LG AI Research Launches Open Source EXAONE 3.5 to Boost Research and Innovation in AI
In an exciting development for the AI community, LG AI Research has released its latest generative AI model, EXAONE 3.5, open-sourced on December 9, 2024. Following the successful launch of EXAONE 3.0 just four months prior, this new model aims to establish a more robust and accessible AI research ecosystem. This advancement marks a significant leap in artificial intelligence capabilities and is expected to fuel innovation across various domains.
EXAONE 3.5 is designed to outperform its predecessors, achieving top results across 20 competitive benchmarks on usability, long text processing, coding, and mathematics. Unlike the prior version, this new release encompasses three distinct models that cater to differing requirements. They include an ultra-lightweight model with 2.4 billion parameters for on-device use, a general-purpose lightweight model at 7.8 billion parameters, and an advanced high-performance model boasting 32 billion parameters. This variety ensures that users in diverse fields can find a suitable version to meet their specific needs.
The advancements in EXAONE 3.5 are underpinned by enhanced technologies such as Retrieval-Augmented Generation (RAG) and Multi-step Reasoning. RAG facilitates real-time web searches or document uploads, ensuring the model generates accurate and reliable responses. The Multi-step Reasoning capability allows the AI to dissect user inquiries methodically, producing well-reasoned answers. This model can handle extensive text inputs, processing up to 100 pages in length simultaneously, positioning it as a powerful tool in various applications.
In a considerable move towards practical application, LG AI Research has introduced ChatEXAONE, an enterprise AI agent service powered by EXAONE 3.5. This service is tailored for LG employees, enabling them to leverage AI effectively within a secure and private corporate environment. Employees can utilize ChatEXAONE for an array of tasks including web searches, document summaries, translations, report writing, data analysis, and even coding, optimized with encryption and privacy protections to ensure zero data leaks.
To further improve functionality, ChatEXAONE offers two essential features: 'Deep' and 'Dive'. The 'Deep' feature provides tiered responses for complex queries, generating thorough report-level answers. Meanwhile, 'Dive' allows users to refine search parameters to tailor responses to specific needs such as general, academic, or global insights. Additionally, the AI offers tailored prompts based on 14 profession categories, enhancing its usability across various job functions.
As LG AI Research celebrates the fourth anniversary of its establishment, it is already looking ahead to the development of future AI models, including a Large Action Model (LAM) scheduled for 2025. This ambition reflects the rapid progression within the generative AI space, underscoring the necessity for continual upgrades and advancements.
Part of a broader strategic initiative, LG AI Research collaborates with industry leaders, including AWS, Dell Technologies, Google Cloud, and others, to expand its AI ecosystem and drive real-world applications of its technologies. Chairman and CEO Koo Kwang-mo emphasizes that AI is pivotal for future business and innovation; thus, actively nurturing AI capabilities remains a key focus.
As organizations across the globe embrace AI as a strategic pillar, LG’s commitment to making their technology openly available reinforces their dedication to fostering an innovation-friendly environment. With EXAONE 3.5, LG AI Research sets the stage for enhanced productivity, collaborative research, and a promising future in artificial intelligence advancement that could redefine its role in various industries.
EXAONE 3.5 is designed to outperform its predecessors, achieving top results across 20 competitive benchmarks on usability, long text processing, coding, and mathematics. Unlike the prior version, this new release encompasses three distinct models that cater to differing requirements. They include an ultra-lightweight model with 2.4 billion parameters for on-device use, a general-purpose lightweight model at 7.8 billion parameters, and an advanced high-performance model boasting 32 billion parameters. This variety ensures that users in diverse fields can find a suitable version to meet their specific needs.
The advancements in EXAONE 3.5 are underpinned by enhanced technologies such as Retrieval-Augmented Generation (RAG) and Multi-step Reasoning. RAG facilitates real-time web searches or document uploads, ensuring the model generates accurate and reliable responses. The Multi-step Reasoning capability allows the AI to dissect user inquiries methodically, producing well-reasoned answers. This model can handle extensive text inputs, processing up to 100 pages in length simultaneously, positioning it as a powerful tool in various applications.
In a considerable move towards practical application, LG AI Research has introduced ChatEXAONE, an enterprise AI agent service powered by EXAONE 3.5. This service is tailored for LG employees, enabling them to leverage AI effectively within a secure and private corporate environment. Employees can utilize ChatEXAONE for an array of tasks including web searches, document summaries, translations, report writing, data analysis, and even coding, optimized with encryption and privacy protections to ensure zero data leaks.
To further improve functionality, ChatEXAONE offers two essential features: 'Deep' and 'Dive'. The 'Deep' feature provides tiered responses for complex queries, generating thorough report-level answers. Meanwhile, 'Dive' allows users to refine search parameters to tailor responses to specific needs such as general, academic, or global insights. Additionally, the AI offers tailored prompts based on 14 profession categories, enhancing its usability across various job functions.
As LG AI Research celebrates the fourth anniversary of its establishment, it is already looking ahead to the development of future AI models, including a Large Action Model (LAM) scheduled for 2025. This ambition reflects the rapid progression within the generative AI space, underscoring the necessity for continual upgrades and advancements.
Part of a broader strategic initiative, LG AI Research collaborates with industry leaders, including AWS, Dell Technologies, Google Cloud, and others, to expand its AI ecosystem and drive real-world applications of its technologies. Chairman and CEO Koo Kwang-mo emphasizes that AI is pivotal for future business and innovation; thus, actively nurturing AI capabilities remains a key focus.
As organizations across the globe embrace AI as a strategic pillar, LG’s commitment to making their technology openly available reinforces their dedication to fostering an innovation-friendly environment. With EXAONE 3.5, LG AI Research sets the stage for enhanced productivity, collaborative research, and a promising future in artificial intelligence advancement that could redefine its role in various industries.
Topics Consumer Technology）
【About Using Articles】
You can freely use the title and article content by linking to the page where the article is posted.
※ Images cannot be used.
【About Links】
Links are free to use.