AI Briefing#kakao
The latest AI and developer news about #kakao, with the original source and a short summary.
The latest AI and developer news about #kakao, with the original source and a short summary.
Kakao has open-sourced the 'Kanana Safeguard' series, consisting of three AI guardrail models specialized for Korean user environments, under the Apache 2.0 license.
Kakao has released Kanana 1.5 8B and 2.1B models under the Apache 2.0 license, allowing for commercial use.
Kakao has open-sourced the 'Kanana 1.5' 8B and 2.1B models under the Apache 2.0 license, enhancing math and coding performance and long-context understanding to support Agentic AI.
Kakao has released 'Kanana-o', a multimodal LLM that integrates its voice model (Kanana-a) and vision model (Kanana-v) using model merging techniques, demonstrating superior performance compared to global models in Korean speech recognition (CER 8.5).
Wrote 1,500 lines of code in 6 hours using Cursor and Claude 3.7, but concluded that complete replacement is impossible
In experiments targeting the OLIVE platform, tasks were completed in less than half the estimated time while maintaining verified quality.
In server log Iceberg APPEND loading tests, applying zstd level 9 reduced file size by 51.4% compared to ORC.
Shares operational experience and insights gained from adopting Apache Iceberg for DB and server log management.
Kakao CodeBuddy utilizes LLM as a Judge for performance evaluation, proposing methods to enhance evaluation reliability by improving prompt structure and mitigating position bias.
Kanana Flag demonstrated overwhelming performance on Korean benchmarks while reducing training costs by over 50% compared to third-party models.
Over 150 participants attended the competition held at BEXCO in Busan to develop ESG-based web and mobile apps in a hackathon.
Lighthouse scores improved from the 70s to the 90s by applying virtual scrolling, uncontrolled components, and Message Queue batching.
The average performance score rose from 81 to 91, with all pages achieving a score of 80 or higher.
Kanana Nano-2.1B maintains 93% of Kanana Essence's Korean performance despite a 26% size reduction.
Carib v2 reduces Redis load to less than 1% and processes 600,000 messages per second without loss on a 2-core VM.
Buffering and marshaling optimizations using gobwas reduce message loss rate by 60% and CPU usage by 25%
Kakao has unveiled a security and privacy protection system that applies Privacy by Design principles to prevent privacy infringements from the service planning stage and enhances expertise by separating the roles of the CPO and CISO.
Kakao operates a BCM framework certified under ISO 22301 and conducts company-wide disaster recovery drills at least once a year.
Kakao has revealed its strategy to enhance service stability, featuring AI-assisted code writing, open source management via the OLIVE Platform, and harmful content blocking using Safebot.
Kakao has built a multi-data center platform to enable integrated monitoring and automated incident response for approximately 60 cloud services.
Redundancy in power, telecommunications, and cooling, along with seismic resistance up to Richter 6.5, ensures uninterrupted operations.
Kakao presented a resilience strategy managing three key elements—failure frequency, impact scope, and recovery time—within KakaoTalk's server environment, which sees an average of 11 deployments per day.
With an average score of 75.1 on Korean benchmarks, surpassing GPT-4o (61.9), the model adopts Native resolution and the C-Abstractor architecture.
Achieved a 50% reduction in server resource usage and a significant decrease in memory usage from 90% to 15%