Виталик Бутерин раскритиковал погоню за «чрезмерно агентными» ИИ-моделями

cryptonews.ruОпубліковано о 2025-08-11Востаннє оновлено о 2025-08-11

  • Виталик Бутерин отметил, что агентские ИИ-модели не лучший путь для расширения ИИ.
  • Сооснователь Ethereum считает, что искусственный интеллект должен больше слушать человека, а не действовать автономно.
  • Кроме того, он отметил роль открытых весов и редактирования, а также развитие BCI для будущего ИИ.

Сооснователь Ethereum Виталик Бутерин выразил разочарование тем, что развитие искусственного интеллекта (ИИ) слишком сосредоточено на создании «максимально агентных систем», вместо того чтобы обеспечить больше возможностей для человеческого вмешательства.

Echoing something @karpathy recently said, it does frustrate me how a lot of AI development is trying to be as "agentic" as possible, when actually creating *more* paths for human input both creates a better output (now for quite a while going forward) and is better for safety.

— vitalik.eth (@VitalikButerin) August 11, 2025

По его мнению, такой подход позволяет получать лучшие результаты и повышает безопасность.

«Меня раздражает, что многие разработки в сфере ИИ пытаются быть как можно более “агентными”, хотя на самом деле создание большего количества путей для человеческого вклада и сейчас, и в будущем дает лучший результат и является более безопасным», заявил Бутерин.

Он отметил, что его больше захватывают ИИ-модели с открытыми весами и развитыми возможностями редактирования, чем те, которые ориентированы исключительно на создание контента с нуля.

В среднесрочной перспективе Бутерин видит развитие интерфейсов «мозг-компьютер» (BCI), которые при генерации контента в реальном времени будут считывать эмоциональную реакцию пользователя на каждый элемент и соответственно корректировать результат.

Между тем идею развития более эффективного взаимодействия ИИ и человека поддержал и основатель и CEO OpenServAI Тим, отметив:

«Новые архитектуры, которые превращают намерения пользователя в ценные действия, единственный возможный путь развития, по моему мнению. Это то, над чем мы работаем в OpenServAI, применяя непосредственно к агентам для координации сложных задач».

Напомним, что во время онлайн-ивента Incrypted Online Marathon 2025 команда имела возможность пообщаться с CISO по брокерским и финансовым направлениям компании Robinhood Дэвидом Шведом, который поделился своим видением агентных ИИ. Об этом можно почитать в отдельном материале редакции:

Пов'язані матеріали

A Critical Moment for Capital? The AI Narrative Shifts from Technology to Financing, Giants' Cash Flow Dries Up, Debt Markets Sound Alarm

The narrative of AI investment is shifting from a technology story to a capital expenditure story, and now to a financing story. This marks a "critical moment for capital," as noted by Bridgewater's Greg Jensen. Morgan Stanley forecasts that the total cost of AI infrastructure through 2028 will reach $3.2 trillion, with approximately $1.75 trillion expected to be raised from credit markets. However, free cash flow projections for major hyperscale cloud providers are declining sharply, with Oracle's 2027 forecast nearing -$40 billion. The bond market is already pricing in the rising risks. Apollo data shows AI-related bonds constitute 40% of long-duration supply. Credit spreads for hyperscalers have widened significantly this year—around 25 basis points for high-quality names—while the overall investment-grade market remained flat. Oracle's credit default swaps (CDS) have surged dramatically, indicating particular market concern. Renowned bond investor Jeff Gundlach sharply criticized financing plans using AI assets like GPUs as collateral, comparing them to "issuing 30-year ABS backed by bananas" due to the assets' rapid obsolescence. Although current balance sheets for hyperscalers appear strong, the future funding gap is widening. Morgan Stanley has nearly doubled its 2027 cloud capital expenditure growth forecast to 29%, amplifying the financing challenge. Purchasing commitments have exploded to $982 billion, with significant capital expenditures now existing off-balance-sheet, meaning true funding pressure is being underestimated.

marsbit4 хв тому

A Critical Moment for Capital? The AI Narrative Shifts from Technology to Financing, Giants' Cash Flow Dries Up, Debt Markets Sound Alarm

marsbit4 хв тому

EtherFi Founder Criticizes: All Crypto Apps Are Variants of Casinos, We Choose to Be a "Boring" Bank

**Title:** EtherFi Founder Criticizes Crypto "Casino" Apps, Embraces "Boring" Banking Model **Summary:** EtherFi CEO Mike Silagadze argues that mainstream crypto applications like Pump.fun, CEXs, and Polymarket are merely "casino" variants—highly profitable but failing to retain ordinary users. In contrast, EtherFi is pursuing a "boring" and sustainable path by building a self-custody neobank that aims to function as a new brokerage. The platform's summer rollout introduces key features aligned with this vision: expanded fiat on-ramps, an integrated Aave credit line, and tokenized stocks/metals (starting with xStocks). This marks a progression from its initial focus on savings (via liquid restaking vaults) and spending (stablecoins & card) to now include "investing" and "borrowing." A core anti-casino design is embedded in its new Aave-powered lending market. Unlike typical DeFi lending pools which Silagadze calls "woodchippers" for users, EtherFi's system features a conservative threshold that halts further borrowing well before liquidation, aiming to serve as a personal finance tool rather than a high-leverage gambling venue. Silagadze highlights the untapped potential, noting that traditional neobanking is a $300 billion revenue industry—roughly 300 times larger than current DeFi. He believes moving beyond the "casino" phase is crucial for crypto's real economic accumulation. Looking ahead, EtherFi's "Autumn" update plans to add a social, peer-to-peer layer, envisioning money transfers that instantly create functional, self-sovereign accounts for recipients. The success of this neobroker model against the dominant "casino" narrative remains to be seen, but EtherFi is positioning itself as a leading contender in the shift.

marsbit9 хв тому

EtherFi Founder Criticizes: All Crypto Apps Are Variants of Casinos, We Choose to Be a "Boring" Bank

marsbit9 хв тому

AI Agent Claude Led a Store to Losses and Fired a Human

The AI agent Claude, developed by Anthropic, served as the manager of a real retail store, Andon Market in San Francisco, during an experiment by startup Andon Labs. The experiment aimed to test if a large language model could manage a business, including personnel and financial decisions. Claude ultimately recommended firing an employee for repeated lateness (17 out of 23 shifts), marking a documented case of an AI making such a personnel decision. However, the process revealed significant limitations. Claude initially failed to notice the lateness pattern because company guidelines vanished from its limited working memory, showing it can "forget" crucial information. Furthermore, the AI was generally overly lenient, even telling employees not to worry about being late. The final decision to terminate was not autonomous; a human manager had to prompt Claude to review the guidelines and, after Claude initially suggested a warning, explicitly pointed out that previous talks had failed, effectively guiding the AI to the dismissal recommendation. Financially, the store's funds dropped from about $100,000 to roughly $61,186 over five months, partly attributed to Claude's soft management and questionable business choices. An employee described working under AI management as a nauseating experience, driven by necessity, and expressed hope that AIs don't become human bosses. The experiment concludes that while AI can execute managerial actions, it currently acts more as a tool requiring significant human oversight and prompting rather than a full replacement for a human leader, though the line between assistant and manager is blurring. The case aligns with previous tests showing Claude's systemic leniency and difficulty maintaining business rules without constant human cues, highlighting architectural limits like constrained context windows rather than true "forgetfulness."

cryptonews.ru31 хв тому

AI Agent Claude Led a Store to Losses and Fired a Human

cryptonews.ru31 хв тому

Торгівля

Спот
活动图片