FT: главный исследователь Meta по ИИ покинет корпорацию ради своего стартапа

cryptonews.ruPublicado a 2025-09-11Actualizado a 2025-11-12

  • Лауреат премии Тьюринга и руководитель FAIR Ян Лекун планирует покинуть Meta в ближайшие месяцы.
  • Он создает стартап, посвященный разработке «моделей мира» — нового направления в ИИ.
  • Уход усиливает кризис в стратегии Meta, которая перестраивает свои ИИ-подразделения.

Корпорация Meta может лишиться одного из ключевых архитекторов своей стратегии искусственного интеллекта. По данным Financial Times, главный специалист по ИИ компании Янн Лекун намерен уйти, чтобы основать собственный стартап.

Источники издания утверждают, что он уже ведет переговоры с инвесторами и планирует запустить проект в течение ближайших месяцев.

Лекун является профессором Нью-Йоркского университета и лауреатом премии имени Тьюринга. Он занимает должность старшего научного сотрудника Meta и возглавляет исследовательское подразделение Fundamental AI Research (FAIR).

Будущий стартап специалиста, как сообщается, сосредоточится на разработке «моделей мира». Это системы, которые формируют внутреннее представление об окружающей среде и способны прогнозировать причинно-следственные связи.

Подобными исследованиями также занимаются Google DeepMind и World Labs.

Потенциальный уход Лекуна стал символом напряженности внутри Meta, где в последние месяцы происходит масштабная перестройка ИИ-направления. Руководство компании стремится сократить отставание от OpenAI, Google и Anthropic, создав новое подразделение Meta Superintelligence Labs (MSL). В него пригласили более 50 инженеров и исследователей из других компаний.

В июне Meta инвестировала $14,3 млрд в компанию Scale AI и назначила ее генерального директора Александра Вана во главе MSL. Однако перестановки лишь усилили внутренние конфликты, считают в FT.

Новые сотрудники якобы недовольны корпоративной бюрократией, а старая команда генеративного ИИ столкнулась с ограничениями в доступе к ресурсам.

На фоне этих перемен исследовательская работа FAIR, ориентированная на долгосрочные научные цели, отошла на второй план. Провал семейства моделей Llama 4 показал, что Meta уступает лидерам индустрии, подчеркнули эксперты.

Решение CEO компании Марка Цукерберга ускорить прикладные разработки фактически ослабило фундаментальные исследования, которыми занимался Лекун, говорится в материале. На момент подготовки материала сообщения о возможном уходе специалиста из корпорации представители Meta не комментировали.

Сам ученый неоднократно выражал скепсис в отношении современных больших языковых моделей. Он утверждает, что индустрия преувеличивает их интеллектуальные возможности.

«Нам бы хотя бы создать систему умнее домашней кошки, прежде чем говорить о сверхразуме», — писал Лекун.

Напомним, мы писали, что Цукерберг заявил о планах Meta создать «персональный суперинтеллект».

Lecturas Relacionadas

AI Boosting Efficiency and Cutting Costs Makes VC Increasingly Expensive

"AI for Cost Reduction Makes VC Funding More Expensive" Despite the "cost-reduction and efficiency" narrative of AI, venture capital (VC) investment in the AI sector is becoming increasingly costly. While AI tools lower the initial costs for many startups—with team sizes shrinking across funding stages—the market is polarizing. For top-tier AI teams, especially those from leading companies like Google and OpenAI, funding rounds are now larger and valuations are higher than ever at the seed and early stages. For example, new ventures by prominent researchers are securing billions in funding with valuations reaching tens of billions before having a mature product. This creates a "barbell" market: lightweight startups need less capital, while elite AI firms attract massive investments early on. This dynamic raises the cost for VCs to acquire and maintain meaningful ownership stakes. As valuations soar early, securing the same equity percentage requires significantly larger capital commitments. VCs must now invest more upfront and reserve substantial funds for follow-on rounds to avoid dilution, prompting large firms like Accel and a16z to raise massive new funds. Consequently, capital is concentrating intensely in a few perceived winners like OpenAI and Anthropic, widening the gap between large and small VC funds. While high valuations bake in future growth expectations, they also compress potential returns, demanding that portfolio companies achieve unprecedented scale. For major VCs, the core strategy is clear: secure early positions in potential winners and maintain the capital to keep investing as valuations rapidly escalate.

marsbitHace 21 min(s)

AI Boosting Efficiency and Cutting Costs Makes VC Increasingly Expensive

marsbitHace 21 min(s)

An Eight-Year Investment Takes a Sharp Turn: Why Did Ethereum Suddenly Abandon Poseidon?

On August 13, Ethereum researcher Justin Drake announced a significant shift in Ethereum's Layer-1 cryptographic roadmap: abandoning the SNARK-friendly hash function Poseidon in favor of traditional functions like SHA2 or BLAKE2. This decision ends eight years of research and investment, marking a major revision to the post-quantum security strategy. Poseidon, introduced in 2019, was highly efficient for zkRollups and zkVMs within SNARK circuits. However, its need for prolonged cryptanalysis and the pressing timeline for quantum resistance revealed limitations. Recent breakthroughs in SNARK design, specifically using binary fields, now enable traditional, battle-tested hash functions to perform as efficiently as Poseidon within SNARKs. Benchmarks show modern laptops can now verify over a million traditional hash calls per second. This change is partly driven by accelerated concerns over quantum computing threats. Reports warn that "Cryptographically Relevant Quantum Computers" could break current blockchain signatures like ECDSA by the early 2030s, risking trillions in assets. Ethereum's response focuses on hash-based post-quantum signature schemes, deemed more quantum-resistant than some lattice-based alternatives under pressure from AI cryptanalysis. Ethereum's updated post-quantum roadmap targets a production-ready "leanVM" for signature aggregation by 2027, with full deployment across consensus, execution, and data layers by 2028. The shift to mature hash functions like SHA2 reduces reliance on newer algorithms and aligns with the goal of using widely analyzed cryptographic primitives. Other major blockchains are also preparing. Solana's core teams have independently chosen the NIST-standardized Falcon signature scheme for its compact size. Starknet plans a phased migration, starting with replacing its Pedersen hash with BLAKE2. Ethereum's move signifies a strategic pivot towards proven security foundations for the quantum era.

marsbitHace 1 hora(s)

An Eight-Year Investment Takes a Sharp Turn: Why Did Ethereum Suddenly Abandon Poseidon?

marsbitHace 1 hora(s)

Programmers Worldwide Are Wasting Money on Anthropic! The Company Can't Stand It Anymore

Anthropic recently published guidelines to help developers using Claude Code reduce unnecessary token costs. The key recommendations include: 1) Clear (/clear) conversations after completing a task to avoid carrying irrelevant file reads and command outputs into the next task. 2) Set the model and reasoning effort level at the start of a session, as switching mid-session invalidates the prompt cache, requiring a full-price recalculation of the entire dialog history. 3) Attach files using @ references instead of typing paths manually to avoid extra tool calls and searches that bloat the context. 4) Add quiet flags to verbose commands (e.g., in CLAUDE.md) to minimize lengthy output in the dialog history. 5) Use /compact while the session cache is still warm (before breaks) to compress the dialog at one-tenth the cost. 6) Offload large-output tasks to a sub-agent, which runs in an isolated context and only returns conclusions, preventing intermediate outputs from polluting the main dialog. The article explains token pricing: input tokens (prefill) are processed in parallel, while output tokens (decode) are generated serially, making output tokens five times more expensive. Caching is crucial for savings—if a request's prefix (system prompt, CLAUDE.md, dialog history) matches the previous one byte-for-byte, reading it costs only 10% of the standard input price. However, cache invalidation occurs when changing models, effort levels, fast mode, compressing dialogs, after cache expiration, or when resuming old sessions. Dialog history also grows quadratically (O(n²)) as file contents and command outputs accumulate, increasing costs per round. Proactive context management—like isolating noisy tasks, using /rewind to trim unproductive turns, and task-based session clearing—is becoming an essential skill for cost-effective AI-assisted development.

marsbitHace 3 hora(s)

Programmers Worldwide Are Wasting Money on Anthropic! The Company Can't Stand It Anymore

marsbitHace 3 hora(s)

Pax Silica vs. WAICO: The US Wants to Prohibit Europe from Using Chinese Artificial Intelligence

The United States is preparing to demand that European and other partners abandon Chinese artificial intelligence initiatives, threatening exclusion from the American-led "Pax Silica" coalition, according to a leaked U.S. State Department document. This ultimatum forces signatories of the "AI Opportunity Statement" to choose between the Western technological ecosystem and alternative frameworks, with China not explicitly named but clearly targeted. Pax Silica is a U.S. strategy for AI and semiconductor hegemony, launched in late 2025. Its European presence expanded significantly in mid-2026. Concurrently, China, Russia, and 27 other nations established the World AI Cooperation Organization (WAICO) in July 2026 as an independent intergovernmental platform promoting AI governance based on UN principles. This situation creates a difficult choice, especially for European nations balancing strategic autonomy with dependence on U.S. tech and security. It also pressures Global South countries with pragmatic ties to both Washington and Beijing. The formation of competing blocks risks fragmenting the global tech landscape, forcing companies to split supply chains, increasing costs, and potentially leading to incompatible standards and protocols. The era of open globalization in AI may be ending, replaced by geopolitical confrontation where technological sovereignty trumps economic efficiency. The decisions made will shape the global digital economy for decades.

cryptonews.ruHace 5 hora(s)

Pax Silica vs. WAICO: The US Wants to Prohibit Europe from Using Chinese Artificial Intelligence

cryptonews.ruHace 5 hora(s)

Trading

Spot
活动图片