviors. I think if this Durandal is the one from the end of Marathon Infinity, then perhaps he's found a way to influence the Flower Game, and is using that influence to re-write the universe to create a reality that allows his escape from reality itself.

John Titor, Timewalker.

49,083 次观看 • 5 个月前

Chas McCormick tried to hit the ball out of Ty France's glove after a great play by JP Crawford. The following game, he got drilled in the back by a 99mph heater. In my opinion, as long as it’s not at someone’s head, I have no problem with this. Play the game the right way and this won’t happen to you.
0:19

Chas McCormick tried to hit the ball out of Ty France's glove after a great play by JP Crawford. The following game, he got drilled in the back by a 99mph heater. In my opinion, as long as it’s not at someone’s head, I have no problem with this. Play the game the right way and this won’t happen to you.

BaseballHistoryNut

56,947 次观看 • 6 个月前

This workflow is perfect for creating short fashion-style cinematic videos. I simplified the original prompts based on willie’s method, and the whole process is now much faster and more stable: 1. Generate a 3×3 keyframe grid (Nano Banana Pro only) Use this simple prompt: “In a 3x3 grid, show this character in different angles, keep the scene the same, random poses. This is far simpler and more efficient than my old prompts. You can generate multiple times and just pick the keyframes you like most. 2. Extract a high-res keyframe (Super stable trick) Take a screenshot of the keyframe you want from the 3×3 grid, send it back to Nano Banana Pro, and simply say: “Give me a high-resolution version.” This method is much more stable than relying on complex upscaling prompts. 3. Generate the video with Kling 2.5 Turbo Upload the first and last frames to Kling 2.5 Turbo and use this prompt: “The camera very slowly and smoothly lowers on a boom.” From my testing, Kling 2.5 Turbo offers the best balance of stability and cost — other models are either less consistent or noticeably more expensive. 4. Final speed adjustment with willie’s tool (Critical step) Use the tool built by willie to fine-tune the playback speed of each clip. This step is essential for getting that premium cinematic feel. I’ll drop the tool link in the comments.
0:10

This workflow is perfect for creating short fashion-style cinematic videos. I simplified the original prompts based on willie’s method, and the whole process is now much faster and more stable: 1. Generate a 3×3 keyframe grid (Nano Banana Pro only) Use this simple prompt: “In a 3x3 grid, show this character in different angles, keep the scene the same, random poses. This is far simpler and more efficient than my old prompts. You can generate multiple times and just pick the keyframes you like most. 2. Extract a high-res keyframe (Super stable trick) Take a screenshot of the keyframe you want from the 3×3 grid, send it back to Nano Banana Pro, and simply say: “Give me a high-resolution version.” This method is much more stable than relying on complex upscaling prompts. 3. Generate the video with Kling 2.5 Turbo Upload the first and last frames to Kling 2.5 Turbo and use this prompt: “The camera very slowly and smoothly lowers on a boom.” From my testing, Kling 2.5 Turbo offers the best balance of stability and cost — other models are either less consistent or noticeably more expensive. 4. Final speed adjustment with willie’s tool (Critical step) Use the tool built by willie to fine-tune the playback speed of each clip. This step is essential for getting that premium cinematic feel. I’ll drop the tool link in the comments.

underwood

22,173 次观看 • 8 个月前

Ruins Of Tomorrow In this scene, I aimed to seamlessly blend the future and present through dynamic motion. As the woman strides across a battlefield, the future gradually transforms into a haunting remnant of the present, with her relentless forward movement symbolizing the inevitable decay. She cradles a child, her symbol of hope, rescued amid the chaos. Hailuo AI (MiniMax) managed to demonstrate a fusion of realities in a single generation. Prompt: A cinematic side view tracking shot shows a female soldier holding her baby. The person remains centered in the frame. The person is walking between two realities. The person starts on the left which shows a world in chaos with burning tanks and debris. They walk to the other side which shows a world flourishing with green grass, flowers and trees.
0:10

Ruins Of Tomorrow In this scene, I aimed to seamlessly blend the future and present through dynamic motion. As the woman strides across a battlefield, the future gradually transforms into a haunting remnant of the present, with her relentless forward movement symbolizing the inevitable decay. She cradles a child, her symbol of hope, rescued amid the chaos. Hailuo AI (MiniMax) managed to demonstrate a fusion of realities in a single generation. Prompt: A cinematic side view tracking shot shows a female soldier holding her baby. The person remains centered in the frame. The person is walking between two realities. The person starts on the left which shows a world in chaos with burning tanks and debris. They walk to the other side which shows a world flourishing with green grass, flowers and trees.

Travis Davids

11,857 次观看 • 1 年前

"After me comes Terminator 4, and now here is Terminator 6!" — Yuvraj Singh crowns Vaibhav Suryavanshi as India's next hitting machine.🚨😱🗣️ "I always call myself a Terminator. Now there is Terminator 4, Abhishek Sharma, who is four times better than me. And after him comes Terminator 6, Vaibhav Suryavanshi, who has evolved even more. I did my part in my time, Abhishek elevated it, and now Vaibhav is setting new standards. This is the third phase of the same journey. It is great to see the sport evolving! When I watch Sinner and Alcaraz, I see how tennis is changing. I see the same evolution in Abhishek and Vaibhav. I have spent a lot of time training Abhishek, and I would love to spend time with Vaibhav as well. He has a great career ahead of him. He is a serious player in the making!"
0:13

"After me comes Terminator 4, and now here is Terminator 6!" — Yuvraj Singh crowns Vaibhav Suryavanshi as India's next hitting machine.🚨😱🗣️ "I always call myself a Terminator. Now there is Terminator 4, Abhishek Sharma, who is four times better than me. And after him comes Terminator 6, Vaibhav Suryavanshi, who has evolved even more. I did my part in my time, Abhishek elevated it, and now Vaibhav is setting new standards. This is the third phase of the same journey. It is great to see the sport evolving! When I watch Sinner and Alcaraz, I see how tennis is changing. I see the same evolution in Abhishek and Vaibhav. I have spent a lot of time training Abhishek, and I would love to spend time with Vaibhav as well. He has a great career ahead of him. He is a serious player in the making!"

sportified

47,838 次观看 • 1 个月前

It is a true honor to be recognized by the The New York Times, and I am deeply grateful to everyone who voted for me. To be counted among colleagues that I love and have had the privilege to work with throughout my career makes this moment even more special - I couldn’t be happier to share it with such great company. NYT Magazine #Grateful #Honored #NileRodgers #NYTMagazine #AmericanSongwriters​​​​​​​​​​​​​​​​
0:17

It is a true honor to be recognized by the The New York Times, and I am deeply grateful to everyone who voted for me. To be counted among colleagues that I love and have had the privilege to work with throughout my career makes this moment even more special - I couldn’t be happier to share it with such great company. NYT Magazine #Grateful #Honored #NileRodgers #NYTMagazine #AmericanSongwriters​​​​​​​​​​​​​​​​

Nile Rodgers

22,642 次观看 • 3 个月前

Good morning and where do I start? I’m so proud of the man you’re becoming and I’m so thankful to have you in my life. The first player I met and the first player I had dinner with after I became the receivers coach at Baylor. Yes I’m speaking it into existence but it’s also the feeling I have when I tell you God truly has a calling on your life. The person you are lets me know the lord is and will continue to bless you more than you can imagine. I love you like a son, little brother, nephew, etc…. Played receiver for a big time program like Baylor, played in the NFL, from the great state of Texas but more importantly a GREAT person and man of God. So you have all the tools you need to be successful in this coaching profession but more importantly LIFE! I’m so thankful to be a part of your journey. I’m going to do EVERYTHING in my power with the help of God to make sure you succeed. I told you this and I truly believe it. You’re the next great coach/receivers coach. Love you G and this post doesn’t do justice for how I feel about you or the kind of person you are✊🏾🙏🏾❤️💯 Gavin Holmes
0:11

Good morning and where do I start? I’m so proud of the man you’re becoming and I’m so thankful to have you in my life. The first player I met and the first player I had dinner with after I became the receivers coach at Baylor. Yes I’m speaking it into existence but it’s also the feeling I have when I tell you God truly has a calling on your life. The person you are lets me know the lord is and will continue to bless you more than you can imagine. I love you like a son, little brother, nephew, etc…. Played receiver for a big time program like Baylor, played in the NFL, from the great state of Texas but more importantly a GREAT person and man of God. So you have all the tools you need to be successful in this coaching profession but more importantly LIFE! I’m so thankful to be a part of your journey. I’m going to do EVERYTHING in my power with the help of God to make sure you succeed. I told you this and I truly believe it. You’re the next great coach/receivers coach. Love you G and this post doesn’t do justice for how I feel about you or the kind of person you are✊🏾🙏🏾❤️💯 Gavin Holmes

Dallas Baker

78,273 次观看 • 1 年前

My heart was about to explode here in Thailand. My videographers were crying and I was holding back my tears for so many times. From Malaysia, now here to Thailand, I was overwhelmed with the community that I was able to reach here through the years. The most fulfilling part as a contents creator is knowing that your contents have made a great positive impact on people from communities far from you. I can't wait to tell you more of my experience 🙏🏼❤️
0:19

My heart was about to explode here in Thailand. My videographers were crying and I was holding back my tears for so many times. From Malaysia, now here to Thailand, I was overwhelmed with the community that I was able to reach here through the years. The most fulfilling part as a contents creator is knowing that your contents have made a great positive impact on people from communities far from you. I can't wait to tell you more of my experience 🙏🏼❤️

Prince De Guzman Transformations

14,496 次观看 • 1 年前

🖤 Partner Push Announcement 🖤 I wanted to officially announce that over the next few months, I will be partner pushing on the purple app! Many of you already know this as I’ve been talking about it on the purple app already, but I wanted to finally announce it on here as well!! Implings and friends! All lurks, raids, etc will be greatly appreciated during the next few months, our numbers are so close!! I want to try to overshoot a bit so that the chances of me getting denied are smaller. 💙 Hitting partner would mean so much to me as I continue to work hard at making this my full time job. Content creation is my dream and passion, I can’t wait to see what great heights we reach together! I have been working incredibly hard to make this dream a reality, and I couldn’t have done with without you guys. My goal is to hit it by the end of the year, possibly by my birthday in November!! Thank you so much for all the love and support, I’m super excited to shoot for this next milestone with you, my implings. 🥰
0:16

🖤 Partner Push Announcement 🖤 I wanted to officially announce that over the next few months, I will be partner pushing on the purple app! Many of you already know this as I’ve been talking about it on the purple app already, but I wanted to finally announce it on here as well!! Implings and friends! All lurks, raids, etc will be greatly appreciated during the next few months, our numbers are so close!! I want to try to overshoot a bit so that the chances of me getting denied are smaller. 💙 Hitting partner would mean so much to me as I continue to work hard at making this my full time job. Content creation is my dream and passion, I can’t wait to see what great heights we reach together! I have been working incredibly hard to make this dream a reality, and I couldn’t have done with without you guys. My goal is to hit it by the end of the year, possibly by my birthday in November!! Thank you so much for all the love and support, I’m super excited to shoot for this next milestone with you, my implings. 🥰

♱ 𝕷𝖞𝖑𝖆𝖓𝖝𝖎𝖆 💀💙🥀 ♱

17,203 次观看 • 1 年前

i built an open world minigame that's controlled with hand movements only here's a step-by-step tutorial: > started with my mediapipe + threejs template (see QT) > added a 3D model made by quaternius > sent a few prompts to gemini 3... --- prompt #1: repurpose the attached script, but now I want to import a gltf model (assets/model.gltf) and use that in the scene instead of the cube when the user moves their hand, a waypoint indicator should move around the scene, and the 3D model should smoothly move there when the user makes a fist, the model should jump the 3D model contains bundled animations in it. use the "idle", "run", and "jump" animations --- prompt #2: make this a procedural open world adventure. generate very simple procedural voxel terrain when the model gets near the edges of the screen, the camera should move, allowing the model to keep going in that direction --- prompt #3: add some floating glowing gems around the map that the user can collect --- i made some manual tweaks for styling and game feel, but that's the gist of it thanks for reading if you got all the way here full code is available at my link in bio
0:15

i built an open world minigame that's controlled with hand movements only here's a step-by-step tutorial: > started with my mediapipe + threejs template (see QT) > added a 3D model made by quaternius > sent a few prompts to gemini 3... --- prompt #1: repurpose the attached script, but now I want to import a gltf model (assets/model.gltf) and use that in the scene instead of the cube when the user moves their hand, a waypoint indicator should move around the scene, and the 3D model should smoothly move there when the user makes a fist, the model should jump the 3D model contains bundled animations in it. use the "idle", "run", and "jump" animations --- prompt #2: make this a procedural open world adventure. generate very simple procedural voxel terrain when the model gets near the edges of the screen, the camera should move, allowing the model to keep going in that direction --- prompt #3: add some floating glowing gems around the map that the user can collect --- i made some manual tweaks for styling and game feel, but that's the gist of it thanks for reading if you got all the way here full code is available at my link in bio

AA

131,799 次观看 • 8 个月前

ode falling in love with his synth sound 🥲 🦊 before, i used to think, “what’s the point of this?” i used to be quite passive back then, so i would always turn the sound down bit by bit, so it was barely audible. but now that this instrument has finally started to feel like something of my own, i began thinking, “oh, this actually works with our music,” and i’ve started turning the volume up a bit more
0:18

ode falling in love with his synth sound 🥲 🦊 before, i used to think, “what’s the point of this?” i used to be quite passive back then, so i would always turn the sound down bit by bit, so it was barely audible. but now that this instrument has finally started to feel like something of my own, i began thinking, “oh, this actually works with our music,” and i’ve started turning the volume up a bit more

ten 🪐

21,409 次观看 • 4 个月前

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!
0:19

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 次观看 • 5 个月前