Loading your language..
Anthropic says its model Claude is helping to develop its own next version.

Anthropic says its model Claude is helping to develop its own next version.

C1🇹🇼 中文🇺🇸 English

September 18th, 2026

Anthropic says its model Claude is helping to develop its own next version.

C1
Please note: This article has been simplified for language learning purposes. Some context and nuance from the original text may have been modified or removed.

🇺🇸 English

In an announcement on Thursday, the AI lab Anthropic stated that its Claude is helping the company develop a smarter next version of the model.

人工智慧實驗室 Anthropic 週四在公告中表示,其 Claude 正在協助公司開發該模型更聰明的下一版本。

Claude is currently responsible for leading 26% of Anthropic's model research and development work; the company points out that this means it can, under human supervision, complete most assigned tasks "end-to-end from a high-level prompt."

Claude 目前負責主導 Anthropic 26%模型研發工作公司指出這表示能在人類監督下,「從高階提示端到端完成大多數指定任務

However, the model does not yet operate fully autonomously.

然而該模型尚未完全自主運作

However, roughly 90% of the company's research and development work is carried out in "collaboration" with Claude;

然而公司約 90%研發工作是在與 Claude「協作」下完成;

Anthropic says this means the model can complete "a substantial amount of work" under close human guidance.

Anthropic 表示這意味著該模型可在人類密切指導下完成「大量工作」

This announcement came at a time when several leading figures in the field of AIlargely led by Anthropic CEO Dario Amodeiwere calling for a slowdown in the pace of the technology's development, citing safety concerns.

這項公告發布之際,若干 AI 領域的領軍人物——在很大程度上由 Anthropic 執行長 Dario Amodei 帶頭——正出於安全方面的疑慮,呼籲放緩該技術的發展步伐。

As corporate leaders weigh the pace of AI development, the company said in a blog post: "We should do everything we can to close the gap between what frontier labs know and what the public knows."

在公司領導人權衡 AI 發展步調之際,該公司於一篇部落格文章中表示:「我們應竭盡所能縮小前沿實驗室所知公眾所知之間的落差。」

"This means we must measure AI progress more effectively, report on it publicly, and give society the opportunity to decide how this information is used."

這代表我們必須更妥善地衡量 AI 的發展、公開報告相關情況讓社會有機會決定如何運用這些資訊。」

The company stated that models accelerating their own development could lead to "humans finding it harder to understand or control these systems."

該公司表示模型加速自身發展可能導致人類更難理解或控制這些系統」。

It argued that sharing these metrics would help people better understand how close leading AI labs are to achieving recursive self-improvementthat is, a state in which a model can autonomously build its own successor.

它主張分享這些指標有助於人們更了解領先 AI 實驗室距離達成遞迴自我改進還有多近;所謂遞迴自我改進,是指模型能自主打造其後繼版本

Anthropic also urges other AI developers to regularly share similar metrics, and encourages the adoption of open methodologies so that these figures can be compared over time, and possibly even between different laboratories.

Anthropic 亦敦促其他 AI 開發者定期分享類似指標並鼓勵採用公開方法論使這些數字得以隨時間比較甚至可能在不同實驗室之間比較。

According to Anthropic's disclosures, it remains unclear how close the company believes it is to achieving recursive self-improvement; however, the pace of Claude's growing contribution to research and development is noteworthy.

根據 Anthropic 的揭露該公司認為自身距離實現遞迴自我改進有多近仍不明朗然而,Claude 對研發日益增加的貢獻其速度值得注意

The share of work that Claude "leads"that is, completes extensively while still under human oversightstood at zero in February.

Claude「主導」——即在仍受人類監督下大量完成——的工作占比在 2 月時為零

Six months later, it was already leading a quarter of research and development work, reaching this benchmark in August.

六個月後它已主導四分之一的研發工作並在 8 月達到這項基準

The company also disclosed details of its current agent oversight measures, noting that as of August, approximately 30,000 agents were engaged in research and engineering work.

該公司亦披露其現行代理監督措施的細節指出截至8月約有30,000個代理正從事研究與工程工作

The company stated that these oversight measures are crucial for understanding how often the monitoring system detects agent misconduct.

公司表示這些監督措施對於了解監測系統多常偵測到代理不當行為至關重要。

Anthropic recently committed to installing external third-party evaluators, who will be stationed within the company to oversee safety-related work.

Anthropic近日承諾設置外部第三方評估人員他們將進駐公司內部監督安全相關工作

When a researcher at Anthropic resigned last week, he issued a stark warning about the threat this technology poses to humanity, which sparked a great deal of recent discussion about AI safety.

Anthropic 一名研究人員上週辭職時嚴厲警告這項技術對人類構成的威脅,引發了近來關於 AI 安全的大量討論。

Since then, Amodei, OpenAI CEO Sam Altman, Elon Musk, and other tech leaders have all endorsed the idea of slowing down development, but other tech leaders and US President Donald Trump have taken the opposite stance.

此後AmodeiOpenAI 執行長 Sam AltmanElon Musk 和其他科技領袖都支持放慢發展的想法但其他科技領袖與美國總統 Donald Trump 則持反對立場

September 18th, 2026

Trending Articles

Buffett steps down as Berkshire chairman, succession plan moves to the next stage

Buffett steps down as Berkshire chairman, succession plan moves to the next stage

巴菲特卸任波克夏董事長,接班計畫邁向下一步

C1Sep 18
Trump expands media ban, barring CNN, MS NOW and Politico from the White House

Trump expands media ban, barring CNN, MS NOW and Politico from the White House

川普擴大媒體禁令,禁止CNN、MS NOW和Politico進入白宮

C1Sep 18
Bank of Japan raises interest rates to 1.25%, a 31-year high

Bank of Japan raises interest rates to 1.25%, a 31-year high

日本央行加息至1.25%,31年新高

C1Sep 18
Why is the Fed raising rates now? What does this mean?

Why is the Fed raising rates now? What does this mean?

聯準會為何現在升息?這代表什麼意義

C1Sep 17
As oil prices and bond yields retreated, US stocks rebounded, posting their best single-day performance in six weeks.

As oil prices and bond yields retreated, US stocks rebounded, posting their best single-day performance in six weeks.

油價與債券殖利率回落,美股反彈創六週最佳單日表現

C1Sep 17
The Fed raises interest rates for the first time in three years, ignoring Trump's calls for a rate cut.

The Fed raises interest rates for the first time in three years, ignoring Trump's calls for a rate cut.

聯準會3年來首度升息,無視川普降息呼籲

C1Sep 17
Jobless claims fall to 196,000, lowest since mid-July, as layoffs remain low

Jobless claims fall to 196,000, lowest since mid-July, as layoffs remain low

失業救濟申請降至19.6萬,創7月中以來新低,裁員仍處低位

C1Sep 17
Trump calls AI risk a "hoax," pointing to a "sick conspiracy" against AI and data centers.

Trump calls AI risk a "hoax," pointing to a "sick conspiracy" against AI and data centers.

川普稱AI風險是「騙局」,並指存在針對AI與資料中心的「病態陰謀」

C1Sep 14
The Supreme Court will rule on a case in which Maine lobster fishermen challenge the requirement for GPS vessel trackers.

The Supreme Court will rule on a case in which Maine lobster fishermen challenge the requirement for GPS vessel trackers.

最高法院將裁定緬因州龍蝦漁民挑戰GPS船隻追蹤器要求一案

C1Sep 14