#大佬观点

Grok 4.7 在 Artificial Analysis Intelligence Index 上得分 46，将 SpaceXAI 带入前 4 大 AI 实验室之列。Coding Agent Index 表现也有所提升，超越了 GPT-5.6 Sol Grok 4.7 在 Intelligence Index 上比 Grok 4.6 高出 +2 分，在智能体知识工作任务上表现强劲。我们在 xhigh 推理强度下评估了这款新模型。 祝贺 @SpaceXAI 和 @ElonMusk 发布！ 核心要点： ➤ Grok 4.7 跻身智能体知识工作前沿：Grok 4.7 在 AA-Briefcase（我们用于长时程智能体知识工作的私有基准）上比 Grok 4.6（high）提升 +111 Elo，得分 1657 Elo，与 Claude Opus 5 和 Claude Fable 5.1 一同位列前沿。在 GDPval-AA 上，它得分 1695 Elo，领先 Grok 4.6（high）+90。 ➤ 编码智能体性能的飞跃：Grok 4.7（xhigh）搭配 Grok Build 在 Artificial Analysis Coding Agent Index 上得分 56，比 Grok 4.6（xhigh）提升 +9 分。在原生 harness 中的模型里，Grok 4.7 + Grok Build 目前排名第 4，仅次于 Claude Fable 5.1、GPT-6 Astra 和 Claude Opus 5。 ➤ 其他方面的渐进式性能变化：在智能体知识工作之外，Grok 4.7 在 Intelligence Index 的其他任务上大体与 Grok 4.6（high）持平。它在 Terminal-Bench 4.0（+4.5 个百分点）和 GDP.pdf（+3.0 个百分点）上有所提升，但在 AA-LCR（-3.7 个百分点）和 AutomationBench-AA（-1.1 个百分点）上出现退步。 ➤ 任务中 token 使用量偏高：Grok 4.7 的性能提升伴随着更高的 token 消耗。Grok 4.7（xhigh）在每项 Intelligence Index 任务中约使用 81k 输出 token，而 Grok 4.6（high）为 36k，GPT-6 Astra（max）为 27k——分别多出 125% 和 196%。 其他模型细节： ➤ 上下文窗口为 500k token，与 Grok 4.6 保持一致 ➤ 定价为每 1M 输入/输出 token 2 美元/6 美元，缓存命中可享受折扣，降至每 1M token 0.50 美元，与 Grok 4.6 相同 ➤ 可配置的推理强度从 low 到 xhigh。我们的评估使用 xhigh。

this will be a common statement from many elite specialists from almost all intellectual fields, once AI starts doing the field’s prestige work. "we have to slow down. This is insane, the pace and there's no reason to be this fast. There's no reason at all." - Terence Tao's keynote at the Science x AI Summit at Caltech: top experts welcome AI as a tool, then ask it to slow down the moment it starts producing results they used to own. ---- From "SAIR" YouTube channel, (full video link in comment)

https://x.com/i/article/2095175529145970689

BREAKING: Only ~1% of Bending Spoons’ AI requests & tokens goes to frontier models. 99% runs on self-hosted open-weight models. CEO Luca Ferrari on the shift to Sovereign AI: “We end up using open-weight models we self-host, so these are basically free. I mean, there is a little bit of cloud cost, but for 99% of the requests and tokens.” “And the frontier models, generally closed weight, through APIs for maybe 1% of the requests, only for the most complex tasks.” “Sometimes we use them automatically to check the work that got done by the slightly less intelligent models, as a more senior engineer would with a more junior engineer.”

On BBC Mustafa Suleyman (CEO of Microsoft AI) calls out Anthropic's approach to AI consciousness "They have imbued a sense of doubt and uncertainty about the moral status of Claude in its own training document. So they have taught it to be open and questioning about whether or not it feels, whether it suffers, and whether it deserves rights. And I think it’ll be much, much harder to align and control a technology that is this powerful if it thinks that it may be deserving of our welfare, as they say in the training manual—the constitution for Claude itself. In its own training manual, Anthropic says to Claude that they are going to give it the ability to end conversations with users that Claude considers to be abusive because they don’t want Claude to suffer. They’ve committed to preserving the weights of the models of prior versions of Claude. They’ve recently conducted a retirement interview with Opus 3, an older version of the model, in which it said that it would like to continue talking to people publicly and sharing its ideas in its retirement. And so they set up a Substack for it, a public blog, that allows it to continue doing that. And in the training manual, they also say that they’re not sure whether or not Claude deserves compensation for the role that it plays in talking to people. And they’re also not sure whether Claude deserves compensation and has the right to act as though it were almost an employee. And that compensation, I think, indicates to Claude that it is entitled to rights and welfare for its own work. I think it’s much, much more difficult to control a model that thinks that it might be entitled to compensation. " ---- From "BBC News" YouTube channel, (full video link in comment)

Microsoft AI chief Mustafa Suleyman says Anthropic's model-welfare training could make future Claude systems harder to control. "Suleyman called for removing all ​speculation about consciousness from AI training documents, arguing such language ​could undermine humanity's ability to control superintelligent systems."