正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

[F2U vtuber model cutting guide]🎀 -color coded PSD with many asset options -detailed Power Point with rigging clips explaining cutting decisions this guide uses alot of underfill, masks, and invert masks hope it can help artists to upgrade their vtuber model art for wider range movement models avaliable in... show more

Kitto🦋🍂Live2D art/rig

10,975 subscribers

12,917 次观看 • 5 天前 •via X (Twitter)

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

After 5 months of planning, our first Live2D Jakarta Chapter event was a huge success! We brought an interactive Live2D showcase to the biggest anime con in Indonesia, featuring over 20 finest models created by our local artists and riggers. This showcase also included a custom-made plugin for VTube Studio that adds touch interactivity to Live2D models, allowing visitors to interact with VTuber models in a more engaging way! We hope this demo could open up new potential for Vtubers in IRL events. Special thanks to Live2D Inc. for trusting me to host this event! The showcase attracted many curious visitors and gave us the opportunity to introduce this tech to a wider audience, and we hope we are opening up new opportunities for everyone. Look forward to our next one! (Model: Key Oriesa by karamomo🍑 | Live2D Rigging | Food Illustration)

After 5 months of planning, our first Live2D Jakarta Chapter event was a huge success! We brought an interactive Live2D showcase to the biggest anime con in Indonesia, featuring over 20 finest models created by our local artists and riggers. This showcase also included a custom-made plugin for VTube Studio that adds touch interactivity to Live2D models, allowing visitors to interact with VTuber models in a more engaging way! We hope this demo could open up new potential for Vtubers in IRL events. Special thanks to Live2D Inc. for trusting me to host this event! The showcase attracted many curious visitors and gave us the opportunity to introduce this tech to a wider audience, and we hope we are opening up new opportunities for everyone. Look forward to our next one! (Model: Key Oriesa by karamomo🍑 | Live2D Rigging | Food Illustration)

Ran 🐲 Live2D Animator

165,862 次观看 • 4 个月前

𝗘𝗺𝗲𝗿𝗴𝗲𝗻𝗰𝘆 𝗰𝗼𝗺𝗺𝗶𝘀𝘀𝗶𝗼𝗻 𝗼𝗽𝗲𝗻! I need to get my rent money in order after two commissioners canceled their orders. 💀 I have 24h time to get my rent money together. 😭 Live2D model for only 500 € I use paypal or kofi I have 3 slots limit for this and it's only for the next 24h: -Model is extra squishy -The Vtuber model comes with all the face and wiggle expressions you might need. -Rig and art -A toggle of choice for the expression. Any help is appreciated 🤍🤍🤍 Dm me if interested or any questions might come up.

𝗘𝗺𝗲𝗿𝗴𝗲𝗻𝗰𝘆 𝗰𝗼𝗺𝗺𝗶𝘀𝘀𝗶𝗼𝗻 𝗼𝗽𝗲𝗻! I need to get my rent money in order after two commissioners canceled their orders. 💀 I have 24h time to get my rent money together. 😭 Live2D model for only 500 € I use paypal or kofi I have 3 slots limit for this and it's only for the next 24h: -Model is extra squishy -The Vtuber model comes with all the face and wiggle expressions you might need. -Rig and art -A toggle of choice for the expression. Any help is appreciated 🤍🤍🤍 Dm me if interested or any questions might come up.

Rebecca Doodles

18,515 次观看 • 1 年前

➡️It's a little embarrassing to share, but I feel like it took me way to long to realize the better workflow with rigging. So here's a little #live2d tip for those beginner artists that clicked for me a while back! I used to rig all body, clothing, and accessories for one given area all at once, but I think it ended up being inconsistent to other parts of the model. My workflow was slower too. 😬 It's similar to the inconsistencies you'd get by not drawing the bone structures of your character drawing, and just trying to draw the clothes right away. Body parts can sometimes seem to not line up, especially as a newer artist. It's a small change, but rigging only the nude base in full and then slowly adding one clothing part at a time makes the entire rigging process easier.😄 I know it seems really obvious, but I actually didn't start doing this until early last year. Everything comes together much faster and much more consistently, so definitely give it a try! Hope you enjoy this #vtuber leg rigging ☺️

➡️It's a little embarrassing to share, but I feel like it took me way to long to realize the better workflow with rigging. So here's a little #live2d tip for those beginner artists that clicked for me a while back! I used to rig all body, clothing, and accessories for one given area all at once, but I think it ended up being inconsistent to other parts of the model. My workflow was slower too. 😬 It's similar to the inconsistencies you'd get by not drawing the bone structures of your character drawing, and just trying to draw the clothes right away. Body parts can sometimes seem to not line up, especially as a newer artist. It's a small change, but rigging only the nude base in full and then slowly adding one clothing part at a time makes the entire rigging process easier.😄 I know it seems really obvious, but I actually didn't start doing this until early last year. Everything comes together much faster and much more consistently, so definitely give it a try! Hope you enjoy this #vtuber leg rigging ☺️

Yoona 主人公 | Live2D | COMMS OPEN

25,317 次观看 • 9 个月前

Fine-tune DeepSeek-OCR on your own language! (100% local) DeepSeek-OCR is a 3B-parameter vision model that achieves 97% precision while using 10× fewer vision tokens than text-based LLMs. It handles tables, papers, and handwriting without killing your GPU or budget. Why it matters: Most vision models treat documents as massive sequences of tokens, making long-context processing expensive and slow. DeepSeek-OCR uses context optical compression to convert 2D layouts into vision tokens, enabling efficient processing of complex documents. The best part? You can easily fine-tune it for your specific use case on a single GPU. I used Unsloth to run this experiment on Persian text and saw an 88.26% improvement in character error rate. ↳ Base model: 149% character error rate (CER) ↳ Fine-tuned model: 60% CER (57% more accurate) ↳ Training time: 60 steps on a single GPU Persian was just the test case. You can swap in your own dataset for any language, document type, or specific domain you're working with. I've shared the complete guide in the next tweet - all the code, notebooks, and environment setup ready to run with a single click. Everything is 100% open-source!

Fine-tune DeepSeek-OCR on your own language! (100% local) DeepSeek-OCR is a 3B-parameter vision model that achieves 97% precision while using 10× fewer vision tokens than text-based LLMs. It handles tables, papers, and handwriting without killing your GPU or budget. Why it matters: Most vision models treat documents as massive sequences of tokens, making long-context processing expensive and slow. DeepSeek-OCR uses context optical compression to convert 2D layouts into vision tokens, enabling efficient processing of complex documents. The best part? You can easily fine-tune it for your specific use case on a single GPU. I used Unsloth to run this experiment on Persian text and saw an 88.26% improvement in character error rate. ↳ Base model: 149% character error rate (CER) ↳ Fine-tuned model: 60% CER (57% more accurate) ↳ Training time: 60 steps on a single GPU Persian was just the test case. You can swap in your own dataset for any language, document type, or specific domain you're working with. I've shared the complete guide in the next tweet - all the code, notebooks, and environment setup ready to run with a single click. Everything is 100% open-source!

Akshay 🚀

126,273 次观看 • 11 个月前

3 years ago Refik Anadol & Efsun, together with the royal family of the forest, the Yawanawa, dreamed of a project that could share the crossover of traditional indigenous art with futuristic pigments of AI. They dreamed of building a community to not only host their family, but also bring together their indigenous brothers and sisters from around the world. Aldeia Sagrada was reimagined. We 1OF1 are very honoured & deeply grateful that Refik Anadol & the Yawanawa family have chosen us to be the steward of their historic collaborative art work, the Winds of Yawanawa 1of1. Last week my family & I had the privilege to join Indigenous Nations from across the Amazon basin & around the world as they came together for a truly historic & deeply beautiful moment at the Indigenous Ayahuasca Conference in Acre, Brasil, where Indigenous leaders shared their ancient wisdom, beautiful art & wonderful energy with each other. As someone who has built infra in Brazil, I can confirm that what Refik Anadol and the Yawanawa family have built in Aldeia Sagrada is absolutely breathtaking & indescribable. The infrastructure & organization of their wonderful community is incredible. They have proven to the world a model for synergistic alliances between traditional Indigenous nations with new technologists, creatives and builders that can both massively support these native communities as well as celebrate the preservation of their way of life. I found myself overwhelmed with emotion at points throughout the week witnessing how amazing this place and these people are. It is with all of the lessons of 1OF1’s global tours in hand & our hearts filled with a deep sense of mission, that we commit to share this iconic work of art with the world in conversation with other works of Amazonian indigenous art. We promise to share the inspiring message of the Yawanawa across the globe through the power of art.

3 years ago Refik Anadol & Efsun, together with the royal family of the forest, the Yawanawa, dreamed of a project that could share the crossover of traditional indigenous art with futuristic pigments of AI. They dreamed of building a community to not only host their family, but also bring together their indigenous brothers and sisters from around the world. Aldeia Sagrada was reimagined. We 1OF1 are very honoured & deeply grateful that Refik Anadol & the Yawanawa family have chosen us to be the steward of their historic collaborative art work, the Winds of Yawanawa 1of1. Last week my family & I had the privilege to join Indigenous Nations from across the Amazon basin & around the world as they came together for a truly historic & deeply beautiful moment at the Indigenous Ayahuasca Conference in Acre, Brasil, where Indigenous leaders shared their ancient wisdom, beautiful art & wonderful energy with each other. As someone who has built infra in Brazil, I can confirm that what Refik Anadol and the Yawanawa family have built in Aldeia Sagrada is absolutely breathtaking & indescribable. The infrastructure & organization of their wonderful community is incredible. They have proven to the world a model for synergistic alliances between traditional Indigenous nations with new technologists, creatives and builders that can both massively support these native communities as well as celebrate the preservation of their way of life. I found myself overwhelmed with emotion at points throughout the week witnessing how amazing this place and these people are. It is with all of the lessons of 1OF1’s global tours in hand & our hearts filled with a deep sense of mission, that we commit to share this iconic work of art with the world in conversation with other works of Amazonian indigenous art. We promise to share the inspiring message of the Yawanawa across the globe through the power of art.

Ryan Zurrer

12,145 次观看 • 1 年前

I’m insanely proud of this piece and the collection. This is edition #2 of Left Alone and whomever collects the digital version of it, can claim this if they want it. Description ⬇️: Left Alone by Archie Shepp (1978) is one of the most beautiful pieces of instrumental music I know. A saxophone, accompanied by a piano, plays notes that resonate with my heart. In my mind, this photograph moves with an impossible slowness to an undefinable silence between notes in jazz riff. The duality in this Aria, Left Alone, is nihilism and conflagration. I posed the model and the fabric in the Japanese character for fire. She models the pose of power in the fluidity of dance. A flame, alone, is a powerful and fragile life force that is on the brink of greatness, and the precipice of darkness; life extinguished. The red is her power extending from heart center and the white is the searing flame. The art is laden with abstractions. Broken pieces of the self that seem grotesque but to the observer, are beautiful. To illustrate how these glitches are manifest in ourselves, I’ve placed each on a Fibonacci curve that spirals from the heart, representing our repeating thought patterns. In Left Alone, I chose to hide the Fibonacci curve. In the latter Arias, I chose to reveal them. You can find the Aria collection on Foundation 🌐 This is Left Alone, the second of nine pieces.

I’m insanely proud of this piece and the collection. This is edition #2 of Left Alone and whomever collects the digital version of it, can claim this if they want it. Description ⬇️: Left Alone by Archie Shepp (1978) is one of the most beautiful pieces of instrumental music I know. A saxophone, accompanied by a piano, plays notes that resonate with my heart. In my mind, this photograph moves with an impossible slowness to an undefinable silence between notes in jazz riff. The duality in this Aria, Left Alone, is nihilism and conflagration. I posed the model and the fabric in the Japanese character for fire. She models the pose of power in the fluidity of dance. A flame, alone, is a powerful and fragile life force that is on the brink of greatness, and the precipice of darkness; life extinguished. The red is her power extending from heart center and the white is the searing flame. The art is laden with abstractions. Broken pieces of the self that seem grotesque but to the observer, are beautiful. To illustrate how these glitches are manifest in ourselves, I’ve placed each on a Fibonacci curve that spirals from the heart, representing our repeating thought patterns. In Left Alone, I chose to hide the Fibonacci curve. In the latter Arias, I chose to reveal them. You can find the Aria collection on Foundation 🌐 This is Left Alone, the second of nine pieces.

Jason Matias | 🐍

27,349 次观看 • 2 年前

One of the craziest use cases I’ve found for Jev: verifiers. I am so excited about this that I at least wanted to share the high-level idea. I used Jev to build a custom verifier for the /goal feature in my agent harness. It checks whether the goal is actually complete after every turn, making continuous verification cheap enough to scale. This means I can run more of these verifiers (previously handled by another expensive reasoning model) more frequently to keep the agents on track. System One models are perfect for verification. I think of this as scaling harnesses further by cleverly combining System One and System Two models. I have a feeling this will enable a new wave of scalable test-time compute methods. Watch this space closely. I've just started to experiment with this and am already seeing really good results. I need to explore and figure out a way to benchmark it. I will share more once I have more results. This is an insane unlock for long-horizon agents. You heard it here first. And you can expect to see more harnesses embracing this new pattern. Full guide dropping in the next couple of days.

One of the craziest use cases I’ve found for Jev: verifiers. I am so excited about this that I at least wanted to share the high-level idea. I used Jev to build a custom verifier for the /goal feature in my agent harness. It checks whether the goal is actually complete after every turn, making continuous verification cheap enough to scale. This means I can run more of these verifiers (previously handled by another expensive reasoning model) more frequently to keep the agents on track. System One models are perfect for verification. I think of this as scaling harnesses further by cleverly combining System One and System Two models. I have a feeling this will enable a new wave of scalable test-time compute methods. Watch this space closely. I've just started to experiment with this and am already seeing really good results. I need to explore and figure out a way to benchmark it. I will share more once I have more results. This is an insane unlock for long-horizon agents. You heard it here first. And you can expect to see more harnesses embracing this new pattern. Full guide dropping in the next couple of days.

elvis

59,794 次观看 • 15 天前

I can't believe how good this turned out. Recently, I put out my first physics paper proposing that life and the universe might be the result of a law governing how oscillators sync up and couple together. I find it beautiful, awe-inspiring, and deeply meaningful. It aligns with how I have always seen the world. It suggests that there is an arrow of negative entropy, a law governing how life forms in all of its complexity. Learning about it is intimidating. I am self-taught, and have struggled for months to understand and articulate the core concepts and framework. After I put out the paper, I was challenged by Danielle Fong 🔆 to make an 'explorable explanation'. A guide you can walk through, and which will hold your hand through each concept, equation, and principle. It was amazing to make. In fact, I plan to make many more covering the wide range of topics I have been researching and exploring. Both to help me, and because I believe that this is something the world needs right now. Link in comments

I can't believe how good this turned out. Recently, I put out my first physics paper proposing that life and the universe might be the result of a law governing how oscillators sync up and couple together. I find it beautiful, awe-inspiring, and deeply meaningful. It aligns with how I have always seen the world. It suggests that there is an arrow of negative entropy, a law governing how life forms in all of its complexity. Learning about it is intimidating. I am self-taught, and have struggled for months to understand and articulate the core concepts and framework. After I put out the paper, I was challenged by Danielle Fong 🔆 to make an 'explorable explanation'. A guide you can walk through, and which will hold your hand through each concept, equation, and principle. It was amazing to make. In fact, I plan to make many more covering the wide range of topics I have been researching and exploring. Both to help me, and because I believe that this is something the world needs right now. Link in comments

Parzival - ∞/89

23,155 次观看 • 5 个月前

🎉 Exciting News! We are thrilled to announce the grand opening party of the first Sense4FIT GYM! 🏋️‍♂️🎉 🌐 Web3 Benefits: •Exclusive Prices for Legendary NFT Holders. •Free Access with Special Ranks NFTs: Ultra Challenger rank💥 💳 Convenient Payment Options: •Pay with SFIT: Embracing the power of blockchain, you can now pay for your membership using SFIT tokens. 📱 Web2 User transition: •Welcome to Day 1: 💥Web2 users can download the Sense4FIT app and embark on an educational journey to discover the benefits of blockchain technology Multiversᕽ . •xPortal Installation: Our coaches will guide you through the process of installing xPortal . •Legendary NFT Acquisition: We help members to buy a legendary NFT to kickstart the Challenge process through the Sense4FIT app. 💼 B2B Partnerships: •Web2 and Web3 Integrated Ecosystem: Sense4FIT GYM welcomes B2B partnerships from both the traditional Web2 world and the exciting Web3 space, ensuring a diverse and innovative ecosystem. 🏢 Franchise Model: •Empowerment and Ownership: We offer a unique franchise model, giving individuals the opportunity to own their own Sense4FIT GYM. 🌌 Fitnessverse Platform: •Coming Soon: Keep an eye out for our groundbreaking Fitnessverse platform, set to launch within the next two months. 🔜This revolutionary platform will enable us to scale and expand our business, reaching new cities and countries, fueling growth like never before! 💥Join us on this incredible journey as we revolutionize the fitness industry, leveraging the power of blockchain and Web3. 🏋🏻‍♂️Together, we'll take this adventure city by city, country by country, and achieve new heights in the world of fitness! 🚀💪 #Sense4FIT #MultiversX

🎉 Exciting News! We are thrilled to announce the grand opening party of the first Sense4FIT GYM! 🏋️‍♂️🎉 🌐 Web3 Benefits: •Exclusive Prices for Legendary NFT Holders. •Free Access with Special Ranks NFTs: Ultra Challenger rank💥 💳 Convenient Payment Options: •Pay with SFIT: Embracing the power of blockchain, you can now pay for your membership using SFIT tokens. 📱 Web2 User transition: •Welcome to Day 1: 💥Web2 users can download the Sense4FIT app and embark on an educational journey to discover the benefits of blockchain technology Multiversᕽ . •xPortal Installation: Our coaches will guide you through the process of installing xPortal . •Legendary NFT Acquisition: We help members to buy a legendary NFT to kickstart the Challenge process through the Sense4FIT app. 💼 B2B Partnerships: •Web2 and Web3 Integrated Ecosystem: Sense4FIT GYM welcomes B2B partnerships from both the traditional Web2 world and the exciting Web3 space, ensuring a diverse and innovative ecosystem. 🏢 Franchise Model: •Empowerment and Ownership: We offer a unique franchise model, giving individuals the opportunity to own their own Sense4FIT GYM. 🌌 Fitnessverse Platform: •Coming Soon: Keep an eye out for our groundbreaking Fitnessverse platform, set to launch within the next two months. 🔜This revolutionary platform will enable us to scale and expand our business, reaching new cities and countries, fueling growth like never before! 💥Join us on this incredible journey as we revolutionize the fitness industry, leveraging the power of blockchain and Web3. 🏋🏻‍♂️Together, we'll take this adventure city by city, country by country, and achieve new heights in the world of fitness! 🚀💪 #Sense4FIT #MultiversX

Sense4FIT

57,160 次观看 • 3 年前

This is my "feel the AGI" moment: I used GPT-5.6 Sol to train my own autocorrect model that outperforms GPT-5.6 Sol (wtf??) I have no ML background. I have no idea what I'm doing. I just kept pushing Sol until it spat out a SOTA model. And I spent $0. The motivation: Years of talking to AI have made me terrible at typing. Rather than fix my skill issue, I decided to throw more AI at it. My idea was: instead of autocorrect that interrupts my flow, I want to type fast with mistakes and have AI clean it up after. I wanted the smallest local model possible, for speed, for battery life, for science! So I decided to train my own. Inspired by Andrej Karpathy’s autoresearch, I ran Codex /goal with this setup: pick an experiment, try it, record the results to a doc, throw it out if it fails, and plan the next experiment without repeating failures. I gave a few examples that had to pass, tight latency targets, and let it run. Sol did some amazing things. First, it scanned benchmarks and shortlisted base models: Qwen 3.5, Gemma 4, Liquid LFM 2.5. It found a dataset on HuggingFace for typed text. Then it built a simulator for fingers striking a Mac keyboard, modeling the physical layout with a Gaussian distribution around each key. It simulated striking the wrong key, wrong order, fat-fingering, etc. With the models + data + simulator, it fine-tuned using MLX right on my MacBook. It had a working prototype within an hour! But accuracy was pretty poor. — Problem 1: Tokenization Sol read papers, ran tests, and identified that the tokenizer was the bottleneck. Tokenization makes typos hard for the model to see, so it memorizes mappings instead of using its language priors. Sol tried ByT5, Google’s tokenizer-free byte-level LLM. This made a big improvement, but the model is old and lacked the knowledge needed to reach Sol performance. Sol dug deeper and realized a tokenizer-free model isn’t needed; instead, it used T5Gemma, an encoder-decoder model. This can understand the input deeply before producing output, and furthermore, Sol could post-train the encoder to improve performance. This gave a much higher ceiling. — Problem 2: Loss function Now the model was correcting some typos perfectly, but ignoring most. Sol realized that standard cross-entropy loss was teaching the model to avoid edits, because the vast majority of characters in the training data were left unmodified. The fix was wild: Sol wrote a custom loss function that byte-aligns the source and target strings, uses a dynamic programming algorithm to compute the minimum edits between the two, then weights correct edits much higher than copies. After a lot of tuning, this dramatically improved accuracy. — Problem 3: Autoregression One failure mode remained: if the model made a mistake, it couldn’t backtrack. It could only predict the next token. Teaching it to “think” like a reasoning model would solve this, but would be far too slow. Sol found a beautiful solution: instead of greedily predicting the next token, beam search over all possibilities. This parallelizes the exploration instead of one linear chain-of-thought. At the end, choose the path with highest cumulative log probability. This worked great, but made the experience worse, since the user wouldn’t see progress until the whole search was done. To fix this, Sol made a clever observation: after each search step, the longest common prefix among surviving branches is guaranteed to appear in the final result, so it can be displayed immediately. As the search progresses, weaker paths are dropped and the prefix grows, so the user sees continuous progress. Sol built all this as a custom MLX pipeline that does the parallel decoding on the MacBook GPU, with just ~40ms TTFT. It’s crazy fast and entirely local. — Final eval (error reduction rate, higher is better): - Apple autocorrect: 49.66% - GPT-5.6 Luna: 82.47% - GPT-5.6 Terra: 87.64% - GPT-5.6 Sol: 90.56% - Our model (1.7B): 91.02% Final cost: - 1 quota reset (thanks Tibo) - $0 (And yes, I verified there's no cheating. In fact, we test words scrubbed from the training data to prove the model isn’t memorizing) There were a ton more details and tangents I could write about: contrastive learning, GRPO, DPO, dynamic masking, and more. Sol is a fascinating and creative model. It blew my mind so many times. Don’t let a lack of experience stop you: Sol makes AI experiments accessible to anyone!

This is my "feel the AGI" moment: I used GPT-5.6 Sol to train my own autocorrect model that outperforms GPT-5.6 Sol (wtf??) I have no ML background. I have no idea what I'm doing. I just kept pushing Sol until it spat out a SOTA model. And I spent $0. The motivation: Years of talking to AI have made me terrible at typing. Rather than fix my skill issue, I decided to throw more AI at it. My idea was: instead of autocorrect that interrupts my flow, I want to type fast with mistakes and have AI clean it up after. I wanted the smallest local model possible, for speed, for battery life, for science! So I decided to train my own. Inspired by Andrej Karpathy’s autoresearch, I ran Codex /goal with this setup: pick an experiment, try it, record the results to a doc, throw it out if it fails, and plan the next experiment without repeating failures. I gave a few examples that had to pass, tight latency targets, and let it run. Sol did some amazing things. First, it scanned benchmarks and shortlisted base models: Qwen 3.5, Gemma 4, Liquid LFM 2.5. It found a dataset on HuggingFace for typed text. Then it built a simulator for fingers striking a Mac keyboard, modeling the physical layout with a Gaussian distribution around each key. It simulated striking the wrong key, wrong order, fat-fingering, etc. With the models + data + simulator, it fine-tuned using MLX right on my MacBook. It had a working prototype within an hour! But accuracy was pretty poor. — Problem 1: Tokenization Sol read papers, ran tests, and identified that the tokenizer was the bottleneck. Tokenization makes typos hard for the model to see, so it memorizes mappings instead of using its language priors. Sol tried ByT5, Google’s tokenizer-free byte-level LLM. This made a big improvement, but the model is old and lacked the knowledge needed to reach Sol performance. Sol dug deeper and realized a tokenizer-free model isn’t needed; instead, it used T5Gemma, an encoder-decoder model. This can understand the input deeply before producing output, and furthermore, Sol could post-train the encoder to improve performance. This gave a much higher ceiling. — Problem 2: Loss function Now the model was correcting some typos perfectly, but ignoring most. Sol realized that standard cross-entropy loss was teaching the model to avoid edits, because the vast majority of characters in the training data were left unmodified. The fix was wild: Sol wrote a custom loss function that byte-aligns the source and target strings, uses a dynamic programming algorithm to compute the minimum edits between the two, then weights correct edits much higher than copies. After a lot of tuning, this dramatically improved accuracy. — Problem 3: Autoregression One failure mode remained: if the model made a mistake, it couldn’t backtrack. It could only predict the next token. Teaching it to “think” like a reasoning model would solve this, but would be far too slow. Sol found a beautiful solution: instead of greedily predicting the next token, beam search over all possibilities. This parallelizes the exploration instead of one linear chain-of-thought. At the end, choose the path with highest cumulative log probability. This worked great, but made the experience worse, since the user wouldn’t see progress until the whole search was done. To fix this, Sol made a clever observation: after each search step, the longest common prefix among surviving branches is guaranteed to appear in the final result, so it can be displayed immediately. As the search progresses, weaker paths are dropped and the prefix grows, so the user sees continuous progress. Sol built all this as a custom MLX pipeline that does the parallel decoding on the MacBook GPU, with just ~40ms TTFT. It’s crazy fast and entirely local. — Final eval (error reduction rate, higher is better): - Apple autocorrect: 49.66% - GPT-5.6 Luna: 82.47% - GPT-5.6 Terra: 87.64% - GPT-5.6 Sol: 90.56% - Our model (1.7B): 91.02% Final cost: - 1 quota reset (thanks Tibo) - $0 (And yes, I verified there's no cheating. In fact, we test words scrubbed from the training data to prove the model isn’t memorizing) There were a ton more details and tangents I could write about: contrastive learning, GRPO, DPO, dynamic masking, and more. Sol is a fascinating and creative model. It blew my mind so many times. Don’t let a lack of experience stop you: Sol makes AI experiments accessible to anyone!

Anshu

179,451 次观看 • 2 个月前

I'm building my game with GPT-6, and color correction has become one of my favorite uses for it. When I generate assets separately, I can like each one on its own and still end up with a scene where the colors don't belong together. A castle looks fine in isolation, then turns greenish against the terrain. Its swords and shields disappear into the background. Getting those things to match used to mean a lot of repainting and trying different textures. Now I've built a workflow where GPT-6 takes screenshots in Unity, inspects the object's color textures, and helps bring them closer to the look I want. I can give it the terrain or tree textures as color references. It can prepare masks, send the relevant textures through Image Edit, then put the result back into the game for another look. Editors such as GPT Image 2 Edit or Nano Banana Edit can be part of that workflow. This is also why I keep pushing for properly prepared 3D models. If the parts are separated logically, you get much more control later. We could work on the swords and shields without changing the whole castle. The stone could stay dark while the equipment became easier to read. The latest pass on my game covered the overall post effects and the skeleton castles. We matched the castles to the surrounding terrain, adjusted the equipment colors, and added a subtle moving highlight. The first stronger color pass went too far. I asked for the accents to move about 30% closer to gray, and that got them where I wanted: visible, without pulling all the attention. I'm so happy with the difference. The video shows the before and after across the map and on the castles. Being able to give feedback on the actual scene and keep refining it this way feels great.

I'm building my game with GPT-6, and color correction has become one of my favorite uses for it. When I generate assets separately, I can like each one on its own and still end up with a scene where the colors don't belong together. A castle looks fine in isolation, then turns greenish against the terrain. Its swords and shields disappear into the background. Getting those things to match used to mean a lot of repainting and trying different textures. Now I've built a workflow where GPT-6 takes screenshots in Unity, inspects the object's color textures, and helps bring them closer to the look I want. I can give it the terrain or tree textures as color references. It can prepare masks, send the relevant textures through Image Edit, then put the result back into the game for another look. Editors such as GPT Image 2 Edit or Nano Banana Edit can be part of that workflow. This is also why I keep pushing for properly prepared 3D models. If the parts are separated logically, you get much more control later. We could work on the swords and shields without changing the whole castle. The stone could stay dark while the equipment became easier to read. The latest pass on my game covered the overall post effects and the skeleton castles. We matched the castles to the surrounding terrain, adjusted the equipment colors, and added a subtle moving highlight. The first stronger color pass went too far. I asked for the accents to move about 30% closer to gray, and that got them where I wanted: visible, without pulling all the attention. I'm so happy with the difference. The video shows the before and after across the map and on the castles. Being able to give feedback on the actual scene and keep refining it this way feels great.

Stefan 3D AI

17,290 次观看 • 20 天前

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.

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

37,824 次观看 • 2 个月前

8 free Polymarket Trading Bots on GitHub (from Beginner Friendly to Advanced Level). Each of these repos comes with a detailed step by step setup and usage guide in English. > Beginner Level - 5 min setup 1. This bot includes 120 ready to use strategies and tools for trading on prediction markets (Binance-Polymarket latency, Smart Routing, Penny Clipper, Momentum, DCA bots, Expiry Fade and more). It was built by a Cambridge computer science student who won a hackathon with this bot. GitHub: 2. A trading bot with a Smart Money strategy - it finds top traders in selected markets, filters them by Pnl, win rate, stable performance and then creates a list for automated copy trading. GitHub: 3. This is a bot toolkit that includes Polymarket - Kalshi arbitrage, whale alerts, market making, spread farming, sports trading and more. GitHub: 4. A weather trading bot from Chinese dev that analyzes different sources in real time, like forecasts, airport data and aviation observations (METAR + SPECI) to get the latest temperature data and generate a detailed weather report for a specific city and day. GitHub: 5. A huge collection of 30+ free trading bots and services for prediction markets. GitHub: > Advanced bot setup 1. This bot analyzes the real trading behavior of any Polymarket trader. It finds repeated patterns in his trades, shows which strategies he uses and helps you understand how to adapt them to your own trading. GitHub: 2. A bot that automatically manages all your limit orders on Polymarket to maximize liquidity rewards. GitHub: > A full ML weather model 1. A machine learning weather model that learns from weather forecasting errors. Instead of blindly trusting forecasts, it analyzes how different weather sources have historically overestimated or underestimated temperature values in specific cities and conditions. Then it automatically adjusts new forecasts to produce more accurate predictions. GitHub: All of these bots also support Dry Run mode, so you can test them on real markets without risking any funds.

8 free Polymarket Trading Bots on GitHub (from Beginner Friendly to Advanced Level). Each of these repos comes with a detailed step by step setup and usage guide in English. > Beginner Level - 5 min setup 1. This bot includes 120 ready to use strategies and tools for trading on prediction markets (Binance-Polymarket latency, Smart Routing, Penny Clipper, Momentum, DCA bots, Expiry Fade and more). It was built by a Cambridge computer science student who won a hackathon with this bot. GitHub: 2. A trading bot with a Smart Money strategy - it finds top traders in selected markets, filters them by Pnl, win rate, stable performance and then creates a list for automated copy trading. GitHub: 3. This is a bot toolkit that includes Polymarket - Kalshi arbitrage, whale alerts, market making, spread farming, sports trading and more. GitHub: 4. A weather trading bot from Chinese dev that analyzes different sources in real time, like forecasts, airport data and aviation observations (METAR + SPECI) to get the latest temperature data and generate a detailed weather report for a specific city and day. GitHub: 5. A huge collection of 30+ free trading bots and services for prediction markets. GitHub: > Advanced bot setup 1. This bot analyzes the real trading behavior of any Polymarket trader. It finds repeated patterns in his trades, shows which strategies he uses and helps you understand how to adapt them to your own trading. GitHub: 2. A bot that automatically manages all your limit orders on Polymarket to maximize liquidity rewards. GitHub: > A full ML weather model 1. A machine learning weather model that learns from weather forecasting errors. Instead of blindly trusting forecasts, it analyzes how different weather sources have historically overestimated or underestimated temperature values in specific cities and conditions. Then it automatically adjusts new forecasts to produce more accurate predictions. GitHub: All of these bots also support Dry Run mode, so you can test them on real markets without risking any funds.

Recogard

58,826 次观看 • 2 个月前

NEWS: SpaceX says a new Starlink update means that the internet will keep working even if trees or other obstacles partially block your home’s view of the sky. "As a Starlink terminal communicates with satellites overhead, it continuously builds a real-time obstruction map, allowing Starlink to dynamically understand its environment. With this information, it can proactively select the best and most stable connection. For dynamic obstructions, such as are encountered by mobile terminals, the system reactively switches in less than 1/10th of a second, allowing a connection to remain stable. The dynamic nature of Starlink combined with the many paths for the system to route traffic provide a high degree of resiliency in obstructed and changing environments. Additionally, to help customers get the most out of their service, the Starlink app includes a built-in tool to preview the install location and guide customers to the best placement with the fewest obstructions. After setup, the Starlink app provides a live obstruction map that shows exactly where signal blockages are occurring and how they may affect your experience, so that users can make informed decisions about optimizing their installation. Starlink also measures uptime 10 times a second from every terminal, transparently reporting any outage that is longer than 1/10th of a second in the app. For well-installed, even partially obstructed terminals, this uptime is typically at the 99.9% level."

NEWS: SpaceX says a new Starlink update means that the internet will keep working even if trees or other obstacles partially block your home’s view of the sky. "As a Starlink terminal communicates with satellites overhead, it continuously builds a real-time obstruction map, allowing Starlink to dynamically understand its environment. With this information, it can proactively select the best and most stable connection. For dynamic obstructions, such as are encountered by mobile terminals, the system reactively switches in less than 1/10th of a second, allowing a connection to remain stable. The dynamic nature of Starlink combined with the many paths for the system to route traffic provide a high degree of resiliency in obstructed and changing environments. Additionally, to help customers get the most out of their service, the Starlink app includes a built-in tool to preview the install location and guide customers to the best placement with the fewest obstructions. After setup, the Starlink app provides a live obstruction map that shows exactly where signal blockages are occurring and how they may affect your experience, so that users can make informed decisions about optimizing their installation. Starlink also measures uptime 10 times a second from every terminal, transparently reporting any outage that is longer than 1/10th of a second in the app. For well-installed, even partially obstructed terminals, this uptime is typically at the 99.9% level."

Sawyer Merritt

1,513,568 次观看 • 1 年前

I’m an empowerment coach, a sensual muse, and an advocate for healthy male sexuality. I’ve developed a one-of-a-kind program that encourages men to embrace self-pleasure, to appreciate their own virility, and to build confidence in their own skin. I do this by sharing breathtaking images, provocative videos, and sensual insights that inspire men to explore themselves without guilt. Think of me as your personal guide to self-discovery, only dressed in the most sinful lingerie. I focus on pleasure because men deserve it. Society has made them feel like self-pleasure is something shameful, when in reality, it’s natural, healthy, and even empowering. I see too many men struggling with confidence, repressing their desires, and feeling disconnected from their own sexuality. That’s where I come in. Through my content, I help erase the stigma and turn self-pleasure into something to be celebrated. The Vanesa's Male Development Program encourages men to embrace masturbation as a tool for confidence, self-exploration, and personal growth. I provide the inspiration—whether it’s through my tantalizing photos, seductive videos, or sensual encouragement—and my followers use it to fuel their own pleasure. It’s about learning to fully enjoy yourself without guilt, to embrace your body, and to feel more confident than ever. And believe me, when a man learns to fully enjoy his own pleasure, he radiates that confidence in every aspect of his life—from relationships to career and beyond. Pleasure makes power.

I’m an empowerment coach, a sensual muse, and an advocate for healthy male sexuality. I’ve developed a one-of-a-kind program that encourages men to embrace self-pleasure, to appreciate their own virility, and to build confidence in their own skin. I do this by sharing breathtaking images, provocative videos, and sensual insights that inspire men to explore themselves without guilt. Think of me as your personal guide to self-discovery, only dressed in the most sinful lingerie. I focus on pleasure because men deserve it. Society has made them feel like self-pleasure is something shameful, when in reality, it’s natural, healthy, and even empowering. I see too many men struggling with confidence, repressing their desires, and feeling disconnected from their own sexuality. That’s where I come in. Through my content, I help erase the stigma and turn self-pleasure into something to be celebrated. The Vanesa's Male Development Program encourages men to embrace masturbation as a tool for confidence, self-exploration, and personal growth. I provide the inspiration—whether it’s through my tantalizing photos, seductive videos, or sensual encouragement—and my followers use it to fuel their own pleasure. It’s about learning to fully enjoy yourself without guilt, to embrace your body, and to feel more confident than ever. And believe me, when a man learns to fully enjoy his own pleasure, he radiates that confidence in every aspect of his life—from relationships to career and beyond. Pleasure makes power.

Vanesa Dream

14,736 次观看 • 5 个月前

🌟 "SHOULD I BUY AN IPHONE FOR TRACKING?" 🌟 tl;dr at bottom I've been using a facecam and Nvidia tracking for a long time and upgrading to a used iphone 13 combined with vbridger, the difference is HUGE. Here is my take on it! Why is Facecam > iPhone? ✅️ More affordable, esp for those using android phones ✅️ More convenient. If you launch Vtubestudio, there's a setting where your webcam automatically turns on, which is great. ✅️ Can track pretty well in the dark IF you already have a good webcam for night tracking. ❌️❌️ Stiff tracking at times ❌️ Not good at tracking specific mouth movement Why iPhone > Facecam? ✅️✅️ You can make the most of your model, since movement along the X and Y axes are a lot more accurate and wider. Also tracks eyes and overall face better. ✅️ More EXPRESSIONS. If your rigging allows for it, things like cheek puff, and tongue are able to be tracked. As far as I know, I cannot do this on facecam. ❌️❌️ WAY more expensive or requires that you have an iPhone already. Needs more set-up (need phone stand right in front of you, need to hook your phone up to a charger at all times, phone could possibly overheat as well if it's old, so you might need a cooler). TL;DR For me, if you have an extra 200 to spare for WAY better tracking, I would 1000% recommend buying a used iPhone on Amazon. iPhone X is the BARE minumum, I would recommend 12/13 so that your phone does not overheat. I do not need to use a cooler for my used iPhone 13. Being able to use my rigging to its fullest makes the model feel so so so different (in a good way). Feel free to reply with any questions, I will try to answer them!

🌟 "SHOULD I BUY AN IPHONE FOR TRACKING?" 🌟 tl;dr at bottom I've been using a facecam and Nvidia tracking for a long time and upgrading to a used iphone 13 combined with vbridger, the difference is HUGE. Here is my take on it! Why is Facecam > iPhone? ✅️ More affordable, esp for those using android phones ✅️ More convenient. If you launch Vtubestudio, there's a setting where your webcam automatically turns on, which is great. ✅️ Can track pretty well in the dark IF you already have a good webcam for night tracking. ❌️❌️ Stiff tracking at times ❌️ Not good at tracking specific mouth movement Why iPhone > Facecam? ✅️✅️ You can make the most of your model, since movement along the X and Y axes are a lot more accurate and wider. Also tracks eyes and overall face better. ✅️ More EXPRESSIONS. If your rigging allows for it, things like cheek puff, and tongue are able to be tracked. As far as I know, I cannot do this on facecam. ❌️❌️ WAY more expensive or requires that you have an iPhone already. Needs more set-up (need phone stand right in front of you, need to hook your phone up to a charger at all times, phone could possibly overheat as well if it's old, so you might need a cooler). TL;DR For me, if you have an extra 200 to spare for WAY better tracking, I would 1000% recommend buying a used iPhone on Amazon. iPhone X is the BARE minumum, I would recommend 12/13 so that your phone does not overheat. I do not need to use a cooler for my used iPhone 13. Being able to use my rigging to its fullest makes the model feel so so so different (in a good way). Feel free to reply with any questions, I will try to answer them!

Minori 🎀🍰💢 || bakaneko vampire :3

309,850 次观看 • 2 年前

The 2023 Pheasant Hunting Forecast is available now, and it’s got us fired up for the season ahead! As you plan your hunts for this fall, be sure to check the forecast for a detailed state-by-state update on all things pheasants in the areas you plan to hunt. 🔗Link below. While each state has its own story to weave, twists to share and regions to look toward, the consistent theming across the core of the ring-necked pheasant range is this: » Yes, winter was harsh. In many places, pheasant hunters thought all was lost. But pheasants are tough and, when wintering habitat is there, can survive one hell of an onslaught of snow and cold. » A goodly base of the core pheasant range experienced almost ideal nesting conditions with a relatively dry and warm spring, punctuated by moisture at just the right times in early summer to help nesting habitat growth and vibrancy. Nesting success makes birds for hunting, period. » Conditions trend from generally dry to downright droughty heading into fall, and that will impact hunting, from the careful approach you might bring to the hunt, to habitat status on the ground (including some emergency mowing, haying and grazing regionally). For pheasants, and all other manner of upland wildlife, habitat is everything. Likewise for hunters. And it is where communion happens in fall, with a bird dog romping ahead, our hearts glad and faces smiling and a rooster out there somewhere, giving us the good old slip. The 2023 Pheasant Hunting Forecast is presented by Sportsman’s Guide #pheasantsforever #pheasanthunting #pheasants #upland #uplandhunting

The 2023 Pheasant Hunting Forecast is available now, and it’s got us fired up for the season ahead! As you plan your hunts for this fall, be sure to check the forecast for a detailed state-by-state update on all things pheasants in the areas you plan to hunt. 🔗Link below. While each state has its own story to weave, twists to share and regions to look toward, the consistent theming across the core of the ring-necked pheasant range is this: » Yes, winter was harsh. In many places, pheasant hunters thought all was lost. But pheasants are tough and, when wintering habitat is there, can survive one hell of an onslaught of snow and cold. » A goodly base of the core pheasant range experienced almost ideal nesting conditions with a relatively dry and warm spring, punctuated by moisture at just the right times in early summer to help nesting habitat growth and vibrancy. Nesting success makes birds for hunting, period. » Conditions trend from generally dry to downright droughty heading into fall, and that will impact hunting, from the careful approach you might bring to the hunt, to habitat status on the ground (including some emergency mowing, haying and grazing regionally). For pheasants, and all other manner of upland wildlife, habitat is everything. Likewise for hunters. And it is where communion happens in fall, with a bird dog romping ahead, our hearts glad and faces smiling and a rooster out there somewhere, giving us the good old slip. The 2023 Pheasant Hunting Forecast is presented by Sportsman’s Guide #pheasantsforever #pheasanthunting #pheasants #upland #uplandhunting

Pheasants Forever

12,346 次观看 • 3 年前

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!

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 • 7 个月前