• AI KATANA
  • Posts
  • Meta releases an AI model that can transcribe and translate close to 100 languages

Meta releases an AI model that can transcribe and translate close to 100 languages

Also: If AI becomes conscious, how will we know?

Author

AI KATANA
August 23, 2023

Welcome!

In today's AI news, Meta unveils its new SeamlessM4T model, capable of transcribing and translating nearly 100 languages, though it comes with certain challenges and ethical concerns. Meanwhile, the concept of AI consciousness remains elusive despite a 14-point checklist proposed by experts. In partnership news, NVIDIA and VMware are ushering in a new era for enterprises with an optimized generative AI platform, and MediaTek aims to transition AI processing to local devices. Additionally, ElevenLabs expands its AI voice tech to 30 languages, while Solar AI boosts its endeavors with a fresh injection of USD 1.5 million in seed funding.

Sliced:

  • 🌍 Meta releases an AI model that can transcribe and translate close to 100 languages

  • 🤖 If AI becomes conscious, how will we know?

  • 🔓 VMware and NVIDIA Unlock Generative AI for Enterprises

  • 📱 Could you soon be running AI tasks right on your iPhone? MediaTek says yes

Meta has launched SeamlessM4T, an innovative AI model capable of transcribing and translating nearly 100 languages, marking a significant advancement in AI-driven speech-to-text and speech-to-speech. This model, which is open-sourced alongside the new translation dataset SeamlessAlign, can identify source languages without a separate model. Building on previous projects like No Language Left Behind and Universal Speech Translator, SeamlessM4T stands out due to its ambitious fusion of translation and transcription. Meta developed it using vast amounts of publicly sourced text and speech, though there are concerns about the ethics of using such data for commercial purposes. Furthermore, while boasting impressive performance metrics, there are biases within the model, particularly regarding gender representations. While AI translation systems like SeamlessM4T can provide efficient and accurate translations, they risk erasing the nuanced, unique interpretations human translators bring. Meta consequently recommends against using SeamlessM4T for critical tasks like medical or legal translations.

In 2021, Google's Blake Lemoine claimed that LaMDA, a chatbot, was sentient, sparking discussions about how to determine consciousness in artificial intelligence (AI). A multidisciplinary group of 19 experts recently proposed a comprehensive checklist to evaluate potential AI consciousness based on 14 criteria rooted in theories of human consciousness. While they concluded that current AI models, including ChatGPT, are unlikely to be conscious, the framework provides a systematic method for evaluating AI systems. However, distinguishing consciousness remains challenging as present understanding is based on human experiences, and consciousness could manifest differently in other entities, leaving the question open and continuously evolving.

VMware and NVIDIA have expanded their partnership to equip enterprises for the generative AI era on VMware's cloud infrastructure. Their integrated platform, "VMware Private AI Foundation with NVIDIA," will facilitate businesses in running generative AI applications, such as intelligent chatbots, tailored to their needs using custom data models. This solution combines NVIDIA's accelerated computing with VMware Cloud Foundation, optimized for AI. The CEOs of both companies emphasize the potential of this collaboration in transforming industries like finance, healthcare, and manufacturing by leveraging generative AI with custom applications. The platform offers benefits like data privacy, a range of model-building choices, top-tier performance, scalable GPU optimization, cost-efficiency, and rapid deployment. It incorporates NVIDIA NeMo, a framework that simplifies the adoption of generative AI, and NeMo's use of TensorRT optimizes AI performance on NVIDIA GPUs. Furthermore, significant tech players like Dell, HPE, and Lenovo will support the platform, enhancing its reach and performance capabilities. The product is expected to launch in early 2024.

Generative AI, a rapidly expanding technology used in chat and image generation systems, currently relies on extensive cloud-based data centers for processing. However, MediaTek, a leading semiconductor company from Taiwan, is pioneering a shift towards on-device AI. Collaborating with Meta, they're developing a means to run generative AI tasks on local devices, such as smartphones, without total reliance on external data centers. While this doesn't eliminate the need for data centers, due to the vastness of Large Language Model (LLM) datasets, it significantly reduces the dependency. MediaTek anticipates this Llama 2-based AI to be integrated into smartphones by year-end. On-device AI offers benefits like reduced latency, enhanced data privacy, and energy efficiency. Yet, challenges persist, including hardware limitations, potential security vulnerabilities, and the financing of mini edge datacenters. Nonetheless, the industry is gearing up for an era where your device's AI assistant operates independently, signaling an imminent paradigm shift in AI processing.

🛠️ AI tools updates

ElevenLabs, a startup for its AI-driven voice-generating platform, is extending its technology to support 30 languages through its newly introduced deep-learning model, Eleven Multilingual v2. This advancement allows for authentic voice generation, maintaining the speaker's unique voice characteristics consistently across all languages, thus ensuring the same speech style, including accents. This leap aligns with the company's vision of universal content accessibility. The application can benefit indie authors, assist in game translations, and aid those with visual impairments and learning needs. The platform recognizes a vast array of languages, from Korean and Finnish to Hindi and Portuguese.

💵 Venture Capital updates

Singapore-based solar photovoltaics startup, Solar AI, secured USD 1.5 million in a seed funding round led by Earth Venture Capital, and joined by Undivided Ventures, Investible, and climate tech angel investor David Pardo. The investment will support the expansion of Solar AI's rent-to-own solar initiative in Singapore and pave the way for regional growth in the upcoming year.

🫡 Meme of the day

⭐️ Generative AI image of the day