Latest news with #YannLeCun
Yahoo
5 days ago
- Science
- Yahoo
Meta chief AI scientist Yann LeCun says current AI models lack 4 key human traits
Yann LeCun says there are four traits of human intelligence. Meta's chief AI scientist says AI lacks these traits, requiring a shift in training methods. Meta's V-JEPA is a non-generative AI model that aims to solve the problem. What do all intelligent beings have in common? Four things, according to Meta's chief AI scientist, Yann LeCun. At the AI Action Summit in Paris earlier this year, political leaders and AI experts gathered to discuss AI development. LeCun shared his baseline definition of intelligence with IBM's AI leader, Anthony Annunziata. "There's four essential characteristics of intelligent behavior that every animal, or relatively smart animal, can do, and certainly humans," he said. "Understanding the physical world, having persistent memory, being able to reason, and being able to plan, and planning complex actions, particularly planning hierarchically." LeCun said AI, especially large language models, have not hit this threshold, and incorporating these capabilities would require a shift in how they are trained. That's why many of the biggest tech companies are cobbling capabilities onto existing models in their race to dominate the AI game, he said. "For understanding the physical world, well, you train a separate vision system. And then you bolt it on the LLM. For memory, you know, you use RAG, or you bolt some associative memory on top of it, or you just make your model bigger," he said. RAG, which stands for retrieval augmented generation, is a way to enhance the outputs of large language models using external knowledge sources. It was developed at Meta. All those, however, are just "hacks," LeCun said. LeCun has spoken on several occasions about an alternative he calls world-based models. These are models trained on real-life scenarios and have higher levels of cognition than pattern-based AI. LeCun, in his chat with Annunziata, offered another definition. "You have some idea of the state of the world at time T, you imagine an action it might take, the world model predicts what the state of the world is going to be from the action you took," he said. But, he said, the world evolves according to an infinite and unpredictable set of possibilities, and the only way to train for them is through abstraction. Meta is already experimenting with this through V-JEPA, a model it released to the public in February. Meta describes it as a non-generative model that learns by predicting missing or masked parts of a video. "The basic idea is that you don't predict at the pixel level. You train a system to run an abstract representation of the video so that you can make predictions in that abstract representation, and hopefully this representation will eliminate all the details that cannot be predicted," he said. The concept is similar to how chemists established a fundamental hierarchy for the building blocks of matter. "We created abstractions. Particles, on top of this, atoms, on top of this, molecules, on top of this, materials," he said. "Every time we go up one layer, we eliminate a lot of information about the layers below that are irrelevant for the type of task we're interested in doing." That, in essence, is another way of saying we've learned to make sense of the physical world by creating hierarchies. Read the original article on Business Insider
Business Insider
6 days ago
- Science
- Business Insider
Meta chief AI scientist Yann LeCun says current AI models lack 4 key human traits
What do all intelligent beings have in common? Four things, according to Meta's chief AI scientist, Yann LeCun. At the AI Action Summit in Paris earlier this year, political leaders and AI experts gathered to discuss AI development. LeCun shared his baseline definition of intelligence with IBM's AI leader, Anthony Annunziata. "There's four essential characteristics of intelligent behavior that every animal, or relatively smart animal, can do, and certainly humans," he said. "Understanding the physical world, having persistent memory, being able to reason, and being able to plan, and planning complex actions, particularly planning hierarchically." LeCun said AI, especially large language models, have not hit this threshold, and incorporating these capabilities would require a shift in how they are trained. That's why many of the biggest tech companies are cobbling capabilities onto existing models in their race to dominate the AI game, he said. "For understanding the physical world, well, you train a separate vision system. And then you bolt it on the LLM. For memory, you know, you use RAG, or you bolt some associative memory on top of it, or you just make your model bigger," he said. RAG, which stands for retrieval augmented generation, is a way to enhance the outputs of large language models using external knowledge sources. It was developed at Meta. All those, however, are just "hacks," LeCun said. LeCun has spoken on several occasions about an alternative he calls world-based models. These are models trained on real-life scenarios and have higher levels of cognition than pattern-based AI. LeCun, in his chat with Annunziata, offered another definition. "You have some idea of the state of the world at time T, you imagine an action it might take, the world model predicts what the state of the world is going to be from the action you took," he said. But, he said, the world evolves according to an infinite and unpredictable set of possibilities, and the only way to train for them is through abstraction. Meta is already experimenting with this through V-JEPA, a model it released to the public in February. Meta describes it as a non-generative model that learns by predicting missing or masked parts of a video. "The basic idea is that you don't predict at the pixel level. You train a system to run an abstract representation of the video so that you can make predictions in that abstract representation, and hopefully this representation will eliminate all the details that cannot be predicted," he said. The concept is similar to how chemists established a fundamental hierarchy for the building blocks of matter. "We created abstractions. Particles, on top of this, atoms, on top of this, molecules, on top of this, materials," he said. "Every time we go up one layer, we eliminate a lot of information about the layers below that are irrelevant for the type of task we're interested in doing." That, in essence, is another way of saying we've learned to make sense of the physical world by creating hierarchies.
Yahoo
14-05-2025
- Business
- Yahoo
Samaya AI, startup building AI for financial services, raises $43.5 million in VC funding
Samaya AI, a startup that creates AI models that assist financial analysts and which is backed by a number of leading figures both in Silicon Valley and on Wall Street, has raised $43.5 in new venture capital financing. Venture capital firm New Enterprise Associates led the funding round. The valuation of Samaya following the new funding was not disclosed. Also participating in the new investment round were former Google CEO turned prominent Silicon Valley investor Eric Schmidt; AI 'godfather' Yann LeCun, who is Meta's chief AI scientist; David Siegel, who is a confounder of the hedge fund Two Sigma; and Marty Chavez, the former Goldman Sachs technology executive who is now vice chair of the investment firm Sixth Street Partners. The company was founded in 2022 by AI researchers who were working at leading AI labs, including Google Brain, Meta's Fundamental AI Research lab, Amazon Web Services, and the Allen Institute for Artificial Intelligence. Maithra Raghu, Samaya's co-founder and CEO, said that she and her co-founders were seeing the development of generative AI within these labs and saw obvious applications in financial services. But at the same time, she says, the founders believed that rather than developing broad, general-purpose large language models (LLMs) that were designed to perform any task, better quality results could be obtained from creating products that would be specialized for particular domains. 'We deeply believed back then that expert intelligence emerges from specialization,' she said. 'It is hard to hit the level of quality and reliability [financial firms require] without specialization.' The company's first product is a tool that can conduct financial research and analysis. It can both look across the web for high quality sources of financial data—such as SEC filings—or be attached to a firm's own knowledge base and data sources and use those to find information. The system can be used to find comparable companies and compare their financial valuation and performance—a task financial analysts often undertake. It can also be used to help with due diligence on potential investments. It is already being deployed at Morgan Stanley's International Securities Group and at a number of hedge funds. Katy Huberty, global director of research at Morgan Stanley, said that Samaya is creating 'actionable insights from both our extensive Research library as well as external sources, enhancing our ability to provide world class analysis to our clients.' Today Samaya also announced the debut of a new AI agent it calls Causal World Models. The system excels at modeling economic systems. In a trial project, Samaya used the new AI software to model the effect of the Trump administration's proposed tariffs on the entire U.S. economy. The system produced a complex flow diagram highlighting the interaction between various economic sectors and providing both quantitative and qualitative analysis. Raghu said a preview version of Causal World Models had proved popular with Samaya's customers, many of which have used it to model the effects of Trump's tariffs on different companies and industry sectors. But she said that the system could be used to answer all kinds of different economic questions too. While past LLMs have been good at finding correlations, they have often failed to understand cause and effect. But Raghu said that Causal World Models uses a multi-stage process to build a graph that maps cause and effect and then can reason about the question it is trying to answer using that graph. Samaya's previous analysis tool can produce outputs that are in some ways similar to what one might get from one of the so-called 'deep research' tools that have been debuted by OpenAI, Google DeepMind, Anthropic, or Perplexity—with the difference being that Samaya AI's products seem to have greater fluency with financial information and ability to perform financial analysis accurately. Samaya also has tools that can produce results in specific formats, beyond just research reports, that are common in financial services—such as PowerPoint decks and Excel spreadsheets. Under the hood, Raghu said that its first research tool is engineered differently than the giant LLMs that power OpenAI's ChatGPT or Anthropic's Claude. Samaya uses what Raghu calls a 'a lattice of experts architecture' that includes many smaller language models, each taking on a portion of a research tasks, and with some of the models used to check the output of others. She said that this approach greatly reduces the chances of hallucination—where an AI model invents information. Tiffany Luck, partner at NEA, said Samaya fits her investment firm's thesis that specific AI tools will need to be designed for specific industry verticals. She also said that accuracy was extremely important in financial services. '90% accuracy is not good enough,' she said. NEA liked that Samaya could provide that kind of accuracy and could serve different kinds of financial services users, with both junior analysts and senior leaders within financial services firms getting benefit out of Samaya's tools. 'This is redefining how AI can partner with financial services,' Schmidt said in a statement. Samaya is not the only company targeting the financial services sector. Morgan Stanley has also been using OpenAI's models. JPMorgan has a large team of AI researchers developing tools for use within the bank. Financial data and news giant Bloomberg has been building AI models designed specifically for financial data and financial service firms. And a number of other startups are also competing with Samaya in building AI tools for financial research, including Model ML, V7 Labs, and Rogo. This story was originally featured on Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
Business Wire
14-05-2025
- Business
- Business Wire
Samaya AI Announces $43.5m in Funding Led by NEA, Launches New AI Agent for Financial Services
MOUNTAIN VIEW, Calif.--(BUSINESS WIRE)-- Samaya AI, the expert artificial intelligence platform for financial services, today announced $43.5 million in financing, led by NEA (New Enterprise Associates), with participation from leaders in technology and financial services, including Eric Schmidt (former CEO of Google), Yann LeCun (AI Turing Award winner), David Siegel (Co-founder Two Sigma), and Marty Chavez (Vice Chair Sixth Street), among others. The funds will be used to support Samaya's product development and market expansion. Samaya develops a suite of expert AI agents designed and trained for complex financial workflows, from investment research to client advisory and deal diligence. For example, Samaya's agents can autonomously synthesize a sector-wide investment report, create an investment presentation by reasoning over proprietary documents and provide instant answers to complex questions over millions of real-time sources — all while carefully grounding the output with cited evidence. Samaya has seen incredible market momentum with 100% month-over-month growth in usage, and counts premier financial institutions such as Morgan Stanley as customers. 'With Samaya, customers can create a personalized team of AI knowledge agents that can 1000x the output of a single analyst,' said Maithra Raghu, founder and CEO of Samaya AI. 'Our users access agentic experiences that reliably deliver insights with expert level quality and no hallucinations. Samaya's proprietary AI is designed for factuality over fluency and trained for financial expertise over generic 'internet user' responses. Customers can use Samaya for such tasks as distilling precise investment alpha from huge volumes of real-time information, getting instantaneous portfolio-specific insights, or creating investment memos by leveraging internal and external data.' With the funding, Samaya is also launching its latest AI agent, Causal World Models. Previously in research preview, Causal World Models can autonomously model the entire economy and use this to carry out multi-stage grounded reasoning and provide quantitative predictions for economy-wide questions. For example, the question, 'What is the impact of the tariffs on US GDP and the economy?' generates an interactive diagram, with the AI agent providing qualitative factors and citing predictions and quantitative projections. 'Samaya is revolutionizing decision making for financial institutions with an AI platform that transforms how work gets done and supercharges research and analysis,' said Tiffany Luck, Partner at NEA. 'We are thrilled to partner with Samaya as they build the leading AI for financial services, and we believe that Samaya has the potential to radically disrupt how all companies approach knowledge work.' 'Samaya AI is reinventing how financial experts interact with information. Their AI agents reason through massive volumes of real-time data to deliver high-precision insights, streamlining workflows like synthesizing market-wide trends, identifying comps for deal due diligence, and generating detailed sector reports in ways generic AI assistants can't. This is redefining how AI can partner with financial services,' said Eric Schmidt, former CEO of Google. 'We're excited to partner with Samaya across all divisions of our Institutional Securities Group, and leverage Samaya's cutting-edge AI to power the creation of actionable insights from both our extensive Research library as well as external sources, enhancing our ability to provide world class analysis to our clients,' said Katy Huberty, Global Director of Research, Morgan Stanley. About Samaya AI Samaya AI is the expert artificial intelligence platform for financial services, enabling customers to create personalized teams of AI agents that supercharge financial research, analysis and decision making. Founded in 2022, Samaya AI's platform is custom trained for research and analysis of information dense, real-time financial sources. Prior to Samaya, the founding team held positions at Google Brain, Meta AI, AWS and Stanford. For more information, visit About NEA New Enterprise Associates, Inc. (NEA) is a global venture capital firm focused on helping entrepreneurs build transformational businesses across multiple stages, sectors and geographies. Founded in 1977, NEA has more than $27 billion in assets under management as of December 31, 2024, and invests in technology and healthcare companies at all stages in a company's lifecycle, from seed stage through IPO. The firm's long track record of investing includes more than 280 portfolio company IPOs and more than 470 mergers and acquisitions. For more information, please visit
Bloomberg
08-05-2025
- Business
- Bloomberg
Meta Taps New Head of AI Lab After Staffer's Return From Google
Meta Platforms Inc. told staff that it's chosen Robert Fergus to helm its artificial intelligence research lab, elevating an employee who helped start the lab before a stint at competitor Alphabet Inc. Fergus co-founded the Facebook AI Research lab, known as FAIR, in 2014 with Yann LeCun, who currently serves as the chief AI scientist at Meta. The unit is tasked with Meta's longer-term AI research, including helping create models that help advance robotics, generate audio, understand images and push the boundaries of AI capabilities.
