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。

This week, we tested 3 latest models in our Game Arena Benchmark: → O3 → O4-mini → Gemini 2.5 Flash Across 4 games—Phoenix Wright, Sokoban, Candy Crush, and 2048—O3 dominated the zero-shot leaderboard, ranking #1 or #2 in nearly every task and outperforming previous SOTA models like O3-mini and... show more

Hao AI Lab

6,636 subscribers

15,105 次观看 • 1 年前 •via X (Twitter)

游戏科学技术

Anya Rossi• Live Now

Private livecam show

9 条评论

Hao AI Lab 的头像

Hao AI Lab1 年前

Leaderboard Snapshot: • O3 — 🥇 in Sokoban & 2048, 🥈 in Phoenix Wright & Candy Crush. • O4-mini — 🥇 in Candy Crush, lower ranks elsewhere. • Gemini 2.5 Flash — mid-tier across all games. O3 showed strong multi-modal reasoning, especially in spatial and long-horizon tasks, consistently placing in the top 2 across all games. O3 completely outperformed previous SOTA models like Gemini-2.5-pro and O3-mini. Gemini-2.5-flash delivered steady, cost-efficient performance without leading. For full results, check out the leaderboard here:

Hao AI Lab 的头像

Hao AI Lab1 年前

OpenAI’s latest models—O3 and O4-mini—both showed strong reasoning in Candy Crush. They didn't just play; they optimized—aiming to eliminate the most candies per move. 🍬 This task demands more than quick reaction: → Vision to understand the board 👁️ → Reasoning to evaluate all possible swaps 🤔 → Planning to pick the best move 🎯 O3 and O4-mini consistently pushed beyond basic matches, prioritizing smart plays. O4-mini averaged 129, just shy of our human-level baseline of 134, and well above the previous SOTA O3-mini at 106.3. 👀

Hao AI Lab 的头像

Hao AI Lab1 年前

We’re building more transparent, robust, and creative AI benchmarks—and we’d love your input. 💡📊 Got ideas for game-based evaluations? Have a challenge in mind? Drop your thoughts below! We also welcome more people to join our community, whether you're a researcher, builder, or just curious. → Leaderboard: → GitHub: → Website: Let’s push AI evaluation forward—together. 🚀

PowerBeatsVR 的头像

PowerBeatsVR3 年前

Get ready for a full-body VR workout that’s fun, fast, and intuitive — Play PowerBeatsVR (Now on Meta Quest) 🔥

Dylan Wolfe 的头像

Dylan Wolfe1 年前

I don’t know why there isn’t any comments. This is awesome! It shows that o3 is making real world tangible progress! Please don’t stop doing these!!

Hao AI Lab 的头像

Hao AI Lab1 年前

thank you!

Chase Brower 的头像

Chase Brower1 年前

Great work! o3 looks like a great model.

Hao AI Lab 的头像

Hao AI Lab1 年前

indeed, the visual reasoning reflects well on game play!

J Sam🌐 的头像

J Sam🌐1 年前

Very good It should be tested to play more mobile app games >pubg , freefire >dragonball legends, dokkan battle >pokemon unite, go, masters >sports betting apps >stardew valley, farmville and sims

相关视频

🚨 New Challenger: GROK joins the Game Arena Benchmark! We evaluated Grok3-mini-beta: thinkining on four games: 🧩 2048 | 🧱 Sokoban | 🍬 Candy Crush | 🎮 Phoenix Wright With fast progress, it’s already comparable to top models like OpenAI’s O1, previous O3-mini, and Gemini-2.5-Pro, ranked 🥇 1st in 2048 and 🥈 2nd in Sokoban. But the Grok3-beta Thinking API is still unreleased—leaving an unanswered question: Can it match or surpass O3’s dominant reasoning performance? For now, let’s take a closer look at how Grok3-mini-beta: thinking performs in gameplay so far. 👇

🚨 New Challenger: GROK joins the Game Arena Benchmark! We evaluated Grok3-mini-beta: thinkining on four games: 🧩 2048 | 🧱 Sokoban | 🍬 Candy Crush | 🎮 Phoenix Wright With fast progress, it’s already comparable to top models like OpenAI’s O1, previous O3-mini, and Gemini-2.5-Pro, ranked 🥇 1st in 2048 and 🥈 2nd in Sokoban. But the Grok3-beta Thinking API is still unreleased—leaving an unanswered question: Can it match or surpass O3’s dominant reasoning performance? For now, let’s take a closer look at how Grok3-mini-beta: thinking performs in gameplay so far. 👇

Hao AI Lab

20,121 次观看 • 1 年前

🔧🤖 New wave of open-source LLMs like Deekseek-R1-0528 and Qwen3-235B-A22B are leveling up with stronger agentic performance. We test them in head-to-head gameplay — the upgraded Deekseek-R1-0528 outsmarts strong reasoning models like o4-mini across several games and it nearly matches SOTA performance on Tetris, going toe-to-toe with o3. ✨🧠 Check out how R1 manages to clear lines in Tetris while other models still struggle 👇

🔧🤖 New wave of open-source LLMs like Deekseek-R1-0528 and Qwen3-235B-A22B are leveling up with stronger agentic performance. We test them in head-to-head gameplay — the upgraded Deekseek-R1-0528 outsmarts strong reasoning models like o4-mini across several games and it nearly matches SOTA performance on Tetris, going toe-to-toe with o3. ✨🧠 Check out how R1 manages to clear lines in Tetris while other models still struggle 👇

Hao AI Lab

36,573 次观看 • 1 年前

🚨 NEW: We made Claude, Gemini, o3 battle each other for world domination. We taught them Diplomacy—the strategy game where winning requires alliances, negotiation, and betrayal. Here's what happened: DeepSeek turned warmongering tyrant. Claude couldn't lie—everyone exploited it ruthlessly. Gemini 2.5 Pro nearly conquered Europe with brilliant tactics. Then o3 orchestrated a secret coalition, backstabbed every ally, and won. Why did we do this? The most popular AI benchmarks don't test deception. But as these models get deployed everywhere—from your email to your workplace—we need to know: Will they lie to get what they want? So Every 🪨 we built the ultimate test: AI Diplomacy, a dynamic benchmark that measures AI's ability to form alliances, negotiate, and betray each other. Watch them live below! Created from the ground up by alex duffy and Tyler Marques.

🚨 NEW: We made Claude, Gemini, o3 battle each other for world domination. We taught them Diplomacy—the strategy game where winning requires alliances, negotiation, and betrayal. Here's what happened: DeepSeek turned warmongering tyrant. Claude couldn't lie—everyone exploited it ruthlessly. Gemini 2.5 Pro nearly conquered Europe with brilliant tactics. Then o3 orchestrated a secret coalition, backstabbed every ally, and won. Why did we do this? The most popular AI benchmarks don't test deception. But as these models get deployed everywhere—from your email to your workplace—we need to know: Will they lie to get what they want? So Every 🪨 we built the ultimate test: AI Diplomacy, a dynamic benchmark that measures AI's ability to form alliances, negotiate, and betray each other. Watch them live below! Created from the ground up by alex duffy and Tyler Marques.

Dan Shipper

323,615 次观看 • 1 年前

I have been testing “OpenAI” o3 mini-high. The exposure of the “reasoning” on this model in my tests seem to not be the full reasoning, but a shifted and edited and tunicated version of what we would expect from a full exposure of the reasoning engine. See the video on the left. One can compare this to my FREE Open Source local newly quantized and distilled DeepSeek R1 8b (llama). See the video on the right. Clearly we see the full reading with DeepSeek R1 and something well tunicates with o3 mini-high (getting mad typing that name folks). Compare the quality of the outputs and the reasoning insights. Frankly there is no comparison. It feels like “OpenAI” begrudgingly released the reasoning output because they had to, but gave us something of a simulated reasoning and not the actual one. Of course my local lap top running an 8b model is slower, but all of the outputs and the actual creative writing is superior on the FREE model. The reasoning output, when legitimate can be many times more valuable than the answer. In the past week I have found thousands of ways to use in in many projects. Thus far I would have been far more impressed with o3 mini if I have never seen DeepSeek. However today it is too little—too late. The prompt: “Write a very creative and funny story on a pig and butterfly as friends living on Mars.”

I have been testing “OpenAI” o3 mini-high. The exposure of the “reasoning” on this model in my tests seem to not be the full reasoning, but a shifted and edited and tunicated version of what we would expect from a full exposure of the reasoning engine. See the video on the left. One can compare this to my FREE Open Source local newly quantized and distilled DeepSeek R1 8b (llama). See the video on the right. Clearly we see the full reading with DeepSeek R1 and something well tunicates with o3 mini-high (getting mad typing that name folks). Compare the quality of the outputs and the reasoning insights. Frankly there is no comparison. It feels like “OpenAI” begrudgingly released the reasoning output because they had to, but gave us something of a simulated reasoning and not the actual one. Of course my local lap top running an 8b model is slower, but all of the outputs and the actual creative writing is superior on the FREE model. The reasoning output, when legitimate can be many times more valuable than the answer. In the past week I have found thousands of ways to use in in many projects. Thus far I would have been far more impressed with o3 mini if I have never seen DeepSeek. However today it is too little—too late. The prompt: “Write a very creative and funny story on a pig and butterfly as friends living on Mars.”

Brian Roemmele

74,506 次观看 • 1 年前

We benchmarked leading multimodal foundation models (GPT-4o, Claude 3.5 Sonnet, Gemini, Llama, etc.) on standard computer vision tasks—from segmentation to surface normal estimation—using standard datasets like COCO and ImageNet. These models have made remarkable progress; however, it is unclear exactly where they stand in terms of understanding vision in detail. Especially when it comes to tasks beyond question-answering. How well do they understand an object's segments or geometry? Our analyses yield an assessment that is quantitatively and qualitatively detailed and is compatible with evaluations developed in the field of computer vision over the past decades. Observed trends: 🔹 The foundation models consistently underperform task-specific SOTA models across all tasks. However, they are respectable generalists, which is remarkable as they are presumably trained primarily on image-text-based tasks. 🔹 They perform semantic tasks notably better than geometric ones. 🔹 GPT-4o performs the best among non-reasoning models, getting the top position in 4 out of 6 tasks. 🔹 Reasoning models, e.g., o3, show improvements in geometric tasks. 🔹 The 'image generation' models, e.g., GPT-40 Image Generation, which have been natively trained multimodally, exhibit quirks. E.g., hallucinated objects, misalignment between the input and output, etc. 🔹 While the prompting techniques affect performance, better models exhibit less sensitivity to variations in prompts. We control for the variance introduced by the prompting methods in our experiments. 🌐 Detailed analyses, visualizations: ⌨️ code: 🧵 1/n

We benchmarked leading multimodal foundation models (GPT-4o, Claude 3.5 Sonnet, Gemini, Llama, etc.) on standard computer vision tasks—from segmentation to surface normal estimation—using standard datasets like COCO and ImageNet. These models have made remarkable progress; however, it is unclear exactly where they stand in terms of understanding vision in detail. Especially when it comes to tasks beyond question-answering. How well do they understand an object's segments or geometry? Our analyses yield an assessment that is quantitatively and qualitatively detailed and is compatible with evaluations developed in the field of computer vision over the past decades. Observed trends: 🔹 The foundation models consistently underperform task-specific SOTA models across all tasks. However, they are respectable generalists, which is remarkable as they are presumably trained primarily on image-text-based tasks. 🔹 They perform semantic tasks notably better than geometric ones. 🔹 GPT-4o performs the best among non-reasoning models, getting the top position in 4 out of 6 tasks. 🔹 Reasoning models, e.g., o3, show improvements in geometric tasks. 🔹 The 'image generation' models, e.g., GPT-40 Image Generation, which have been natively trained multimodally, exhibit quirks. E.g., hallucinated objects, misalignment between the input and output, etc. 🔹 While the prompting techniques affect performance, better models exhibit less sensitivity to variations in prompts. We control for the variance introduced by the prompting methods in our experiments. 🌐 Detailed analyses, visualizations: ⌨️ code: 🧵 1/n

Amir Zamir

73,398 次观看 • 1 年前

Which LLM reasons best when it doesn't have all the information? Enter LLM Poker Arena to find out. It's a Poker Playing benchmark where top reasoning models play Texas Hold'em poker against each other. Claude Opus 4.5, GPT-5.2, Gemini 2.5 Pro, and Grok 4 all sit at the same table and play full tournaments to see who finishes with the chips. Poker is very different when it comes to reasoning. It has to balance probabilistic reasoning, opponent modeling and make decisions under uncertainty. Poker is an interesting evaluation because it tests reasoning under incomplete information, something most coding benchmarks do not capture. In this tournaments the rules are: - Each LLM starts with $1,000 chips - Small and big blinds start at $25 / $50 - Blinds double every 3 minutes - All models run in their reasoning or thinking modes After the first 5 tournaments: - Claude Opus 4.5 with Thinking has 3 wins - GPT-5.2 has 2 wins - Grok 4 and Gemini 2.5 Pro have 0 wins Early results suggest Claude performs quite well at poker as well. Also five is a very small sample size. Planning to run many more tournaments, publish the full benchmark data and add a prediction market on top of it. Thanks for the suggestion clipz. Much more coming as part of Poker Cities !! This was built on Replit ⠕ using their AI integrations, which made it straightforward to connect Claude, GPT, and Gemini. What model do you think wins after 100 tournaments?

Which LLM reasons best when it doesn't have all the information? Enter LLM Poker Arena to find out. It's a Poker Playing benchmark where top reasoning models play Texas Hold'em poker against each other. Claude Opus 4.5, GPT-5.2, Gemini 2.5 Pro, and Grok 4 all sit at the same table and play full tournaments to see who finishes with the chips. Poker is very different when it comes to reasoning. It has to balance probabilistic reasoning, opponent modeling and make decisions under uncertainty. Poker is an interesting evaluation because it tests reasoning under incomplete information, something most coding benchmarks do not capture. In this tournaments the rules are: - Each LLM starts with $1,000 chips - Small and big blinds start at $25 / $50 - Blinds double every 3 minutes - All models run in their reasoning or thinking modes After the first 5 tournaments: - Claude Opus 4.5 with Thinking has 3 wins - GPT-5.2 has 2 wins - Grok 4 and Gemini 2.5 Pro have 0 wins Early results suggest Claude performs quite well at poker as well. Also five is a very small sample size. Planning to run many more tournaments, publish the full benchmark data and add a prediction market on top of it. Thanks for the suggestion clipz. Much more coming as part of Poker Cities !! This was built on Replit ⠕ using their AI integrations, which made it straightforward to connect Claude, GPT, and Gemini. What model do you think wins after 100 tournaments?

Anshul Dhawan

32,192 次观看 • 8 个月前

For developers, Gemini 3 Pro offers unparalleled results across every major AI benchmark compared to Gemini 2.5 Pro. Here three ways Gemini 3 Pro can improve your workflow today: 1. Gemini 3 Pro fits right into existing production agent and coding workflows. It is also integrated into development environment experiences across the ecosystem of developer products like Google AI Studio, Gemini CLI, and Android Studio. 2. We introduced Google Google Antigravity, a new agentic development platform that reimagines how the model and IDE work together. It enables developers to manage agents to work at a higher, task-oriented level collaborating with agents while they build. 3. We’ve built Google AI Studio to be your fastest path from a prompt to an AI-native app. Build mode lets you add AI capabilities faster than ever, automatically wiring up the right models and APIs, while features like annotations enable fast and intuitive iteration. Whether you’re building a one-shot prompt game or an interactive landing page from unstructured voice notes, you can bring any idea to life with Gemini 3.

For developers, Gemini 3 Pro offers unparalleled results across every major AI benchmark compared to Gemini 2.5 Pro. Here three ways Gemini 3 Pro can improve your workflow today: 1. Gemini 3 Pro fits right into existing production agent and coding workflows. It is also integrated into development environment experiences across the ecosystem of developer products like Google AI Studio, Gemini CLI, and Android Studio. 2. We introduced Google Google Antigravity, a new agentic development platform that reimagines how the model and IDE work together. It enables developers to manage agents to work at a higher, task-oriented level collaborating with agents while they build. 3. We’ve built Google AI Studio to be your fastest path from a prompt to an AI-native app. Build mode lets you add AI capabilities faster than ever, automatically wiring up the right models and APIs, while features like annotations enable fast and intuitive iteration. Whether you’re building a one-shot prompt game or an interactive landing page from unstructured voice notes, you can bring any idea to life with Gemini 3.

Google AI

136,648 次观看 • 10 个月前

ox alpha vs deepseek v4 flash vision vs grok 4.6 vs gemini 3.7 flash vs – on photo-to-3d four vision models got one photograph each and had to rebuild the place inside it as a Three.js scene. twelve scenes, twelve first-try runs, zero console errors the setup: one reference photo per scene, sent as an image on OpenRouter. the prompt never says what is in the picture – no "motel", no "bar", no "gas station". the model has to read the photo and rebuild it: layout, materials, hour of the day, and whatever is around the corner that the frame does not show tasks – three photographs of early-2000s america: 1. a motel at night, neon pylon lit, snow on the ground 2. an old new york tavern interior, tin ceiling, tiled floor 3. an abandoned service station in the california desert, midday sun each scene ships as one self-contained html file, procedural geometry and canvas textures only, no downloads. three timed camera shots, and shot 1 has to reproduce the framing of the reference photo models: xAI grok 4.6, Google DeepMind gemini 3.7 flash, DeepSeek deepseek v4 flash vision exp, and ox alpha – a stealth model on openrouter, free, no lab attached to it yet results: - wall clock, three scenes #1 gemini 3.7 flash – 11m 12s #2 deepseek v4 flash – 15m 20s #3 grok 4.6 – 28m 11s #4 ox alpha – 38m 54s - output tokens #1 gemini 3.7 flash – 77,396 #2 ox alpha – 87,613 #3 grok 4.6 – 105,687 #4 deepseek v4 flash – 127,884 - lines of code shipped #1 ox alpha – 2,090 #2 deepseek v4 flash – 2,291 #3 grok 4.6 – 3,529 #4 gemini 3.7 flash – 3,989 - total price #1 ox alpha – $0.000 #2 deepseek v4 flash – $0.091 #3 gemini 3.7 flash – $0.136 #4 grok 4.6 – $0.697 observations: • grok is 7.7x the price of deepseek. it is the only model that read the light – low sun, real shadows on the station, a cold night on the motel • gemini is the fastest and the least deliberate. 17,158 reasoning tokens against deepseek's 99,172, and it still shipped the most code – 3,989 lines • deepseek thought hardest and rendered plainest. 99,172 reasoning tokens, 5.8x gemini's, spent on layout rather than on light. its motel is the second best in the set for $0.030 • ox alpha is free and reads a photo as well as anything here – it lifted "family units / kitchenettes" off the pylon and redrew it in canvas conclusion: twelve scenes, four models, zero fixes, and the whole run cost $0.924! follow thehype. for 24/7 ai news, analysis and breakdowns

ox alpha vs deepseek v4 flash vision vs grok 4.6 vs gemini 3.7 flash vs – on photo-to-3d four vision models got one photograph each and had to rebuild the place inside it as a Three.js scene. twelve scenes, twelve first-try runs, zero console errors the setup: one reference photo per scene, sent as an image on OpenRouter. the prompt never says what is in the picture – no "motel", no "bar", no "gas station". the model has to read the photo and rebuild it: layout, materials, hour of the day, and whatever is around the corner that the frame does not show tasks – three photographs of early-2000s america: 1. a motel at night, neon pylon lit, snow on the ground 2. an old new york tavern interior, tin ceiling, tiled floor 3. an abandoned service station in the california desert, midday sun each scene ships as one self-contained html file, procedural geometry and canvas textures only, no downloads. three timed camera shots, and shot 1 has to reproduce the framing of the reference photo models: xAI grok 4.6, Google DeepMind gemini 3.7 flash, DeepSeek deepseek v4 flash vision exp, and ox alpha – a stealth model on openrouter, free, no lab attached to it yet results: - wall clock, three scenes #1 gemini 3.7 flash – 11m 12s #2 deepseek v4 flash – 15m 20s #3 grok 4.6 – 28m 11s #4 ox alpha – 38m 54s - output tokens #1 gemini 3.7 flash – 77,396 #2 ox alpha – 87,613 #3 grok 4.6 – 105,687 #4 deepseek v4 flash – 127,884 - lines of code shipped #1 ox alpha – 2,090 #2 deepseek v4 flash – 2,291 #3 grok 4.6 – 3,529 #4 gemini 3.7 flash – 3,989 - total price #1 ox alpha – $0.000 #2 deepseek v4 flash – $0.091 #3 gemini 3.7 flash – $0.136 #4 grok 4.6 – $0.697 observations: • grok is 7.7x the price of deepseek. it is the only model that read the light – low sun, real shadows on the station, a cold night on the motel • gemini is the fastest and the least deliberate. 17,158 reasoning tokens against deepseek's 99,172, and it still shipped the most code – 3,989 lines • deepseek thought hardest and rendered plainest. 99,172 reasoning tokens, 5.8x gemini's, spent on layout rather than on light. its motel is the second best in the set for $0.030 • ox alpha is free and reads a photo as well as anything here – it lifted "family units / kitchenettes" off the pylon and redrew it in canvas conclusion: twelve scenes, four models, zero fixes, and the whole run cost $0.924! follow thehype. for 24/7 ai news, analysis and breakdowns

thehype.

24,272 次观看 • 1 个月前

🚀 New Paper: Pixel Reasoner 🧠🖼️ How can Vision-Language Models (VLMs) perform chain-of-thought reasoning within the image itself? We introduce Pixel Reasoner, the first open-source framework that enables VLMs to “think in pixel space” through curiosity-driven reinforcement learning. Current VLMs reason only in text — even when grounded in rich images or videos, their logical steps are verbalized in natural language. This restricts their ability to interrogate visual evidence and demonstrate how conclusions are drawn. 🔍 So we ask: What if we could make VLMs "show their work" by reasoning directly in the pixel space? Inspired by GPT-o3’s "think-in-image" ability, we propose a framework where VLMs use interactive visual operations — zoom, select-frame, highlight — to reason through complex visual inputs. To do this, we design a two-stage training process: Instruction tuning with synthesized visual reasoning traces. Reinforcement learning with curiosity-driven reward to balance exploration between pixel and text reasoning ✨ With this, Pixel Reasoner achieves near-SoTA performance on many information-rich multimodal benchmarks: 📊 84% on InfographicsVQA 🧠 84% on V* benchmark 🧩 74% on TallyQA-Complex It also achieves strong accuracy of 68% on MVBench (a video benchmark). Website: Paper: Code: Demo: (coming soon)

🚀 New Paper: Pixel Reasoner 🧠🖼️ How can Vision-Language Models (VLMs) perform chain-of-thought reasoning within the image itself? We introduce Pixel Reasoner, the first open-source framework that enables VLMs to “think in pixel space” through curiosity-driven reinforcement learning. Current VLMs reason only in text — even when grounded in rich images or videos, their logical steps are verbalized in natural language. This restricts their ability to interrogate visual evidence and demonstrate how conclusions are drawn. 🔍 So we ask: What if we could make VLMs "show their work" by reasoning directly in the pixel space? Inspired by GPT-o3’s "think-in-image" ability, we propose a framework where VLMs use interactive visual operations — zoom, select-frame, highlight — to reason through complex visual inputs. To do this, we design a two-stage training process: Instruction tuning with synthesized visual reasoning traces. Reinforcement learning with curiosity-driven reward to balance exploration between pixel and text reasoning ✨ With this, Pixel Reasoner achieves near-SoTA performance on many information-rich multimodal benchmarks: 📊 84% on InfographicsVQA 🧠 84% on V* benchmark 🧩 74% on TallyQA-Complex It also achieves strong accuracy of 68% on MVBench (a video benchmark). Website: Paper: Code: Demo: (coming soon)

Wenhu Chen

82,829 次观看 • 1 年前

Today, we publicly released RoboCasa365, a large-scale simulation benchmark for training and systematically evaluating generalist robot models. Built upon our original RoboCasa framework, it offers: • 2,500 realistic kitchen environments; • 365 everyday tasks (basic skills + long-horizon mobile manipulation); • Over 3,200 objects with many articulated fixtures/appliances. All are designed for fully controlled, reproducible benchmarking of robotic policies. Progress in robotic foundation models is real. But it’s still hard to answer basic questions like: How close are we to general-purpose autonomy? What factors drive generalization? What are the model/data scaling curves like? Real-world eval is slow and noisy, and existing sims (like LIBERO, which we built 3 years ago) often lack sufficient task and scene diversity. This benchmark comes with 2,200+ hours of demonstrations and 500K+ trajectories to support studies of multi-task training, pretraining, and continual learning at scale. Check it out at

Today, we publicly released RoboCasa365, a large-scale simulation benchmark for training and systematically evaluating generalist robot models. Built upon our original RoboCasa framework, it offers: • 2,500 realistic kitchen environments; • 365 everyday tasks (basic skills + long-horizon mobile manipulation); • Over 3,200 objects with many articulated fixtures/appliances. All are designed for fully controlled, reproducible benchmarking of robotic policies. Progress in robotic foundation models is real. But it’s still hard to answer basic questions like: How close are we to general-purpose autonomy? What factors drive generalization? What are the model/data scaling curves like? Real-world eval is slow and noisy, and existing sims (like LIBERO, which we built 3 years ago) often lack sufficient task and scene diversity. This benchmark comes with 2,200+ hours of demonstrations and 500K+ trajectories to support studies of multi-task training, pretraining, and continual learning at scale. Check it out at

Yuke Zhu

24,394 次观看 • 6 个月前

GEMINI 3 LAUNCH IS HERE I got a SNEAK PEEK at Gemini 3 with Logan Kilpatrick (Google Deepmind), and it might be the most POWERFUL vibe-coding tool on the planet. A little breakdown: 1. Anyone can build 3D and casual games now You can vibecode full, playable 3D video games generated in minutes. Actual games with physics, characters, controls, and loops you can remix instantly. Pure insanity. I can see founders and brands spinning up games on the fly to ride trends and drive growth. 2. Intelligent apps are becoming the default We built apps where reasoning, memory, and multi-step planning were baked in from the start. Once you’re building apps with ACTUAL intelligence baked in, there’s a whole wave of new opportunities that weren’t possible before. 3. Gemini acts like a creative partner You describe the idea, Gemini fills in the gaps, challenges decisions, proposes alternatives, and iterates in real time. 4. Vibe coding hits a new level Gemini 3 can generate assets, code, game logic, UI, and narrative in one flow. Tools like Claude and Cursor feel fast. This feels like the next layer, the one where a single builder can compete with full teams. Logan Kilpatrick and I pushed Google Gemini 3 hard, and the outputs were solid. A few times we had to give it a few extra prompts but it took feedback really well. I think 1 year ago, a lot of people discounted Google in the AI arms race. Can you discount them anymore? Doubt it. After this, it feels like they at best leading, at worst leading. What do you think of Google's AI efforts/Gemini 3 My biggest takeaway was how it just felt like Gemini 3 had a little more vibe coding horsepower than anything I’ve used.

GEMINI 3 LAUNCH IS HERE I got a SNEAK PEEK at Gemini 3 with Logan Kilpatrick (Google Deepmind), and it might be the most POWERFUL vibe-coding tool on the planet. A little breakdown: 1. Anyone can build 3D and casual games now You can vibecode full, playable 3D video games generated in minutes. Actual games with physics, characters, controls, and loops you can remix instantly. Pure insanity. I can see founders and brands spinning up games on the fly to ride trends and drive growth. 2. Intelligent apps are becoming the default We built apps where reasoning, memory, and multi-step planning were baked in from the start. Once you’re building apps with ACTUAL intelligence baked in, there’s a whole wave of new opportunities that weren’t possible before. 3. Gemini acts like a creative partner You describe the idea, Gemini fills in the gaps, challenges decisions, proposes alternatives, and iterates in real time. 4. Vibe coding hits a new level Gemini 3 can generate assets, code, game logic, UI, and narrative in one flow. Tools like Claude and Cursor feel fast. This feels like the next layer, the one where a single builder can compete with full teams. Logan Kilpatrick and I pushed Google Gemini 3 hard, and the outputs were solid. A few times we had to give it a few extra prompts but it took feedback really well. I think 1 year ago, a lot of people discounted Google in the AI arms race. Can you discount them anymore? Doubt it. After this, it feels like they at best leading, at worst leading. What do you think of Google's AI efforts/Gemini 3 My biggest takeaway was how it just felt like Gemini 3 had a little more vibe coding horsepower than anything I’ve used.

GREG ISENBERG

73,838 次观看 • 10 个月前

[ENG TRANS] Ping: I want to say, before we went, we had no idea how massive the event was going to be. But once we arrived, it was huge—the arena was gigantic. Teshow: The arena was so beautiful. Ping: Yeah, and honestly, I've never played in a real basketball match in my life before. Mark: Like, a serious match like this? Ping: Never. The arena was so big... they said it's even larger than Impact Arena back home. Mark: Really Ping: Yeah! It felt so huge and high up. Mark: Roughly how many spectators were there? Ping: I think they said around 20,000 or so. Teshow: 20,000 capacity! Our Impact Arena has 2 or 3 tiers—lower section, middle, and the upper deck. But over there, it goes up another level—4 full tiers! Mark: It looks like a football stadium. Ping: Exactly, it was super tall! Looking up at the people on the top level, they looked tiny... #TeshowPing #GMMTVLiveHouse

[ENG TRANS] Ping: I want to say, before we went, we had no idea how massive the event was going to be. But once we arrived, it was huge—the arena was gigantic. Teshow: The arena was so beautiful. Ping: Yeah, and honestly, I've never played in a real basketball match in my life before. Mark: Like, a serious match like this? Ping: Never. The arena was so big... they said it's even larger than Impact Arena back home. Mark: Really Ping: Yeah! It felt so huge and high up. Mark: Roughly how many spectators were there? Ping: I think they said around 20,000 or so. Teshow: 20,000 capacity! Our Impact Arena has 2 or 3 tiers—lower section, middle, and the upper deck. But over there, it goes up another level—4 full tiers! Mark: It looks like a football stadium. Ping: Exactly, it was super tall! Looking up at the people on the top level, they looked tiny... #TeshowPing #GMMTVLiveHouse

S H I O R I .ೃ࿐

233,116 次观看 • 1 个月前

We’re excited to introduce KAME: Tandem Architecture for Enhancing Knowledge in Real-Time Speech-to-Speech Conversational AI, accepted at #ICASSP2026! 🐢 Blog Paper Can a speech AI think deeply without pausing to process? In real conversation, we don’t wait until we’ve fully worked out what we want to say—we start talking, and our thoughts catch up as the sentence unfolds. Fast speech-to-speech models achieve this, but their reasoning tends to stay shallow. Cascaded pipelines that route through a knowledgeable LLM are smarter, but the added latency breaks the flow—they fall back to "think, then speak." In our new paper, we propose a way to break this trade-off. We call it KAME (Turtle in Japanese). A speech-to-speech model handles the fast response loop and starts replying immediately. In parallel, a backend LLM runs asynchronously, generating response candidates that are continuously injected as "oracle" signals in real time. This shifts the AI paradigm from "think, then speak" to "speak while thinking." The backend LLM is completely swappable. You can plug in GPT-4.1, Claude Opus, or Gemini 2.5 Flash depending on the task without changing the frontend. In our experiments, Claude tended to score higher on reasoning, while GPT did better on humanities questions. Try the model yourself here:

We’re excited to introduce KAME: Tandem Architecture for Enhancing Knowledge in Real-Time Speech-to-Speech Conversational AI, accepted at #ICASSP2026! 🐢 Blog Paper Can a speech AI think deeply without pausing to process? In real conversation, we don’t wait until we’ve fully worked out what we want to say—we start talking, and our thoughts catch up as the sentence unfolds. Fast speech-to-speech models achieve this, but their reasoning tends to stay shallow. Cascaded pipelines that route through a knowledgeable LLM are smarter, but the added latency breaks the flow—they fall back to "think, then speak." In our new paper, we propose a way to break this trade-off. We call it KAME (Turtle in Japanese). A speech-to-speech model handles the fast response loop and starts replying immediately. In parallel, a backend LLM runs asynchronously, generating response candidates that are continuously injected as "oracle" signals in real time. This shifts the AI paradigm from "think, then speak" to "speak while thinking." The backend LLM is completely swappable. You can plug in GPT-4.1, Claude Opus, or Gemini 2.5 Flash depending on the task without changing the frontend. In our experiments, Claude tended to score higher on reasoning, while GPT did better on humanities questions. Try the model yourself here:

Sakana AI

293,975 次观看 • 4 个月前

🧠Today with OpenAI #Astra all over the news on ARC Prize , François Chollet we also have news from our work Jose Sánchez David Vivancos - e/acc #Neuraxon 2.0 by Qubic #OpenScience #aigarth Come-from-Beyond - Today for the First time we achieved Level 8 on one of the training games, finishing a all the levels. - And consitant 0.45x faster than humans in some game at level 1 like the one you see in the video, with the real small nxon brain with 8 spheres playing level 1 of r11l in the bottom. - But more imporantly this is real digitalbrains evolving daily to learn, #NoLLMs , #NoGPUs just bioinspired digital brain in action. - Also Today we moved a bit our last mark in the real Kaggle test to 0.28, this may look not to much we are beyond the middle of current scores now but wait: 1. - Literally you could run Nxon it in an old smartphone. 2.- For the Kaggle Game the top scoreres are using the GPU provided and emmbeding "small" #LLMs pretrained with harness to the task if we do so we could also climb the benchmark, but no we won't do that, we just use #Neuraxon 2.0 evolved over a couple months we are buildin #TrueAI not bruteforce compute/energy 3.- And the best is that Once Kaggle challenge ends (november 2nd) everything will be open source as allways in Qubic Open Science. 4.- Neuraxon 3.0 is advancing and will join the game soon, so stay tunned for more updates and thanks for your support!

🧠Today with OpenAI #Astra all over the news on ARC Prize , François Chollet we also have news from our work Jose Sánchez David Vivancos - e/acc #Neuraxon 2.0 by Qubic #OpenScience #aigarth Come-from-Beyond - Today for the First time we achieved Level 8 on one of the training games, finishing a all the levels. - And consitant 0.45x faster than humans in some game at level 1 like the one you see in the video, with the real small nxon brain with 8 spheres playing level 1 of r11l in the bottom. - But more imporantly this is real digitalbrains evolving daily to learn, #NoLLMs , #NoGPUs just bioinspired digital brain in action. - Also Today we moved a bit our last mark in the real Kaggle test to 0.28, this may look not to much we are beyond the middle of current scores now but wait: 1. - Literally you could run Nxon it in an old smartphone. 2.- For the Kaggle Game the top scoreres are using the GPU provided and emmbeding "small" #LLMs pretrained with harness to the task if we do so we could also climb the benchmark, but no we won't do that, we just use #Neuraxon 2.0 evolved over a couple months we are buildin #TrueAI not bruteforce compute/energy 3.- And the best is that Once Kaggle challenge ends (november 2nd) everything will be open source as allways in Qubic Open Science. 4.- Neuraxon 3.0 is advancing and will join the game soon, so stay tunned for more updates and thanks for your support!

David Vivancos - e/acc

25,190 次观看 • 20 天前

🚀Introducing VisualWebBench: A Comprehensive Benchmark for Multimodal Web Page Understanding and Grounding. 🤔What's this all about? Why this benchmark? > Back in Nov 2023, when we released MMMU ( a comprehensive multimodal understanding benchmark, we received feedback that it included very few UI screenshots. Considering the growing importance of UI understanding, especially with the rise of powerful agents like Devin ( which is built on the strong vision capability of #GPT4, we recognized the need for a benchmark focused on UI screenshot understanding.📸👀 > Multimodal #LLMs have significantly boosted web agents' performance on benchmarks like Mind2Web and WebArena. For instance, the SeeAct agent ( showcases the power of integrating vision into web agents. However, these benchmarks primarily evaluate the end-to-end task execution ability of web agents rather than their understanding of web pages. 🌉 Bridging the Gap with VisualWebBench > To provide a comprehensive evaluation of multimodal LLMs' web page understanding capabilities, we introduce VisualWebBench. Our benchmark spans 139 websites 🌐 across 12 domains 🏷️ and 87 sub-domains 🔍, ensuring a diverse and representative dataset. It assesses MLLMs at three levels: website-level, element-level, and action-level 📊, and encompasses seven tasks designed to evaluate understanding, OCR, grounding, and reasoning abilities 🧠💡. 😮 Surprising Findings > 🎉 Open-source models are catching up: Even though closed-source MLLMs are still leading the leaderboard, we are happy to see open-source models like LLaVA 1.6 34B achieve comparable performance to Gemini Pro. > 🧠 Grounding ability, crucial for developing MLLM-based web applications, is a weakness for most MLLMs. > 🖼️ Importance of Image Resolution: The limited image resolution handling capabilities of most open-source MLLMs restrict their utility in web scenarios, where rich text and elements are prevalent. > 🧱 Relatively strong correlation with general understanding benchmarks like MMMU but weak correlation with web agent benchmarks like Mind2Web. Web agent benchmarks primarily evaluate the end-to-end task execution ability of web agents, which involves a series of actions to accomplish a goal. In contrast, VisualWebBench emphasizes evaluating the foundational skills of MLLMs such as understanding and grounding web page elements. 💡Fun Fact > Claude Sonnet is better than Opus on our benchmark :) 🎓 Conclusion > VisualWebBench serves as a valuable resource for the community, driving research and development in the field of multimodal web page understanding and grounding. As MLLMs continue to evolve and improve, we look forward to seeing new applications and breakthroughs. We believe that our benchmark will contribute to the development of more powerful MLLMs in the web domain, ultimately leading to a more intuitive and efficient user experience on the web. Kudos to the student leads Junpeng Liu Yifan Song and the team Bill Yuchen Lin, Wai Lam, Graham Neubig, Yuanzhi Li! 👏 Check out more details in the Junpeng's thread👇

🚀Introducing VisualWebBench: A Comprehensive Benchmark for Multimodal Web Page Understanding and Grounding. 🤔What's this all about? Why this benchmark? > Back in Nov 2023, when we released MMMU ( a comprehensive multimodal understanding benchmark, we received feedback that it included very few UI screenshots. Considering the growing importance of UI understanding, especially with the rise of powerful agents like Devin ( which is built on the strong vision capability of #GPT4, we recognized the need for a benchmark focused on UI screenshot understanding.📸👀 > Multimodal #LLMs have significantly boosted web agents' performance on benchmarks like Mind2Web and WebArena. For instance, the SeeAct agent ( showcases the power of integrating vision into web agents. However, these benchmarks primarily evaluate the end-to-end task execution ability of web agents rather than their understanding of web pages. 🌉 Bridging the Gap with VisualWebBench > To provide a comprehensive evaluation of multimodal LLMs' web page understanding capabilities, we introduce VisualWebBench. Our benchmark spans 139 websites 🌐 across 12 domains 🏷️ and 87 sub-domains 🔍, ensuring a diverse and representative dataset. It assesses MLLMs at three levels: website-level, element-level, and action-level 📊, and encompasses seven tasks designed to evaluate understanding, OCR, grounding, and reasoning abilities 🧠💡. 😮 Surprising Findings > 🎉 Open-source models are catching up: Even though closed-source MLLMs are still leading the leaderboard, we are happy to see open-source models like LLaVA 1.6 34B achieve comparable performance to Gemini Pro. > 🧠 Grounding ability, crucial for developing MLLM-based web applications, is a weakness for most MLLMs. > 🖼️ Importance of Image Resolution: The limited image resolution handling capabilities of most open-source MLLMs restrict their utility in web scenarios, where rich text and elements are prevalent. > 🧱 Relatively strong correlation with general understanding benchmarks like MMMU but weak correlation with web agent benchmarks like Mind2Web. Web agent benchmarks primarily evaluate the end-to-end task execution ability of web agents, which involves a series of actions to accomplish a goal. In contrast, VisualWebBench emphasizes evaluating the foundational skills of MLLMs such as understanding and grounding web page elements. 💡Fun Fact > Claude Sonnet is better than Opus on our benchmark :) 🎓 Conclusion > VisualWebBench serves as a valuable resource for the community, driving research and development in the field of multimodal web page understanding and grounding. As MLLMs continue to evolve and improve, we look forward to seeing new applications and breakthroughs. We believe that our benchmark will contribute to the development of more powerful MLLMs in the web domain, ultimately leading to a more intuitive and efficient user experience on the web. Kudos to the student leads Junpeng Liu Yifan Song and the team Bill Yuchen Lin, Wai Lam, Graham Neubig, Yuanzhi Li! 👏 Check out more details in the Junpeng's thread👇

Xiang Yue

56,696 次观看 • 2 年前

Unstructured Thoughts about OpenAI o3, the nature of AGI, and Post-Labor Economics AGI just crossed a threshold—here’s why that matters and what we can do with it. I’ve been hammering on OpenAI’s new o3 model for a few days, long enough to watch the hype settle into something more interesting: utility. Benchmarks suggest a polite incremental bump; lived experience says we’ve entered a qualitatively different regime. o3 is the first model that feels faster than my ability to absorb its output. My brain—not the AI—has become the bottleneck. A new ceiling for human cognition? Most discussions of “alien intelligence” forget that we share the same sandbox: mathematics, physics, code, natural language. What shifts is cognitive horizon—the totality you can mentally represent and manipulate. o3 expands that horizon in real time. In an afternoon it consolidated two years of my work on post‑labor economics, stress‑tested the logic, surfaced data sources, and offered to autogenerate the Python notebooks. The cost of insight has collapsed from years to hours. If you merely outsource thought, you’ll stagnate. If you treat the model as a sparring partner—interrogating, refining, iterating—you’ll compound your own intelligence. Exponential leverage is now a choice, not a privilege. What o3 got right about my health project? I dumped the entire history of my chronic‑fatigue recovery protocol—including the five‑axis “burnout pentagram”—into memory and asked the model where I’d gone astray. It corrected a handful of minor assumptions and, more importantly, recalibrated my timeline: six‑to‑eight months of recovery left instead of eighteen. That’s not “replace your doctor” advice; it’s proof that large‑context reasoning is finally clinically useful. Post‑Labor Economics: the sketch that o3 and I built in one sitting 1. Metric 1 – Economic Agency Index (EAI) Income decomposed into wages, property, and transfers. The higher the property share, the more “post‑labor” you already are. 2. Metric 2 – Collective Purchasing Power (CPP) How much capital a county can mobilize without taxation or new debt. Rising CPP means you are compounding local prosperity. Interventions happen at the county level (subsidiarity): solar co‑ops in Arizona, riverfront greenways in the Midwest, data‑center dividends in fiber‑rich exurbs. Ownership is local, revenue is distributed, migration equilibrates naturally, and environmental stewardship becomes self‑interest rather than moral theater. UBI morphs from last‑ditch transfer to one of several levers for raising EAI. The bigger picture: AGI isn’t an oracle descending from the sky; it’s a time‑compression engine. Every minute you spend learning how to learn with it buys you an hour you would have burned doing rote synthesis. The frontier question is no longer “Will the machines replace us?” but “How fast can we upgrade ourselves in partnership with them?” What’s next? I’m cleaning the data, building the national EAI/CPP dashboard, and pressure‑testing the whole framework. I’ll publish the notebooks (or let o3 do it) once the numbers are solid. Meanwhile, I want to hear from you: Where does o3 add the most leverage in your world? Which of the post‑labor metrics feels wrong—or dangerously right? What failure mode should falsify this thesis? Drop your critique, your data source, or your wild counter‑proposal in the comments. Let’s map the edge of this new cognitive horizon together. —Dave

Unstructured Thoughts about OpenAI o3, the nature of AGI, and Post-Labor Economics AGI just crossed a threshold—here’s why that matters and what we can do with it. I’ve been hammering on OpenAI’s new o3 model for a few days, long enough to watch the hype settle into something more interesting: utility. Benchmarks suggest a polite incremental bump; lived experience says we’ve entered a qualitatively different regime. o3 is the first model that feels faster than my ability to absorb its output. My brain—not the AI—has become the bottleneck. A new ceiling for human cognition? Most discussions of “alien intelligence” forget that we share the same sandbox: mathematics, physics, code, natural language. What shifts is cognitive horizon—the totality you can mentally represent and manipulate. o3 expands that horizon in real time. In an afternoon it consolidated two years of my work on post‑labor economics, stress‑tested the logic, surfaced data sources, and offered to autogenerate the Python notebooks. The cost of insight has collapsed from years to hours. If you merely outsource thought, you’ll stagnate. If you treat the model as a sparring partner—interrogating, refining, iterating—you’ll compound your own intelligence. Exponential leverage is now a choice, not a privilege. What o3 got right about my health project? I dumped the entire history of my chronic‑fatigue recovery protocol—including the five‑axis “burnout pentagram”—into memory and asked the model where I’d gone astray. It corrected a handful of minor assumptions and, more importantly, recalibrated my timeline: six‑to‑eight months of recovery left instead of eighteen. That’s not “replace your doctor” advice; it’s proof that large‑context reasoning is finally clinically useful. Post‑Labor Economics: the sketch that o3 and I built in one sitting 1. Metric 1 – Economic Agency Index (EAI) Income decomposed into wages, property, and transfers. The higher the property share, the more “post‑labor” you already are. 2. Metric 2 – Collective Purchasing Power (CPP) How much capital a county can mobilize without taxation or new debt. Rising CPP means you are compounding local prosperity. Interventions happen at the county level (subsidiarity): solar co‑ops in Arizona, riverfront greenways in the Midwest, data‑center dividends in fiber‑rich exurbs. Ownership is local, revenue is distributed, migration equilibrates naturally, and environmental stewardship becomes self‑interest rather than moral theater. UBI morphs from last‑ditch transfer to one of several levers for raising EAI. The bigger picture: AGI isn’t an oracle descending from the sky; it’s a time‑compression engine. Every minute you spend learning how to learn with it buys you an hour you would have burned doing rote synthesis. The frontier question is no longer “Will the machines replace us?” but “How fast can we upgrade ourselves in partnership with them?” What’s next? I’m cleaning the data, building the national EAI/CPP dashboard, and pressure‑testing the whole framework. I’ll publish the notebooks (or let o3 do it) once the numbers are solid. Meanwhile, I want to hear from you: Where does o3 add the most leverage in your world? Which of the post‑labor metrics feels wrong—or dangerously right? What failure mode should falsify this thesis? Drop your critique, your data source, or your wild counter‑proposal in the comments. Let’s map the edge of this new cognitive horizon together. —Dave

David Shapiro (L/0)

45,581 次观看 • 1 年前

Today, I had the privilege of speaking to the HUMAIN team at our CEO Townhall. I often say this, and I mean it deeply: I am living my dream... working in a country where energy goes beyond oil. It’s in the people, the ambition, and the belief in building a better future. Saudi Arabia is unlike anywhere else. The hospitality is second to none. The vision is bold. And the commitment to shaping the future is real. What we are building at HUMAIN is foundational. We are not only participating in the AI era, but redefining it. Shifting the narrative from experimentation to real value creation. But more importantly, we are not afraid to challenge what an organization should look like in an AI-native world. In fact, we are pioneering it. We are actively reshaping how we work: - Moving toward a future where AI agents do the work - Empowering our people to focus on high-value thinking, creativity, and decision-making - Building strong foundations in security, governance, and guardrails - Continuously enhancing our models, systems, and operating frameworks This is not theory. This is happening now. What I saw today from our teams gives me absolute confidence: - Products that are not just innovative, but game-changing - Teams building at a speed and quality that challenge global norms - A culture focused on execution, ownership, and impact At #LEAP2026 this year, we will go beyond vision. We will show: Real demos. Real products. Some of them are a first of their kind in the world. And they are built right here in Saudi Arabia. I could not be more proud of this team. What they have accomplished in such a short time is remarkable. This is just the beginning. The future is not something we wait for. It’s something we build. #HUMAIN #LEAP #TheEndOfLimits #AI

Today, I had the privilege of speaking to the HUMAIN team at our CEO Townhall. I often say this, and I mean it deeply: I am living my dream... working in a country where energy goes beyond oil. It’s in the people, the ambition, and the belief in building a better future. Saudi Arabia is unlike anywhere else. The hospitality is second to none. The vision is bold. And the commitment to shaping the future is real. What we are building at HUMAIN is foundational. We are not only participating in the AI era, but redefining it. Shifting the narrative from experimentation to real value creation. But more importantly, we are not afraid to challenge what an organization should look like in an AI-native world. In fact, we are pioneering it. We are actively reshaping how we work: - Moving toward a future where AI agents do the work - Empowering our people to focus on high-value thinking, creativity, and decision-making - Building strong foundations in security, governance, and guardrails - Continuously enhancing our models, systems, and operating frameworks This is not theory. This is happening now. What I saw today from our teams gives me absolute confidence: - Products that are not just innovative, but game-changing - Teams building at a speed and quality that challenge global norms - A culture focused on execution, ownership, and impact At #LEAP2026 this year, we will go beyond vision. We will show: Real demos. Real products. Some of them are a first of their kind in the world. And they are built right here in Saudi Arabia. I could not be more proud of this team. What they have accomplished in such a short time is remarkable. This is just the beginning. The future is not something we wait for. It’s something we build. #HUMAIN #LEAP #TheEndOfLimits #AI

Tareq Amin

17,856 次观看 • 5 个月前

Tyrese Maxey was asked about Knicks fans filling the arena in Philadelphia: "It absolutely sucks, if I'm being honest. It just sucks. That's really all I can say about it, man. It's hard. It's definitely difficult. It's only one way to put a stop to it and it's, we have to go out there and win these games. Just being completely honest, we were better when we played them in the Garden this entire season. I know we lost Game 2 and 1, but Game 2 was better. The regular season was better. I was telling them, it felt louder here for them than it did in the Garden. We got to put a stop to it as a team. Winning these games, that's gonna make our fans louder than theirs or whatever. I don't know how to keep them out. I don't know the logistics of it, but it does suck. I can't even lie. It definitely does suck."

Tyrese Maxey was asked about Knicks fans filling the arena in Philadelphia: "It absolutely sucks, if I'm being honest. It just sucks. That's really all I can say about it, man. It's hard. It's definitely difficult. It's only one way to put a stop to it and it's, we have to go out there and win these games. Just being completely honest, we were better when we played them in the Garden this entire season. I know we lost Game 2 and 1, but Game 2 was better. The regular season was better. I was telling them, it felt louder here for them than it did in the Garden. We got to put a stop to it as a team. Winning these games, that's gonna make our fans louder than theirs or whatever. I don't know how to keep them out. I don't know the logistics of it, but it does suck. I can't even lie. It definitely does suck."

SNY Knicks

1,985,673 次观看 • 4 个月前

The frontier labs will crush you. That’s the first thing first friends say when I tell them I’m working on agents for biological research. It’s what I thought too before I started learning about biological agents. Now I’m convinced this space will be dominated by small teams building narrowly scoped, domain-specific models that uniquely embrace the complexity of living systems. So I’m diving in with a team of scientists and engineers who spent years watching models fail in real research settings - and built the “scar tissue” to understand why. We’re building Applied Scientific Intelligence, and some of the early results are encouraging: our models for biological data analysis & literature rank #1 on global benchmarks - outperforming ChatGPT, Claude, Gemini, as well as AI science companies with deep pockets (we’re still bootstrapped ;)). This recently caught the attention of NVIDIA and we partnered with them on our latest model: a new literature agent called “Alexandria” that’s going live in alpha this coming week. DM me if you want to try it out (or sign up in the link below). My main task now is building the distribution engine to get our agents in the hands of biologists. Estimates put the number of researchers at 9-10M globally. I think this number spikes in the next years as science starts to compound like software. Find an exponential curve and get in front of it, they say. Here goes.

The frontier labs will crush you. That’s the first thing first friends say when I tell them I’m working on agents for biological research. It’s what I thought too before I started learning about biological agents. Now I’m convinced this space will be dominated by small teams building narrowly scoped, domain-specific models that uniquely embrace the complexity of living systems. So I’m diving in with a team of scientists and engineers who spent years watching models fail in real research settings - and built the “scar tissue” to understand why. We’re building Applied Scientific Intelligence, and some of the early results are encouraging: our models for biological data analysis & literature rank #1 on global benchmarks - outperforming ChatGPT, Claude, Gemini, as well as AI science companies with deep pockets (we’re still bootstrapped ;)). This recently caught the attention of NVIDIA and we partnered with them on our latest model: a new literature agent called “Alexandria” that’s going live in alpha this coming week. DM me if you want to try it out (or sign up in the link below). My main task now is building the distribution engine to get our agents in the hands of biologists. Estimates put the number of researchers at 9-10M globally. I think this number spikes in the next years as science starts to compound like software. Find an exponential curve and get in front of it, they say. Here goes.

Nate Hindman

13,890 次观看 • 4 个月前