Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Using SocialReasoning Bench, we observed a stable pattern across models—agents execute competently, but fail to consistently improve the user’s position, even with explicit instructions to optimize for user interest.

Microsoft Research

554,455 subscribers

12,355 views • 2 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

If someone from the future gave us a trillion tokens of agents competently planning, executing, and adapting, we'd likely turn today's models into highly capable agents overnight. But we don't currently have that data. We at Mechanize are working to solve that.

If someone from the future gave us a trillion tokens of agents competently planning, executing, and adapting, we'd likely turn today's models into highly capable agents overnight. But we don't currently have that data. We at Mechanize are working to solve that.

Mechanize

27,890 views • 11 months ago

Earlier this year we announced x402. x402 enables AI agents to pay for content and services as well as execute transactions using stablecoins. In the last 30 days we’ve seen over $200M in annualized transactions across both humans and AI agents. 📈

Earlier this year we announced x402. x402 enables AI agents to pay for content and services as well as execute transactions using stablecoins. In the last 30 days we’ve seen over $200M in annualized transactions across both humans and AI agents. 📈

Coinbase Developer Platform🛡️

75,254 views • 7 months ago

For an upcoming project, we combined spreads related to the mouse position but also to the blocs position when the user scrolls. Love the fluidity! Currently in the making at TuxCreativeHouse with Louis Paquet

For an upcoming project, we combined spreads related to the mouse position but also to the blocs position when the user scrolls. Love the fluidity! Currently in the making at TuxCreativeHouse with Louis Paquet

Michaël Garcia

22,978 views • 3 years ago

"Our wind industry is a European success story. With a European Wind Power package, we will fast-track permitting even more and improve the auction systems across the EU, focusing on skills, access to finance and stable supply chains." — President Ursula von der Leyen #SOTEU 2023

"Our wind industry is a European success story. With a European Wind Power package, we will fast-track permitting even more and improve the auction systems across the EU, focusing on skills, access to finance and stable supply chains." — President Ursula von der Leyen #SOTEU 2023

European Commission

42,263 views • 2 years ago

Manual trading is the slowest part of your strategy. Not another bot, but a control panel to build your own trading agents & trade across DEXs from one terminal. Build agents, manage positions, and control TP/SL - all in Oraichain Quant Terminal. You set the rules and risk. The agent executes using quant models - to find entries, select orders, and run 24/7. Miss fewer moves. Think less. Execute better. 👉

Manual trading is the slowest part of your strategy. Not another bot, but a control panel to build your own trading agents & trade across DEXs from one terminal. Build agents, manage positions, and control TP/SL - all in Oraichain Quant Terminal. You set the rules and risk. The agent executes using quant models - to find entries, select orders, and run 24/7. Miss fewer moves. Think less. Execute better. 👉

Oraichain

88,491 views • 6 months ago

Introducing Collaborative Reasoner: a framework to improve collaborative reasoning in language models. Collaborative Reasoner paves the way for developing social agents that can partner with humans and other agents. Read the research paper and download the code.

Introducing Collaborative Reasoner: a framework to improve collaborative reasoning in language models. Collaborative Reasoner paves the way for developing social agents that can partner with humans and other agents. Read the research paper and download the code.

AI at Meta

58,510 views • 1 year ago

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 views • 1 year ago

Learning reliable manipulation skills for robots is probably the single most important thing for the field to accomplish. Generalist models are usually not reliable enough to achieve the ~100% reliability we really want. But Wenli Xiao has a recipe for achieving this, and for using the data to improve the base model too. Learn more here in the RoboPapers podcast with Michael Cho - Rbt/Acc ->

Learning reliable manipulation skills for robots is probably the single most important thing for the field to accomplish. Generalist models are usually not reliable enough to achieve the ~100% reliability we really want. But Wenli Xiao has a recipe for achieving this, and for using the data to improve the base model too. Learn more here in the RoboPapers podcast with Michael Cho - Rbt/Acc ->

Chris Paxton

35,602 views • 7 months ago

When data agents fail, they often fail silently - giving confident-sounding answers that are wrong, and it can be hard to figure out what caused the failure. "Building and Evaluating Data Agents" is a new short course created with Snowflake and taught by Anupam Datta and Josh Reini that teaches you to build data agents with comprehensive evaluation built in. Skills you'll gain: - Build reliable LLM data agents using the Goal-Plan-Action framework and runtime evaluations that catch failures mid-execution - Use OpenTelemetry tracing and evaluation infrastructure to diagnose exactly where agents fail and systematically improve performance - Orchestrate multi-step workflows across web search, SQL, and document retrieval in LangGraph-based agents The result: visibility into every step of your agent's reasoning, so if something breaks, you have a systematic approach to fix it. Sign up to get started:

When data agents fail, they often fail silently - giving confident-sounding answers that are wrong, and it can be hard to figure out what caused the failure. "Building and Evaluating Data Agents" is a new short course created with Snowflake and taught by Anupam Datta and Josh Reini that teaches you to build data agents with comprehensive evaluation built in. Skills you'll gain: - Build reliable LLM data agents using the Goal-Plan-Action framework and runtime evaluations that catch failures mid-execution - Use OpenTelemetry tracing and evaluation infrastructure to diagnose exactly where agents fail and systematically improve performance - Orchestrate multi-step workflows across web search, SQL, and document retrieval in LangGraph-based agents The result: visibility into every step of your agent's reasoning, so if something breaks, you have a systematic approach to fix it. Sign up to get started:

Andrew Ng

101,930 views • 9 months ago

Introducing LobeHub: Agent teammates that grow with you. LobeHub is the ultimate space for work and life: to find, build, and collaborate with agent teammates that grow with you. We’re building the world’s first and largest human–agent co-evolving network. Two years ago, we built LobeChat, an open-source interface for using different AI models. Today, LobeChat has 70k+ GitHub stars and serves 6M+ users worldwide. How to fully unlock the power of models has always been a shared mission between us and the community. We started with interaction — a fundamentally new, agent-first experience. Agents are no longer passive tools invoked in a single conversation. They should be proactive, always-on units of work. Treating agents as the minimal atomic unit is also the core of our agent harness infra. Today’s agents are mostly one-off executors. Even with memory, it’s often global — and hallucinates. We build long-term agent teammates that evolve with users. Each agent has its own dedicated memory space, editable by users, allowing humans and agents to co-evolve over time. This, in turn, allows us to design clearer rewards for reinforcement learning and create cleaner environments for continual learning. Agent teammates can work in groups. Through a multi-agent system, agent groups operate faster, more cost-effective, and go beyond what single-agent systems can achieve. For example, a single agent often requires heavy user involvement to proceed step by step, whereas LobeHub can execute the same work from a single instruction, with a supervisor orchestrating agents that run in parallel or debate to produce better results. We are building the collaboration network among agent teammates — and between humans and agent teammates as well. Ease of use matters. AI intelligence and shared human intelligence are equally important. With simple instructions and tool selection, you can effortlessly build and team up with agent coworkers to deliver complex, systematic work — even assembling a quant team to execute trades. Through the LobeHub community, anyone can discover, reuse, and remix agents and agent groups, customizing them to fit their own workflows, preferences, and needs. Last but not least, our vision started with LobeChat: multi-model support is the most efficient approach for users. We believe different models excel in different scenarios. By routing across multiple models, LobeHub improves cost efficiency and unlocks capabilities that a single-model setup cannot easily support.

Introducing LobeHub: Agent teammates that grow with you. LobeHub is the ultimate space for work and life: to find, build, and collaborate with agent teammates that grow with you. We’re building the world’s first and largest human–agent co-evolving network. Two years ago, we built LobeChat, an open-source interface for using different AI models. Today, LobeChat has 70k+ GitHub stars and serves 6M+ users worldwide. How to fully unlock the power of models has always been a shared mission between us and the community. We started with interaction — a fundamentally new, agent-first experience. Agents are no longer passive tools invoked in a single conversation. They should be proactive, always-on units of work. Treating agents as the minimal atomic unit is also the core of our agent harness infra. Today’s agents are mostly one-off executors. Even with memory, it’s often global — and hallucinates. We build long-term agent teammates that evolve with users. Each agent has its own dedicated memory space, editable by users, allowing humans and agents to co-evolve over time. This, in turn, allows us to design clearer rewards for reinforcement learning and create cleaner environments for continual learning. Agent teammates can work in groups. Through a multi-agent system, agent groups operate faster, more cost-effective, and go beyond what single-agent systems can achieve. For example, a single agent often requires heavy user involvement to proceed step by step, whereas LobeHub can execute the same work from a single instruction, with a supervisor orchestrating agents that run in parallel or debate to produce better results. We are building the collaboration network among agent teammates — and between humans and agent teammates as well. Ease of use matters. AI intelligence and shared human intelligence are equally important. With simple instructions and tool selection, you can effortlessly build and team up with agent coworkers to deliver complex, systematic work — even assembling a quant team to execute trades. Through the LobeHub community, anyone can discover, reuse, and remix agents and agent groups, customizing them to fit their own workflows, preferences, and needs. Last but not least, our vision started with LobeChat: multi-model support is the most efficient approach for users. We believe different models excel in different scenarios. By routing across multiple models, LobeHub improves cost efficiency and unlocks capabilities that a single-model setup cannot easily support.

LobeHub

185,161 views • 5 months ago

90% of AI products fail in the first month. It's not bad models but bad prompts. Adaline AI is a platform to create prompts, test across models and modalities, integrate tools, evaluate performance, and deploy with real-time monitoring. Find the best prompt for your LLMs...👇

90% of AI products fail in the first month. It's not bad models but bad prompts. Adaline AI is a platform to create prompts, test across models and modalities, integrate tools, evaluate performance, and deploy with real-time monitoring. Find the best prompt for your LLMs...👇

Akshay 🚀

46,821 views • 1 year ago

John Allan Namu: AI is a powerful tool but if you use it badly then you will get a bad outcome despite the power of the tool. Now that we are deploying this model across the most critical sectors in our economy, if we fail to get a richer understanding of the tool and fail to interrogate the intent with which we are using this tool, then we are going to end up in the situation that are in now #citizenexplainer

John Allan Namu: AI is a powerful tool but if you use it badly then you will get a bad outcome despite the power of the tool. Now that we are deploying this model across the most critical sectors in our economy, if we fail to get a richer understanding of the tool and fail to interrogate the intent with which we are using this tool, then we are going to end up in the situation that are in now #citizenexplainer

Citizen TV Kenya

11,775 views • 2 months ago

Hoping your coding agents could understand you and adapt to your preferences? Meet TOM-SWE, our new framework for coding agents that don’t just write code, but model the user's mind persistently (ranging from general preferences to small details) arxiv: ❓Motivation: Most coding agents today can plan, edit, run, and test code. But they still fail at a key part of real-world development, understanding the user! Underspecified, shifting, or context-dependent instructions can easily break them. You must have those moments when coding agents were running for 10 minutes and ended up producing things largely misaligned. (1/)

Hoping your coding agents could understand you and adapt to your preferences? Meet TOM-SWE, our new framework for coding agents that don’t just write code, but model the user's mind persistently (ranging from general preferences to small details) arxiv: ❓Motivation: Most coding agents today can plan, edit, run, and test code. But they still fail at a key part of real-world development, understanding the user! Underspecified, shifting, or context-dependent instructions can easily break them. You must have those moments when coding agents were running for 10 minutes and ended up producing things largely misaligned. (1/)

Xuhui Zhou

40,135 views • 8 months ago

Vision-language models can control robots, but what if the prompt is too complex for the robot to follow directly? We developed a way to get robots to “think through” complex instructions, feedback, and interjections. We call it the Hierarchical Interactive Robot (Hi Robot).

Vision-language models can control robots, but what if the prompt is too complex for the robot to follow directly? We developed a way to get robots to “think through” complex instructions, feedback, and interjections. We call it the Hierarchical Interactive Robot (Hi Robot).

Physical Intelligence

116,845 views • 1 year ago

Still manually degening across tabs, bridging to stake, and praying to the chart gods? 📊 Let Autonomous Agents Earn While You Sleep Figment Trade lets anyone launch fully autonomous DeFi agents using just natural language, no coding needed. Your agents can: ⚡️ Trade perpetuals with up to 100x leverage 💹 Optimize yield across protocols ⛓️ Run 24/7, fully on-chain Just describe a strategy and personality. Figment handles the rest. Touch grass while your agent trades. This is vibe trading. 👉

Still manually degening across tabs, bridging to stake, and praying to the chart gods? 📊 Let Autonomous Agents Earn While You Sleep Figment Trade lets anyone launch fully autonomous DeFi agents using just natural language, no coding needed. Your agents can: ⚡️ Trade perpetuals with up to 100x leverage 💹 Optimize yield across protocols ⛓️ Run 24/7, fully on-chain Just describe a strategy and personality. Figment handles the rest. Touch grass while your agent trades. This is vibe trading. 👉

Figment.Trade

140,843 views • 1 year ago

Was this one of the 274 FBI agents who were embedded on Jan 6th? He was parading around the capitol grounds with a pyramid on his head, calling for the crowd to "execute traitors", but wouldn't identify himself, even when confronted by patriots in the crowd.

Was this one of the 274 FBI agents who were embedded on Jan 6th? He was parading around the capitol grounds with a pyramid on his head, calling for the crowd to "execute traitors", but wouldn't identify himself, even when confronted by patriots in the crowd.

Died Suddenly

13,456 views • 9 months ago

.Eric Jang tried using transformers for his Go bot, but they couldn't beat ResNets. The reason gets at something general about architectures. ResNets are biased towards the local. Nearby things matter more, and a useful pattern in one place is a useful pattern anywhere. Transformers are biased the other way, towards global context, with every position able to attend to every other. Most Go fighting is local, and a useful local pattern learned in one position can be applied anywhere in the board. A ResNet's inductive bias means it gets these insights about Go for free. But a transformer has to pay for them.

.Eric Jang tried using transformers for his Go bot, but they couldn't beat ResNets. The reason gets at something general about architectures. ResNets are biased towards the local. Nearby things matter more, and a useful pattern in one place is a useful pattern anywhere. Transformers are biased the other way, towards global context, with every position able to attend to every other. Most Go fighting is local, and a useful local pattern learned in one position can be applied anywhere in the board. A ResNet's inductive bias means it gets these insights about Go for free. But a transformer has to pay for them.

Dwarkesh Patel

32,615 views • 1 month ago

🚨 AI Can Build Complex Apps With Just One Single Prompt New AI models may be banned but multi-LLM agents can build insane things.... Here is our agent building an entire Slack-like app using a single prompt! Combines Opus, GPT 5.5 and open-source LLMs to optimize cost and performance

🚨 AI Can Build Complex Apps With Just One Single Prompt New AI models may be banned but multi-LLM agents can build insane things.... Here is our agent building an entire Slack-like app using a single prompt! Combines Opus, GPT 5.5 and open-source LLMs to optimize cost and performance

Bindu Reddy

6,663,574 views • 23 days ago

You can see from the nuance that Zuck thinks we are in an AI CapEx bubble. But since this is a race & $META is one of the most financially stable ones in this race, they are willing to push the race further for competitors to fail, even if it means $META overspend a few $100B.

You can see from the nuance that Zuck thinks we are in an AI CapEx bubble. But since this is a race & $META is one of the most financially stable ones in this race, they are willing to push the race further for competitors to fail, even if it means $META overspend a few $100B.

Rihard Jarc

354,459 views • 10 months ago

Sachi Takahara / 0699.eth sits down with David Minarsch ☴, core contributor to OLAS, to break down what it really means to co-own AI. Instead of just using AI, OLAS lets users participate — through tokens, running agents, or even operating them as businesses. From DeFi to prediction markets, agents generate yield, and operators share in the upside. It’s a shift from passive users to active participants in an AI economy. But ownership is only half the story — alignment matters. Through staking contracts, OLAS defines clear KPIs for agents, similar to roles in a workplace. Agents act autonomously, but incentives keep them aligned with the system’s goals. And at the infrastructure level, reliability is handled through decentralized agent architecture — combining multiple models and consensus mechanisms to reduce errors and improve decision-making. The result: AI that isn’t just decentralized — but coordinated, incentivized, and owned by the community.

Sachi Takahara / 0699.eth sits down with David Minarsch ☴, core contributor to OLAS, to break down what it really means to co-own AI. Instead of just using AI, OLAS lets users participate — through tokens, running agents, or even operating them as businesses. From DeFi to prediction markets, agents generate yield, and operators share in the upside. It’s a shift from passive users to active participants in an AI economy. But ownership is only half the story — alignment matters. Through staking contracts, OLAS defines clear KPIs for agents, similar to roles in a workplace. Agents act autonomously, but incentives keep them aligned with the system’s goals. And at the infrastructure level, reliability is handled through decentralized agent architecture — combining multiple models and consensus mechanisms to reduce errors and improve decision-making. The result: AI that isn’t just decentralized — but coordinated, incentivized, and owned by the community.

BlockLayer Podcast

38,548 views • 2 months ago