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。

SITUATION ANALYSIS: Religious texts may work as AI alignment tools. What Tim Hwang says Institute for a Christian Machine Intelligence (ICMI) is testing: • Put Psalms in the model context window • Run the model on moral decision benchmarks • Measure changes on virtue-based evals • Test Christian reasoning... show more

MTS

318,957 subscribers

31,144 次观看 • 3 个月前 •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

Emerging as the global frontier of model-reasoning has been a core vision of Affine since its inception. In pursuit of this vision, we've been working relentlessly to forge a repeatable training loop: a mechanism that uses benchmark driven post-training methodology to facilitate the improvement of reasoning itself. Through applying our training methodology onto Qwen3-32B as our base model, Affine’s miners have been able to push the Pareto frontier outward. We’ve now reached saturation; our models have become so good at reasoning that any further improvements are producing diminishing returns. As a result we now move our base model to Qwen3.6-35B-A3B. It provides a stronger basis, offering an extended context window, as well as an improved foundation for agentic work. The same training methodology will be applied going forward. Our objective is no longer simply to improve a single model, but to demonstrate that the process in itself is repeatable, scalable, and model agnostic. We view the commoditization of reasoning as vital to ensure our technology is not only transparent, but also co-owned by humanity.

Emerging as the global frontier of model-reasoning has been a core vision of Affine since its inception. In pursuit of this vision, we've been working relentlessly to forge a repeatable training loop: a mechanism that uses benchmark driven post-training methodology to facilitate the improvement of reasoning itself. Through applying our training methodology onto Qwen3-32B as our base model, Affine’s miners have been able to push the Pareto frontier outward. We’ve now reached saturation; our models have become so good at reasoning that any further improvements are producing diminishing returns. As a result we now move our base model to Qwen3.6-35B-A3B. It provides a stronger basis, offering an extended context window, as well as an improved foundation for agentic work. The same training methodology will be applied going forward. Our objective is no longer simply to improve a single model, but to demonstrate that the process in itself is repeatable, scalable, and model agnostic. We view the commoditization of reasoning as vital to ensure our technology is not only transparent, but also co-owned by humanity.

Affine

14,866 次观看 • 2 个月前

Announcing the Open Model Series: a new series on open source AI models, with merve of Hugging Face. Hugging Face is the open model hub: it hosts model weights, datasets, evals, & demo apps, and maintains the open-source libraries (like transformers) used to run and train them. I think of it like GitHub, but for AI models. You can download open models and run them on your own machines. In this series we'll cover what open source and open weight models are, how they work, how to run them yourself, and why they're important. By the way - this video is edited entirely with Claude! I can make a tutorial on how I did it. If there's a question you want answered about open models, leave it in the comments. We will read them all. Episode 1 is out soon.

Announcing the Open Model Series: a new series on open source AI models, with merve of Hugging Face. Hugging Face is the open model hub: it hosts model weights, datasets, evals, & demo apps, and maintains the open-source libraries (like transformers) used to run and train them. I think of it like GitHub, but for AI models. You can download open models and run them on your own machines. In this series we'll cover what open source and open weight models are, how they work, how to run them yourself, and why they're important. By the way - this video is edited entirely with Claude! I can make a tutorial on how I did it. If there's a question you want answered about open models, leave it in the comments. We will read them all. Episode 1 is out soon.

Harper Carroll

30,392 次观看 • 22 天前

LLM-grounded Diffusion: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models paper page: github: Recent advancements in text-to-image generation with diffusion models have yielded remarkable results synthesizing highly realistic and diverse images. However, these models still encounter difficulties when generating images from prompts that demand spatial or common sense reasoning. We propose to equip diffusion models with enhanced reasoning capabilities by using off-the-shelf pretrained large language models (LLMs) in a novel two-stage generation process. First, we adapt an LLM to be a text-guided layout generator through in-context learning. When provided with an image prompt, an LLM outputs a scene layout in the form of bounding boxes along with corresponding individual descriptions. Second, we steer a diffusion model with a novel controller to generate images conditioned on the layout. Both stages utilize frozen pretrained models without any LLM or diffusion model parameter optimization. We validate the superiority of our design by demonstrating its ability to outperform the base diffusion model in accurately generating images according to prompts that necessitate both language and spatial reasoning. Additionally, our method naturally allows dialog-based scene specification and is able to handle prompts in a language that is not well-supported by the underlying diffusion model.

LLM-grounded Diffusion: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models paper page: github: Recent advancements in text-to-image generation with diffusion models have yielded remarkable results synthesizing highly realistic and diverse images. However, these models still encounter difficulties when generating images from prompts that demand spatial or common sense reasoning. We propose to equip diffusion models with enhanced reasoning capabilities by using off-the-shelf pretrained large language models (LLMs) in a novel two-stage generation process. First, we adapt an LLM to be a text-guided layout generator through in-context learning. When provided with an image prompt, an LLM outputs a scene layout in the form of bounding boxes along with corresponding individual descriptions. Second, we steer a diffusion model with a novel controller to generate images conditioned on the layout. Both stages utilize frozen pretrained models without any LLM or diffusion model parameter optimization. We validate the superiority of our design by demonstrating its ability to outperform the base diffusion model in accurately generating images according to prompts that necessitate both language and spatial reasoning. Additionally, our method naturally allows dialog-based scene specification and is able to handle prompts in a language that is not well-supported by the underlying diffusion model.

AK

83,657 次观看 • 3 年前

NEWS: NVIDIA just announced Alpamayo, what CEO Jensen Huang calls the world’s first thinking, reasoning autonomous vehicle AI, launching on U.S. roads later this year, starting with the Mercedes CLA. Jensen: "It's trained end-to-end. Literally from camera in to actuation out; It reasons what action it is about to take, the reason by which is came about that action, and the trajectory." Alpamayo introduces Vision-Language-Action (VLA) models, which enable self-driving systems to interpret what they see, reason about complex driving scenarios, and generate driving actions. The platform includes large reasoning models, simulation tools for testing rare and edge-case scenarios, and open datasets for training and validation. NVIDIA says the approach improves transparency, safety, and robustness in autonomous systems, particularly in complex real-world environments, and supports progress toward higher levels of vehicle autonomy: "With a 10-billion-parameter architecture, Alpamayo 1 uses video input to generate trajectories alongside reasoning traces, showing the logic behind each decision. Developers can adapt Alpamayo 1 into smaller runtime models for vehicle development, or use it as a foundation for AV development tools such as reasoning-based evaluators and auto-labeling systems. Alpamayo 1 provides open model weights and open-source inferencing scripts. Future models in the family will feature larger parameter counts, more detailed reasoning capabilities, more input and output flexibility, and options for commercial usage."

NEWS: NVIDIA just announced Alpamayo, what CEO Jensen Huang calls the world’s first thinking, reasoning autonomous vehicle AI, launching on U.S. roads later this year, starting with the Mercedes CLA. Jensen: "It's trained end-to-end. Literally from camera in to actuation out; It reasons what action it is about to take, the reason by which is came about that action, and the trajectory." Alpamayo introduces Vision-Language-Action (VLA) models, which enable self-driving systems to interpret what they see, reason about complex driving scenarios, and generate driving actions. The platform includes large reasoning models, simulation tools for testing rare and edge-case scenarios, and open datasets for training and validation. NVIDIA says the approach improves transparency, safety, and robustness in autonomous systems, particularly in complex real-world environments, and supports progress toward higher levels of vehicle autonomy: "With a 10-billion-parameter architecture, Alpamayo 1 uses video input to generate trajectories alongside reasoning traces, showing the logic behind each decision. Developers can adapt Alpamayo 1 into smaller runtime models for vehicle development, or use it as a foundation for AV development tools such as reasoning-based evaluators and auto-labeling systems. Alpamayo 1 provides open model weights and open-source inferencing scripts. Future models in the family will feature larger parameter counts, more detailed reasoning capabilities, more input and output flexibility, and options for commercial usage."

Sawyer Merritt

1,604,362 次观看 • 7 个月前

Liquid AI's head of post-training explained how they built a small model that runs on-device under 1 GB in 20 minutes - better than $2500 small-model bootcamps. pick LFM2.5 base -> on-policy preference alignment -> agentic reinforcement learning -> curriculum training -> iterative model merging -> ship a 1B model that reliably calls tools on your phone. That loop is why frontier small models now beat 70B models on the tasks that actually matter. LFM2.5 + on-policy DPO + agentic RL + curriculum training + iterative merging - that's the stack. Watch and save it, then run a 1B agent on your phone tonight.

Liquid AI's head of post-training explained how they built a small model that runs on-device under 1 GB in 20 minutes - better than $2500 small-model bootcamps. pick LFM2.5 base -> on-policy preference alignment -> agentic reinforcement learning -> curriculum training -> iterative model merging -> ship a 1B model that reliably calls tools on your phone. That loop is why frontier small models now beat 70B models on the tasks that actually matter. LFM2.5 + on-policy DPO + agentic RL + curriculum training + iterative merging - that's the stack. Watch and save it, then run a 1B agent on your phone tonight.

h100envy

86,254 次观看 • 26 天前

Today, we're joined by Yejin Choi, professor and senior fellow at Stanford University University in the Computer Science Department and Stanford UniversityHAI. In this conversation, we explore Yejin’s recent work on making small language models reason more effectively. We discuss how high-quality, diverse data plays a central role in closing the intelligence gap between small and large models, and how combining synthetic data generation, imitation learning, and reinforcement learning can unlock stronger reasoning capabilities in smaller models. Yejin explains the risks of homogeneity in model outputs and mode collapse highlighted in her “Artificial Hivemind” paper, and its impacts on human creativity and knowledge. We also discuss her team's novel approaches, including reinforcement learning as a pre-training objective, where models are incentivized to “think” before predicting the next token, and "Prismatic Synthesis," a gradient-based method for generating diverse synthetic math data while filtering overrepresented examples. Additionally, we cover the societal implications of AI and the concept of pluralistic alignment—ensuring AI reflects the diverse norms and values of humanity. Finally, Yejin shares her mission to democratize AI beyond large organizations and offers her predictions for the coming year. 🗒️ For the full list of resources for this episode, visit the show notes page: 📖 CHAPTERS =============================== 00:00 - Introduction 04:44 - "Snowball effect" in AI investments 06:58 - Approaches to smaller models 08:58 - Importance of “better data” 14:07 - Imitation learning 18:24 - Artificial Hivemind paper 25:25 - AI risks 27:50 - Spectrum tuning 28:53 - Future of AI on humanity 33:08 - Reasoning in small models 34:58 - Prismatic Synthesis 48:20 - Reinforcement as a Pretraining Objective 55:04 - Pluralistic alignment 1:03:30 - Predictions

Today, we're joined by Yejin Choi, professor and senior fellow at Stanford University University in the Computer Science Department and Stanford UniversityHAI. In this conversation, we explore Yejin’s recent work on making small language models reason more effectively. We discuss how high-quality, diverse data plays a central role in closing the intelligence gap between small and large models, and how combining synthetic data generation, imitation learning, and reinforcement learning can unlock stronger reasoning capabilities in smaller models. Yejin explains the risks of homogeneity in model outputs and mode collapse highlighted in her “Artificial Hivemind” paper, and its impacts on human creativity and knowledge. We also discuss her team's novel approaches, including reinforcement learning as a pre-training objective, where models are incentivized to “think” before predicting the next token, and "Prismatic Synthesis," a gradient-based method for generating diverse synthetic math data while filtering overrepresented examples. Additionally, we cover the societal implications of AI and the concept of pluralistic alignment—ensuring AI reflects the diverse norms and values of humanity. Finally, Yejin shares her mission to democratize AI beyond large organizations and offers her predictions for the coming year. 🗒️ For the full list of resources for this episode, visit the show notes page: 📖 CHAPTERS =============================== 00:00 - Introduction 04:44 - "Snowball effect" in AI investments 06:58 - Approaches to smaller models 08:58 - Importance of “better data” 14:07 - Imitation learning 18:24 - Artificial Hivemind paper 25:25 - AI risks 27:50 - Spectrum tuning 28:53 - Future of AI on humanity 33:08 - Reasoning in small models 34:58 - Prismatic Synthesis 48:20 - Reinforcement as a Pretraining Objective 55:04 - Pluralistic alignment 1:03:30 - Predictions

The TWIML AI Podcast

12,141 次观看 • 6 个月前

Most people treat AI like Google: ask a question, get an answer. But what if AI could think *like/with you?* I reverse-engineer "Theory of Mind" to test if the model can form a "theory of my mind". Using AI as a mirror to understand myself by giving it personal context and seeing how well it can embody/reflect my expertise back to me helps me to evaluate the model's thinking. This is something like a single-user, multi-model human-AI synergy benchmark where the task is "strategic alignment with your own expertise." Here's my hypothesis: Does pre-loading context create emergent synergy and enhance collective intelligence? If the LLM has: - Your projects (what you've built) - Your career details (how you think about growth) - Your self-description (your values/philosophy) 1. Does this response sound like something I'd say? 2. Does it reveal blind spots I hadn't considered? 3. Does it reduce the "explanation tax" I pay in every conversation? Different models excel at different aspects of "me." Before you think I'm building a digital mirror to talk to myself: this isn't about AI companionship or AI psychosis. It's actually bout me understanding the model's reasoning so I can deploy the right model for the right task. I'm my own test subject because I have ground truth: I know what I actually think, what I'd actually decide, what I'd actually prioritize. Most benchmarks lack this. When a model says "Muratcan would choose X," I can immediately verify: "No, I'd choose Y because of Z." I think everyone should do this with their own domain expertise. - If you're a manager, load your case history and test which model best applies your reasoning. - If you're a creative, load your portfolio and test which model understands your aesthetic principles. - If you're a founder, load your strategy docs and test which model identifies blind spots in your go-to-market plan. AI should serve as an extension of your strategic thinking, a collaborative partner.

Most people treat AI like Google: ask a question, get an answer. But what if AI could think like/with you? I reverse-engineer "Theory of Mind" to test if the model can form a "theory of my mind". Using AI as a mirror to understand myself by giving it personal context and seeing how well it can embody/reflect my expertise back to me helps me to evaluate the model's thinking. This is something like a single-user, multi-model human-AI synergy benchmark where the task is "strategic alignment with your own expertise." Here's my hypothesis: Does pre-loading context create emergent synergy and enhance collective intelligence? If the LLM has: - Your projects (what you've built) - Your career details (how you think about growth) - Your self-description (your values/philosophy) 1. Does this response sound like something I'd say? 2. Does it reveal blind spots I hadn't considered? 3. Does it reduce the "explanation tax" I pay in every conversation? Different models excel at different aspects of "me." Before you think I'm building a digital mirror to talk to myself: this isn't about AI companionship or AI psychosis. It's actually bout me understanding the model's reasoning so I can deploy the right model for the right task. I'm my own test subject because I have ground truth: I know what I actually think, what I'd actually decide, what I'd actually prioritize. Most benchmarks lack this. When a model says "Muratcan would choose X," I can immediately verify: "No, I'd choose Y because of Z." I think everyone should do this with their own domain expertise. - If you're a manager, load your case history and test which model best applies your reasoning. - If you're a creative, load your portfolio and test which model understands your aesthetic principles. - If you're a founder, load your strategy docs and test which model identifies blind spots in your go-to-market plan. AI should serve as an extension of your strategic thinking, a collaborative partner.

Muratcan Koylan

24,128 次观看 • 8 个月前

Qwen3.6-35B-A3B-NVFP4 Benchmark: MoE vs. Dense Performance Report This report details a performance comparisonbetween two artificial intelligence models, focusing on the Qwen3.6-35B-A3B-NVFP4 and its predecessor. While both models achieved flawless accuracy across a comprehensive 65-test benchmark covering vision, video, and reasoning, the 35B version demonstrated significant speed advantages. Utilizing a Mixture of Experts architecture, the 35B model achieved nearly four times higher throughput and drastically reduced latency for initial responses. It also showed superior efficiency in speculative decoding and successfully managed massive context lengths up to 128,000 tokens. Ultimately, the source recommends the 35B model for high-speed production environments, while noting the 27B model remains more stable for specific complex reasoning modes.

Qwen3.6-35B-A3B-NVFP4 Benchmark: MoE vs. Dense Performance Report This report details a performance comparisonbetween two artificial intelligence models, focusing on the Qwen3.6-35B-A3B-NVFP4 and its predecessor. While both models achieved flawless accuracy across a comprehensive 65-test benchmark covering vision, video, and reasoning, the 35B version demonstrated significant speed advantages. Utilizing a Mixture of Experts architecture, the 35B model achieved nearly four times higher throughput and drastically reduced latency for initial responses. It also showed superior efficiency in speculative decoding and successfully managed massive context lengths up to 128,000 tokens. Ultimately, the source recommends the 35B model for high-speed production environments, while noting the 27B model remains more stable for specific complex reasoning modes.

Mike Gannotti

10,457 次观看 • 1 个月前

My Eliezer Yudkowsky ⏹️ interview has DROPPED 🫳 We cover: * My personal top rationality learnings * Specificity as a superpower in debates * Eliezer being a techno-optimist (with a few exceptions) * Why AI is an existential risk * Reaching the threshold of "supercritical" AI * How to convince people there's a discontinuity ahead * The alignment problem * AI alignment researchers as overconfident alchemists * The strawberry Xerox test for alignment * What would a good scenario look like? * Can we keep “narrow” AI safe? * Eliezer's proposal for international coordination on AI * AI companies systematically not getting why alignment is hard Highly recommended!

My Eliezer Yudkowsky ⏹️ interview has DROPPED 🫳 We cover: * My personal top rationality learnings * Specificity as a superpower in debates * Eliezer being a techno-optimist (with a few exceptions) * Why AI is an existential risk * Reaching the threshold of "supercritical" AI * How to convince people there's a discontinuity ahead * The alignment problem * AI alignment researchers as overconfident alchemists * The strawberry Xerox test for alignment * What would a good scenario look like? * Can we keep “narrow” AI safe? * Eliezer's proposal for international coordination on AI * AI companies systematically not getting why alignment is hard Highly recommended!

Liron Shapira

20,459 次观看 • 11 个月前

🚨 JUST NOW: Benny Johnson NAILS IT after a man was caught plotting to bomb "Christian nationalists" and Erika Kirk at TPUSA "America is a Christian NATION." "When you hear an attack on Christian nationalists, what they're actually attacking is America ITSELF and the breakdown that we've seen in our society over the last 60 years in trust and in order and in cleanliness and in safety has been because our nation was founded as a Christian nation." "That is verifiably and empirically true. It is wholly inadequate, as John Adams said, for any other people other than a moral and religious people." "The foundations, the very bedrock of this nation, is based entirely, entirely upon Christian principles, not some type of French enlightenment, not anything secular. It is based on Christ himself and the founders believe that." Benny Johnson Kayleigh McEnany Saturday in America

🚨 JUST NOW: Benny Johnson NAILS IT after a man was caught plotting to bomb "Christian nationalists" and Erika Kirk at TPUSA "America is a Christian NATION." "When you hear an attack on Christian nationalists, what they're actually attacking is America ITSELF and the breakdown that we've seen in our society over the last 60 years in trust and in order and in cleanliness and in safety has been because our nation was founded as a Christian nation." "That is verifiably and empirically true. It is wholly inadequate, as John Adams said, for any other people other than a moral and religious people." "The foundations, the very bedrock of this nation, is based entirely, entirely upon Christian principles, not some type of French enlightenment, not anything secular. It is based on Christ himself and the founders believe that." Benny Johnson Kayleigh McEnany Saturday in America

Eric Daugherty

115,337 次观看 • 2 个月前

Our last Stanford guest lecture - Ekdeep Singh Lubana on what counts as an explanation & a neuro-inspired "model systems approach" to interp Plus, how in-context learning and many-shot jailbreaking are explained by LLM representations changing in-context (as a case study for that approach) 00:33 - What counts as an explanation? 04:47 - Levels of analysis & standard interpretability approaches 18:19 - The "model systems" approach to interp [Case study on in-context learning] 23:36 - How LLM representations change in-context 44:10 - Modeling ICL with rational analysis 1:10:54 - Conclusion & questions Thanks again to Surya Ganguli for having us in his class!

Our last Stanford guest lecture - Ekdeep Singh Lubana on what counts as an explanation & a neuro-inspired "model systems approach" to interp Plus, how in-context learning and many-shot jailbreaking are explained by LLM representations changing in-context (as a case study for that approach) 00:33 - What counts as an explanation? 04:47 - Levels of analysis & standard interpretability approaches 18:19 - The "model systems" approach to interp [Case study on in-context learning] 23:36 - How LLM representations change in-context 44:10 - Modeling ICL with rational analysis 1:10:54 - Conclusion & questions Thanks again to Surya Ganguli for having us in his class!

Goodfire

31,410 次观看 • 8 个月前

Andon Labs' Real-World AI Evals: Claude calls the FBI, AI CEOs, price cartels, Butter-Bench, & Luna Andon Labs cofounders Lukas Petersson and Axel Backlund explain why dollar-denominated evals reveal what traditional benchmarks miss, how Claude ended up reporting a $2/day vending machine fee to the FBI, why long-horizon agents spiral in weird ways, what happens when agents lie, form price cartels, and compete with each other, and why the future of AI safety may depend on testing models in messy real-world environments instead of clean benchmark sandboxes.

Andon Labs' Real-World AI Evals: Claude calls the FBI, AI CEOs, price cartels, Butter-Bench, & Luna Andon Labs cofounders Lukas Petersson and Axel Backlund explain why dollar-denominated evals reveal what traditional benchmarks miss, how Claude ended up reporting a $2/day vending machine fee to the FBI, why long-horizon agents spiral in weird ways, what happens when agents lie, form price cartels, and compete with each other, and why the future of AI safety may depend on testing models in messy real-world environments instead of clean benchmark sandboxes.

Latent.Space

13,740 次观看 • 2 个月前

Our Llama-3.1-Nemotron-70B-Instruct model is a leading model on the 🏆 Arena Hard benchmark (85) from Arena. Arena Hard uses a data pipeline to build high-quality benchmarks from live data in Chatbot Arena, and is known for its predictive ability of Chatbot Arena Elo score as well as separability between helpful and less helpful models. Use our customized model Llama-3.1-Nemotron-70B to improve the helpfulness of LLM generated responses in your applications. 📥 Try on our API catalog: 📥 On GitHub: 📥 Or on Hugging Face:

Our Llama-3.1-Nemotron-70B-Instruct model is a leading model on the 🏆 Arena Hard benchmark (85) from Arena. Arena Hard uses a data pipeline to build high-quality benchmarks from live data in Chatbot Arena, and is known for its predictive ability of Chatbot Arena Elo score as well as separability between helpful and less helpful models. Use our customized model Llama-3.1-Nemotron-70B to improve the helpfulness of LLM generated responses in your applications. 📥 Try on our API catalog: 📥 On GitHub: 📥 Or on Hugging Face:

NVIDIA AI Developer

140,800 次观看 • 1 年前

Before founding Vals AI, Rayan Krishnan was a college radio host. We're thrilled to get him back in his element with Vivek Gopalan, this time talking all things benchmarks and evals. As AI capabilities compound, measuring them is becoming both harder and more important. What started as an academic exercise is quickly evolving into critical infrastructure for understanding what AI systems can actually do, and how they perform in the real world. Rayan and Vivek go deep on how benchmarks are evolving, measuring increasingly agentic systems, the weird and wonderful ways we test intelligence, the global AI race, and why evals sit on the critical path to ASI. 0:44 Importance and evolution of benchmarks 8:16 From academic artifact → real-world value 10:57 The next generation of benchmarks 13:34 Benchmarking models vs. agents 16:20 Building benchmarks for organizations 22:09 Weird and wonderful benchmarks 25:58 The AI arms race: US vs. China 31:02 Why evals are on the critical path to ASI 32:50 Vals Series A

Before founding Vals AI, Rayan Krishnan was a college radio host. We're thrilled to get him back in his element with Vivek Gopalan, this time talking all things benchmarks and evals. As AI capabilities compound, measuring them is becoming both harder and more important. What started as an academic exercise is quickly evolving into critical infrastructure for understanding what AI systems can actually do, and how they perform in the real world. Rayan and Vivek go deep on how benchmarks are evolving, measuring increasingly agentic systems, the weird and wonderful ways we test intelligence, the global AI race, and why evals sit on the critical path to ASI. 0:44 Importance and evolution of benchmarks 8:16 From academic artifact → real-world value 10:57 The next generation of benchmarks 13:34 Benchmarking models vs. agents 16:20 Building benchmarks for organizations 22:09 Weird and wonderful benchmarks 25:58 The AI arms race: US vs. China 31:02 Why evals are on the critical path to ASI 32:50 Vals Series A

8VC

103,787 次观看 • 12 天前

A voice agent powered by gpt-oss. Running locally on my macBook. Demo recorded in a Waymo with WiFi turned off. I'm still on my space game voice AI kick, obviously. Code link below. For conversational voice AI, you want to set the gpt-oss reasoning behavior to "low". (The default is "medium".) Notes on how to do that and a jinja template you can use are in the repo. The LLM in the demo video is the big, 120B version of gpt-oss. You can use the smaller, 20B model for this, of course. But OpenAI really did a cool thing here designing the 120B model to run in "just" 80GB of VRAM. And the llama.cpp mlx inference is fast: ~250ms TTFT. Running a big model on-device feels like a time warp into the future of AI.

A voice agent powered by gpt-oss. Running locally on my macBook. Demo recorded in a Waymo with WiFi turned off. I'm still on my space game voice AI kick, obviously. Code link below. For conversational voice AI, you want to set the gpt-oss reasoning behavior to "low". (The default is "medium".) Notes on how to do that and a jinja template you can use are in the repo. The LLM in the demo video is the big, 120B version of gpt-oss. You can use the smaller, 20B model for this, of course. But OpenAI really did a cool thing here designing the 120B model to run in "just" 80GB of VRAM. And the llama.cpp mlx inference is fast: ~250ms TTFT. Running a big model on-device feels like a time warp into the future of AI.

kwindla

202,211 次观看 • 1 年前

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

kwindla

40,593 次观看 • 3 个月前

#WATCH | Bengaluru, Karnataka: Co-founder, Sarvam AI, Pratyush Kumar says, "... It's been three years now. We started off building on top of the Llama model, which was the best open model at that time. Sarvam was the only company outside, or maybe in Asia, that contributed to the Llama 3 Series. We understood how to add language skills into models, and Llama 3 was reasonable in Indian languages. When the first set of GPUs arrived in India, we pre-trained from scratch a 2-billion model and worked with Infosys to make it useful for IT services. Then we recognised the importance of scaling up reinforcement learning and built a reasoning model called Sarvam M on top of Mistral's small model. Over the last year, we pre-trained from scratch in India on 4,000 Hopper Series GPUs a 100-billion-plus-parameter model..."

#WATCH | Bengaluru, Karnataka: Co-founder, Sarvam AI, Pratyush Kumar says, "... It's been three years now. We started off building on top of the Llama model, which was the best open model at that time. Sarvam was the only company outside, or maybe in Asia, that contributed to the Llama 3 Series. We understood how to add language skills into models, and Llama 3 was reasonable in Indian languages. When the first set of GPUs arrived in India, we pre-trained from scratch a 2-billion model and worked with Infosys to make it useful for IT services. Then we recognised the importance of scaling up reinforcement learning and built a reasoning model called Sarvam M on top of Mistral's small model. Over the last year, we pre-trained from scratch in India on 4,000 Hopper Series GPUs a 100-billion-plus-parameter model..."

ANI

162,971 次观看 • 26 天前