The Telecommunications Expertise Affiliation (TTA) introduced on Dec. 23 that it has granted Synthetic Intelligence (AI) Credibility Evaluation and Testing (CAT) certification to Mi:dm Ok 2.0 Base, a big language mannequin (LLM) independently developed by KT Corp.
CAT is a personal voluntary certification system operated by TTA to reply to moral and technical danger elements in synthetic intelligence and strengthen the competitiveness of the home AI business. It comprehensively evaluates whether or not AI fashions and techniques meet belief necessities similar to governance, transparency, and accountability.
This certification marks the primary case the place worldwide standard-based certification standards have been utilized to large-scale LLMs with huge parameters and general-purpose capabilities. Mi:dm Ok 2.0 Base has realized from large-scale Korean language knowledge and is able to varied pure language processing duties together with classification, question-answering, summarization, era, and transformation. It’s an open-source mannequin with 11.5B parameters presently out there by way of the Hugging Face platform.
TTA defined that by way of this certification, it confirmed that process-specific outputs and achievements are quantitatively and constantly managed, together with coaching knowledge filtering content material, vulnerability remediation by way of mannequin crimson teaming, and comparative testing of assorted reliability analysis benchmark performances. TTA President Son Seung-hyun stated, “We are going to proceed to play a key position in realizing the AI high 3 powerhouse standing and creating South Korea’s AI innovation ecosystem primarily based on our ICT testing and certification experience.”
Bae Quickly-min, head of the AI Future Lab at KT’s Expertise Innovation Division, stated, “This certification is the results of goal verification of the techniques and execution capabilities that KT has ready for accountable AI growth and operation forward of the implementation of the AI Primary Act.”
