谷歌确认新一代旗舰模型 Gemini 4 即将推出,有望不用等到年底Google confirms next-gen flagship model Gemini 4 is coming soon, expected not to wait until end of year


AI free token/AI免费积分
谷歌确认新一代旗舰模型 Gemini 4 即将推出,有望不用等到年底
感谢IT之家网友 Nuc_F、补药吖、华南吴彦祖 的线索投递!

IT之家 9 月 24 日消息,The Information 当地时间 23 日举办的 AI Agenda Live 峰会上,谷歌 DeepMind 负责人科拉伊 · 卡武克奥卢透露,谷歌新一代旗舰模型 Gemini 4 即将推出,现已进入开发流程中的后训练初期。

IT之家注:“后训练”是在基础 AI 模型完成后进一步调整模型,提高其行为可靠性。

卡武克奥卢希望 Gemini 4 能在“远早于年底”的时间点推出。“笔者的打算是尽快拿出后训练阶段的早期成果,因为笔者看到了结果,也很振奋。”

随后,谷歌会继续快速迭代模型。

Gemini 4 将成为谷歌自去年年底推出 Gemini 3 系列后首款下一代旗舰模型。谷歌 2 月发布了 Gemini 3.1 升级版,此后还推出多款规模更小、成本更低,但复杂推理能力也较弱的 Flash 模型。

今年 5 月,谷歌 CEO 桑达尔 · 皮查伊曾在年度 I/O 大会上宣布,重大更新 Gemini 3.5 Pro 将在 6 月发布,但最终没有亮相。

卡武克奥卢称,Gemini 3 和 3.1 推出后,谷歌没有继续发布 Gemini 3.5 Pro,而是“稍微退了一步”,转而把精力放在 Flash 模型上。“对笔者来说,当时最关键的是尽可能加快学习速度。”

卡武克奥卢没有说明谷歌是否正式取消 Gemini 3.5 Pro,而是明确称,当前重点是推出 Gemini 4。

被问到是否担心谷歌已经落后于同行时,卡武克奥卢回复道,自己对团队“百分之百有信心”。“笔者的团队强得不可思议,我对团队有最大的信任。在我看来,笔者始终会站在工艺前沿,这是确定无疑的。”

当下,Gemini 4 正处于防护机制开发和安全测试阶段。谷歌工程师已在内部使用这款模型运行 Antigravity AI 编程工具。

与 AI 行业的一些同行不同,卡武克奥卢对于行业是否达成 AGI 并没有明确倾向。他只是表示:“讨论笔者是否达成了 AGI,并不是正确的讨论方向。真正应该讨论的是,笔者能不能构建出值得信任的智能体?”

本文来源:https://www.ithome.com/1/006/800.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

Thanks to IT Home readers Nuc_F, Buyao Ya, and South China's Wu Yanzu for their tip submissions!

IT Home, September 24 – At The Information's AI Agenda Live summit held on the 23rd local time, Google DeepMind CEO Koray Kavukcuoglu revealed that Google's next-generation flagship model, Gemini 4, is about to launch and has now entered the early post-training phase of the development process.

IT Home note: "Post-training" refers to further adjusting the model after the base AI model is completed to improve the reliability of its behavior.

Kavukcuoglu hopes Gemini 4 will launch "well before the end of the year." "My intention is to deliver early results from the post-training phase as soon as possible, because I have seen the results and I am very excited."

After that, Google will continue to iterate on the model rapidly.

Gemini 4 will be Google's first next-generation flagship model since the launch of the Gemini 3 series late last year. Google released the Gemini 3.1 upgrade in February, and has since also launched several Flash models that are smaller, lower-cost, but with weaker complex reasoning capabilities.

In May this year, Google CEO Sundar Pichai announced at the annual I/O conference that the major update Gemini 3.5 Pro would be released in June, but it ultimately never appeared.

Kavukcuoglu said that after the launch of Gemini 3 and 3.1, Google did not continue to release Gemini 3.5 Pro, but instead "stepped back slightly" and shifted its focus to the Flash models. "For me, the most critical thing at the time was to speed up the learning pace as much as possible."

Kavukcuoglu did not state whether Google has officially canceled Gemini 3.5 Pro, but clearly said that the current focus is on launching Gemini 4.

When asked whether he is worried that Google has fallen behind its peers, Kavukcuoglu replied that he has "100% confidence" in his team. "My team is incredibly strong, and I have the utmost trust in them. In my view, I will always stand at the forefront of the craft – that is certain."

Currently, Gemini 4 is in the stage of safety mechanism development and security testing. Google engineers have already been using this model internally to run the Antigravity AI coding tool.

Unlike some peers in the AI industry, Kavukcuoglu does not have a clear stance on whether the industry has achieved AGI. He simply said: "Discussing whether we have achieved AGI is not the right direction of discussion. What we should really discuss is: can we build intelligent agents that are worthy of trust?"

Google confirms next-gen flagship model Gemini 4 is coming soon, expected not to wait until end of year

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry