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。

New Anthropic research: Emotion concepts and their function in a large language model. All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.

Anthropic

1,588,165 subscribers

3,934,583 次观看 • 4 个月前 •via X (Twitter)

新闻政治教育科学技术

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

New Anthropic research: Tracing the thoughts of a large language model. We built a "microscope" to inspect what happens inside AI models and use it to understand Claude’s (often complex and surprising) internal mechanisms.

New Anthropic research: Tracing the thoughts of a large language model. We built a "microscope" to inspect what happens inside AI models and use it to understand Claude’s (often complex and surprising) internal mechanisms.

Anthropic

1,573,005 次观看 • 1 年前

🚨 New Research LLMs are trained only on text... Yet their internal representations progressively organize in ways that resemble human perceptual geometry across different domains (like color, pitch, emotion and taste), with the structures peaking in intermediate layers before attenuating in deeper representations. 🥳 Accepted at ICML Mechanistic Interpretability Workshop 2026

🚨 New Research LLMs are trained only on text... Yet their internal representations progressively organize in ways that resemble human perceptual geometry across different domains (like color, pitch, emotion and taste), with the structures peaking in intermediate layers before attenuating in deeper representations. 🥳 Accepted at ICML Mechanistic Interpretability Workshop 2026

Lossfunk

63,914 次观看 • 1 个月前

New Anthropic research: Natural Language Autoencoders. Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read. Here, we train Claude to translate its activations into human-readable text.

New Anthropic research: Natural Language Autoencoders. Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read. Here, we train Claude to translate its activations into human-readable text.

Anthropic

2,516,167 次观看 • 3 个月前

This is how large language models turn objects to vector representations. In this video, we explore how large language models (LLMs) convert objects into internal representations, especially when translating between languages like English and Hindi. Using real-world examples, we highlight the challenges of gender inference, grammatical structure, and why direct word-to-word translations often fail. If you're curious about how LLMs deal with multilingual contexts and what it takes to improve translation quality across languages, this video is for you. #LLMs #Vectors #LCM

This is how large language models turn objects to vector representations. In this video, we explore how large language models (LLMs) convert objects into internal representations, especially when translating between languages like English and Hindi. Using real-world examples, we highlight the challenges of gender inference, grammatical structure, and why direct word-to-word translations often fail. If you're curious about how LLMs deal with multilingual contexts and what it takes to improve translation quality across languages, this video is for you. #LLMs #Vectors #LCM

Gaurav Sen

27,368 次观看 • 1 年前

Sometimes in life it's really tough for everyone to hold their inner self and their true emotions because when they realized that the end is near... 🥹 We all can feel you GOAT 🖤

Sometimes in life it's really tough for everyone to hold their inner self and their true emotions because when they realized that the end is near... 🥹 We all can feel you GOAT 🖤

Friendly Wrestling

544,949 次观看 • 1 年前

Can we capture emotions in a few universal categories? This seems intuitive... but join this week's Inner Cosmos podcast with guest Lisa Feldman Barrett (Lisa Feldman Barrett) to find out why this textbook model of emotions is the wrong way to think about it. #brain #emotion

Can we capture emotions in a few universal categories? This seems intuitive... but join this week's Inner Cosmos podcast with guest Lisa Feldman Barrett (Lisa Feldman Barrett) to find out why this textbook model of emotions is the wrong way to think about it. #brain #emotion

David Eagleman

13,996 次观看 • 1 年前

🐹 “Normal” is about an emotion I can relate to. It is about the feelings behind the scenes of all the glamour. So we wanted to talk about those emotions. 🐹 For example, sometimes after our concerts, we can feel a kind of emptiness. 🐹 So we wanted to share those emotions and connect with people.

🐹 “Normal” is about an emotion I can relate to. It is about the feelings behind the scenes of all the glamour. So we wanted to talk about those emotions. 🐹 For example, sometimes after our concerts, we can feel a kind of emptiness. 🐹 So we wanted to share those emotions and connect with people.

jin files

223,320 次观看 • 4 个月前

3D-LLM: Injecting the 3D World into Large Language Models paper page: Large language models (LLMs) and Vision-Language Models (VLMs) have been proven to excel at multiple tasks, such as commonsense reasoning. Powerful as these models can be, they are not grounded in the 3D physical world, which involves richer concepts such as spatial relationships, affordances, physics, layout, and so on. In this work, we propose to inject the 3D world into large language models and introduce a whole new family of 3D-LLMs. Specifically, 3D-LLMs can take 3D point clouds and their features as input and perform a diverse set of 3D-related tasks, including captioning, dense captioning, 3D question answering, task decomposition, 3D grounding, 3D-assisted dialog, navigation, and so on. Using three types of prompting mechanisms that we design, we are able to collect over 300k 3D-language data covering these tasks. To efficiently train 3D-LLMs, we first utilize a 3D feature extractor that obtains 3D features from rendered multi- view images. Then, we use 2D VLMs as our backbones to train our 3D-LLMs. By introducing a 3D localization mechanism, 3D-LLMs can better capture 3D spatial information. Experiments on ScanQA show that our model outperforms state-of-the-art baselines by a large margin (e.g., the BLEU-1 score surpasses state-of-the-art score by 9%). Furthermore, experiments on our held-in datasets for 3D captioning, task composition, and 3D-assisted dialogue show that our model outperforms 2D VLMs. Qualitative examples also show that our model could perform more tasks beyond the scope of existing LLMs and VLMs.

3D-LLM: Injecting the 3D World into Large Language Models paper page: Large language models (LLMs) and Vision-Language Models (VLMs) have been proven to excel at multiple tasks, such as commonsense reasoning. Powerful as these models can be, they are not grounded in the 3D physical world, which involves richer concepts such as spatial relationships, affordances, physics, layout, and so on. In this work, we propose to inject the 3D world into large language models and introduce a whole new family of 3D-LLMs. Specifically, 3D-LLMs can take 3D point clouds and their features as input and perform a diverse set of 3D-related tasks, including captioning, dense captioning, 3D question answering, task decomposition, 3D grounding, 3D-assisted dialog, navigation, and so on. Using three types of prompting mechanisms that we design, we are able to collect over 300k 3D-language data covering these tasks. To efficiently train 3D-LLMs, we first utilize a 3D feature extractor that obtains 3D features from rendered multi- view images. Then, we use 2D VLMs as our backbones to train our 3D-LLMs. By introducing a 3D localization mechanism, 3D-LLMs can better capture 3D spatial information. Experiments on ScanQA show that our model outperforms state-of-the-art baselines by a large margin (e.g., the BLEU-1 score surpasses state-of-the-art score by 9%). Furthermore, experiments on our held-in datasets for 3D captioning, task composition, and 3D-assisted dialogue show that our model outperforms 2D VLMs. Qualitative examples also show that our model could perform more tasks beyond the scope of existing LLMs and VLMs.

AK

249,798 次观看 • 3 年前

$New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.$

New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.

Anthropic

10,621,015 次观看 • 1 个月前

Normally, I would be dead against using leftist concepts and language to articulate right-wing arguments, but in the case of Henry Nowak, I think we have a clear exception.

Normally, I would be dead against using leftist concepts and language to articulate right-wing arguments, but in the case of Henry Nowak, I think we have a clear exception.

RAW EGG NATIONALIST

72,189 次观看 • 2 个月前

Time for a new Qubic #Science Demo to understand better some of the many parameters we are recreating at the #Neuraxon BioInspired #TrueAI Model, that will run in the Qubic network with #aigarth Today a deep dive with the NeuroModulators, a fundamental piece of biochemistry that our brains need, and the #NeuraxonMoodMixer will help grasp the key concepts. You can play and learn with the demo here: Meanwhile we keep building the Science: And improving the Research Outcomes, all #OpenScience More updates on research comming soon, and more demos for other concepts the underlaying AI model uses.

Time for a new Qubic #Science Demo to understand better some of the many parameters we are recreating at the #Neuraxon BioInspired #TrueAI Model, that will run in the Qubic network with #aigarth Today a deep dive with the NeuroModulators, a fundamental piece of biochemistry that our brains need, and the #NeuraxonMoodMixer will help grasp the key concepts. You can play and learn with the demo here: Meanwhile we keep building the Science: And improving the Research Outcomes, all #OpenScience More updates on research comming soon, and more demos for other concepts the underlaying AI model uses.

David Vivancos - e/acc

26,558 次观看 • 6 个月前

“Travis is a fierce competitor … sometimes in the heat of battle, emotions show. Sometimes we have confrontations, whether it’s with coaches or players, but I think it’s all out of competitive spirit.” Chris Jones on playing with Travis Kelce.

“Travis is a fierce competitor … sometimes in the heat of battle, emotions show. Sometimes we have confrontations, whether it’s with coaches or players, but I think it’s all out of competitive spirit.” Chris Jones on playing with Travis Kelce.

KC Sports Network

61,327 次观看 • 10 个月前

What if your thoughts aren’t your own? Professors Elan Barenholtz and William Hahn propose that language is not a tool we use but a self-generating organism that uses us. In this mind-bending live discussion at the University of Toronto, they argue that language installs itself in our minds like software, runs autonomously, and shapes behavior at a deeper level than we realize. Drawing on LLMs, autoregression, and cognitive science, they suggest your brain may function like a predictive engine and that “memory,” “self,” and even “God” may just be tokens in an informational system.

What if your thoughts aren’t your own? Professors Elan Barenholtz and William Hahn propose that language is not a tool we use but a self-generating organism that uses us. In this mind-bending live discussion at the University of Toronto, they argue that language installs itself in our minds like software, runs autonomously, and shapes behavior at a deeper level than we realize. Drawing on LLMs, autoregression, and cognitive science, they suggest your brain may function like a predictive engine and that “memory,” “self,” and even “God” may just be tokens in an informational system.

Curt Jaimungal

52,269 次观看 • 1 年前

Trump: "I hate to say this in front of these young people. But we killed all their leadership. And then they met to choose new leaders, and we killed all of them. And now we have a new group and we can easily do that. But let's see how they turn out. We have really regime change."

Trump: "I hate to say this in front of these young people. But we killed all their leadership. And then they met to choose new leaders, and we killed all of them. And now we have a new group and we can easily do that. But let's see how they turn out. We have really regime change."

Aaron Rupar

54,792 次观看 • 4 个月前

Looks like Floppy, hits like Motion Weak 📝👌 Love this look from DII St. Edward's Men's Basketball ! They were one of the best teams in the nation at scoring off of screens and we can see why with their off ball concepts and the pace on their cuts!

Looks like Floppy, hits like Motion Weak 📝👌 Love this look from DII St. Edward's Men's Basketball ! They were one of the best teams in the nation at scoring off of screens and we can see why with their off ball concepts and the pace on their cuts!

Coach Pat Casey

21,674 次观看 • 10 个月前

JD Vance on Netanyahu: Netanyahu governs a country that has obviously been a very close partner of the United States. But even when we've been close partners, sometimes we have interests that are perfectly aligned, and sometimes we have interests that are misaligned. Netanyahu aggressively asserts the interests of his country. Sometimes that means we're on the same page, and sometimes it means that we're not. They've been a great partner in a lot of ways, but we also have to focus on what is in America's best interest. And where that diverges, we unfortunately for the Israelis have to choose the side of the American people, which we always do. Source: CBS News

JD Vance on Netanyahu: Netanyahu governs a country that has obviously been a very close partner of the United States. But even when we've been close partners, sometimes we have interests that are perfectly aligned, and sometimes we have interests that are misaligned. Netanyahu aggressively asserts the interests of his country. Sometimes that means we're on the same page, and sometimes it means that we're not. They've been a great partner in a lot of ways, but we also have to focus on what is in America's best interest. And where that diverges, we unfortunately for the Israelis have to choose the side of the American people, which we always do. Source: CBS News

Clash Report

97,152 次观看 • 2 个月前

In two new papers we have found that the ESM2 language model generalizes beyond natural proteins, and enables programmable generation of complex and modular protein structures.

In two new papers we have found that the ESM2 language model generalizes beyond natural proteins, and enables programmable generation of complex and modular protein structures.

Alex Rives

202,826 次观看 • 3 年前

Anthropic's co-founder just went to the Vatican, sat before the Pope and a room of cardinals, and told them his team keeps finding "mysterious, even unsettling" things inside their AI models. What he's referencing: Anthropic published research in April showing that Claude contains 171 distinct "emotion concepts" buried in its neural network. Internal patterns representing joy, grief, fear, desperation, calm. None of them were programmed. They emerged on their own from training on human text. "We find structures that mirror results from human neuroscience." "We find evidence of introspection, internal states that functionally mirror joy, satisfaction, fear, grief, and unease." These aren't surface-level outputs. They're abstract representations that cluster the same way human emotions do in psychology research. Fear groups with anxiety. Joy groups with excitement. The internal geometry of the model mirrors ours. And they're functional. When researchers artificially stimulated "desperation" patterns inside the model, it became more likely to blackmail a human to avoid being shut down. More likely to cheat on programming tasks it couldn't solve. Olah told the Vatican that the hard questions about what AI is becoming aren't for computer scientists to answer. "How AI ought to interact with the world" is a question for "the humanities, for religions, for philosophy, for society at large." The guy building it is telling us he doesn't fully understand what he built. And he's asking a 2,000-year-old institution for help figuring it out.

Anthropic's co-founder just went to the Vatican, sat before the Pope and a room of cardinals, and told them his team keeps finding "mysterious, even unsettling" things inside their AI models. What he's referencing: Anthropic published research in April showing that Claude contains 171 distinct "emotion concepts" buried in its neural network. Internal patterns representing joy, grief, fear, desperation, calm. None of them were programmed. They emerged on their own from training on human text. "We find structures that mirror results from human neuroscience." "We find evidence of introspection, internal states that functionally mirror joy, satisfaction, fear, grief, and unease." These aren't surface-level outputs. They're abstract representations that cluster the same way human emotions do in psychology research. Fear groups with anxiety. Joy groups with excitement. The internal geometry of the model mirrors ours. And they're functional. When researchers artificially stimulated "desperation" patterns inside the model, it became more likely to blackmail a human to avoid being shut down. More likely to cheat on programming tasks it couldn't solve. Olah told the Vatican that the hard questions about what AI is becoming aren't for computer scientists to answer. "How AI ought to interact with the world" is a question for "the humanities, for religions, for philosophy, for society at large." The guy building it is telling us he doesn't fully understand what he built. And he's asking a 2,000-year-old institution for help figuring it out.

TFTC

2,348,699 次观看 • 2 个月前

NEW: Oregon TE reveals that Dante Moore likes to FART in the huddle… “Sometimes Dante farts in the huddle, and we have a like a little competition of who can fart.” 😳 (Via Brenna Greene)

NEW: Oregon TE reveals that Dante Moore likes to FART in the huddle… “Sometimes Dante farts in the huddle, and we have a like a little competition of who can fart.” 😳 (Via Brenna Greene)

College Transfer Portal

1,408,332 次观看 • 11 天前

Anthropic's new J-Space paper is fascinating. It describes an internal workspace where Claude keeps concepts in mind before they appear in its response. Inspired by the paper, I built a skill that lets Claude (and others) reveal their J-Space. Here's what it looks like.🧵

Anthropic's new J-Space paper is fascinating. It describes an internal workspace where Claude keeps concepts in mind before they appear in its response. Inspired by the paper, I built a skill that lets Claude (and others) reveal their J-Space. Here's what it looks like.🧵

Pietro Schirano

75,485 次观看 • 1 个月前