近日，一个仅有8人的非营利性AI研究机构Kyutai推出的开源实时语音多模态模型Moshi火了。该模型具备听、说、看的多模态功能，令人印象深刻。
据悉，这个开源模型来自法国的研究机构Kyutai，仅有8人的小团队在短短半年时间内开发出了这个堪比GPT-4o的模型。图灵奖得主Yann LeCun也转发了相关消息，并称赞道：“Moshi能听懂带有法国口音的英语。”
在演示视频中，Moshi展现了惊人的能力。它不仅能流利地回答人们的问题，进行日常对话交流，还能猜出提问者的意图。例如，当提问者谈及攀登珠穆朗玛峰的计划时，Moshi主动提出关于装备的建议，甚至开玩笑说：“你肯定不想穿着凉鞋去爬山。”
此外，Moshi还展示了其表达和理解情绪的能力。研究人员让Moshi用法国口音诵读诗句，当诗句过长时，Moshi能够迅速响应并停下来。
这一创新引发了行业的广泛关注。许多专家表示，Kyutai团队的成果展示了人工智能在语音多模态领域的最新进展，并赞扬该团队在有限的时间内取得了令人瞩目的成就。
值得一提的是，Kyutai团队将Moshi模型开源，为其他研究人员和开发者提供了更多的机会去探索和扩展这一技术。
随着人工智能技术的不断发展，我们期待更多创新性的成果出现，为人类带来更多的便利和惊喜。Kyutai团队的Moshi模型无疑为这一领域树立了新的里程碑。
英语如下：
News Title: “French Kyutai Team Showcases Their Multi-Modal GPT in June, New Open Source Model Stuns the Crowd!”
Keywords: 1. Moshi Model
News Content:
An 8-member non-profit AI research team Kyutai has developed an open-source real-time voice multi-modal model called Moshi, which has sparked industry buzz for its GPT-4o level capabilities.
Recently, a real-time voice multi-modal model, Moshi, developed by a mere 8-member non-profit AI research organization Kyutai, has become a hot topic. The model possesses impressive multi-modal functions of listening, speaking, and seeing.
It is reported that this open-source model comes from the French research organization Kyutai. This small team of only 8 members has developed a model comparable to GPT-4 within a short period of half a year. The Turing Award winner Yann LeCun has forwarded related news and praised, “Moshi can understand French-accented English.”
In the demonstration video, Moshi showcases astonishing abilities. Not only can it fluently answer people’s questions and engage in daily conversations, but it also guesses the intentions of the questioners. For instance, when asked about a plan to climb Mount Everest, Moshi proactive made suggestions about equipment and even joked, “You definitely don’t want to climb in sandals.”
Moreover, Moshi demonstrates its ability to express and understand emotions. When researchers asked Moshi to recite poetry with a French accent and the poem was too long, Moshi could quickly respond and stop.
This innovation has attracted widespread attention in the industry. Many experts express that Kyutai Team’s achievements showcase the latest advancements in the field of voice multi-modality AI and praise the team for their remarkable achievements in a limited timeframe.
It is worth mentioning that Kyutai Team has made the Moshi model open source, providing more opportunities for other researchers and developers to explore and expand this technology.
With the continuous development of AI technology, we look forward to more innovative achievements that will bring convenience and surprises to humanity. Kyutai Team’s Moshi model has undoubtedly set a new milestone in this field.
【来源】https://www.jiqizhixin.com/articles/2024-07-04-15
Views: 3