
Sony Analysis has inked a partnership to assist take a look at and finetune the Southeast Asian Languages in One Community (SEA-LION) synthetic intelligence (AI) mannequin, specializing in Indian languages.
The AI arm of Sony Analysis will work with AI Singapore (AISG) liable for the event of SEA-LION, to plug gaps in making certain the massive language mannequin (LLM) stands up nicely on the worldwide panorama, representing the area’s populations and languages. The companions stated in a press release Tuesday their analysis collaboration will contain LLMs underneath the SEA-LION umbrella, all of that are pre-trained and instruct-tuned particularly on Southeast Asian cultures and languages.
The open-source LLM has been educated on 981 billion language tokens, which AISG defines as fragments of phrases created from breaking down textual content through the tokenization course of. These fragments embrace 623 billion English tokens, 128 billion Southeast Asia tokens, and 91 billion Chinese language tokens.
Additionally: Misplaced in translation: AI chatbots nonetheless too English-language centric, Stanford examine finds
The partnership entails that Sony will work on assessments and suggestions on the AI mannequin, tapping the Japanese vendor’s analysis presence in India and experience within the growth of LLMs for Indian languages (together with Tamil). Tamil is estimated for use by 60 million to 85 million individuals globally, most of whom are based mostly in India and Southeast Asia.
Sony will alternate finest practices on LLM growth and analysis methodologies, in addition to the appliance of its analysis in speech era, content material evaluation, and recognition.
The mixing of the SEA-LION AI mannequin with Tamil language capabilities has the potential to spice up the efficiency of recent functions, stated AISG’s senior director of AI merchandise Leslie Teo. He added that the Singapore company may even share its data and finest practices in LLM growth.
Additionally: AI leaders urged to combine native information fashions for range’s sake
IBM and Google are amongst different business gamers drawn into finetuning the regional LLM, together with making it obtainable for builders to construct custom-made AI functions.
“Entry to LLMs that handle the worldwide panorama of language and tradition has been a barrier to driving analysis and creating new applied sciences which are consultant and equitable for the worldwide populations we serve,” stated Hiroaki Kitano, president of Sony Analysis. “Range and localization are important forces. In Southeast Asia particularly, there are greater than 1,000 totally different languages spoken by the residents of the area. This linguistic range underscores the significance of making certain AI fashions and instruments are designed to assist the wants of all populations around the globe.”
Additionally: Transparency is sorely missing amid rising AI curiosity
Established in April 2023, Sony Analysis focuses on technological growth that may enhance content material creation and fan engagement, together with within the areas of AI, sensing, and digital areas. For example, its deep studying analysis crew has been engaged on applied sciences that embody, amongst others, mannequin compression and neural rendering, which it hopes will be built-in into Sony’s GUI growth instrument Neural Community Console, and open-source libraries Neural Community Libraries.
These applied sciences can be utilized in AI-powered electronics merchandise spanning numerous sectors, corresponding to video games, films, and music, and video games, Sony stated.
Its interactive leisure unit has filed a patent for a “harassment detection equipment” that features an enter unit constructed to obtain biometric information and with capabilities to generate, based mostly on biometric information, emotion information related to customers, in accordance with an April 2024 publication on World Mental Property Group’s PatentScope search platform.
Additionally: New world customary goals to construct safety round massive language fashions
With the system, Sony hopes to have the ability to detect and mitigate communications between people in multi-player video games or digital actuality experiences which are malicious, corresponding to harassment. Tapping machine studying and AI fashions, the system can detect biometric information corresponding to speech and decide a participant’s emotional state, as an example, by way of sounds corresponding to sobbing and screaming. These could also be used to establish victims of harassment throughout the shared atmosphere, in accordance with the submitting.
In Could, Sony Music Group launched a press release noting that its artists’ copyrighted works, together with compositions, lyrics, and audio recordings, shouldn’t be scraped and used to coach AI fashions except explicitly licensed.