Summary of SenseTime Conference Call Company Overview - **Company**: SenseTime (0020.HK) - **Industry**: Artificial Intelligence (AI) Software Key Points 1. **Generative AI Trend**: Management remains optimistic about the generative AI trend in China, emphasizing the launch of their foundation model, SenseNova V6, which features competitive costs for training and inferencing [1][2][4] 2. **MOU with Chinese University**: SenseTime signed a Memorandum of Understanding (MOU) with the Faculty of Law at the Chinese University of Hong Kong to enhance legal information accessibility through AI [4][7] 3. **Foundation Model - SenseNova V6**: The SenseNova V6 model, introduced in April, boasts multimodal reasoning capabilities and cost efficiency in both training and inferencing. It can handle long-form video understanding and supports various use cases such as role-playing, translation, and cultural tourism guiding [8][4] 4. **AI Supply Chain Outlook**: Management's positive outlook on generative AI aligns with a broader positive view on the China AI supply chain, indicating an increase in entities adopting generative AI technologies [2][4] 5. **Investment Upgrades**: Analysts have upgraded several companies within the AI supply chain, including SMIC, VeriSilicon, AMEC, and Cambricon, reflecting confidence in the sector's growth [2][4] Additional Insights - **Technological Capabilities**: SenseTime's offerings include capabilities in perception intelligence, natural language processing, decision intelligence, and AI-enabled content generation, supported by their SenseCore system [3][4] - **Market Position**: SenseTime is positioned as a leading AI software company, focusing on low-cost and high-efficiency AI solutions [3][4] This summary encapsulates the essential information from the conference call, highlighting SenseTime's strategic initiatives and the overall sentiment in the AI industry.

27 May 2025 | 7:19AM HKT TechNet China 2025: SenseTime (0020.HK) Foundation model introduced; expanding AI-powered user case We hosted SenseTime's management on May 21 at our TechNet Conference China 2025 in Shanghai. Management remains positive on the generative AI trend in China, and highlights their newly launched foundation model, SenseNova V6, carrying upgraded features with competitive costs across training and inferencing. The company also newly signed a MOU with the Faculty of Law at the Chinese Uni ...

- The report introduces the Llama 4 model series, which includes Llama 4 Scout, Llama 4 Maverick, and Llama 4 Behemoth, highlighting their advanced multimodal capabilities and efficiency through the MoE (Mixture of Experts) architecture[10][11][12] - Llama 4 Scout features 16 experts with 17 billion activated parameters, supports a 10M context window, and is optimized for single H100 GPU deployment, achieving state-of-the-art (SOTA) performance in various benchmarks[11][12] - Llama 4 Maverick employs 128 routed experts and a shared expert, activating only a subset of total parameters during inference, which reduces service costs and latency. It also incorporates post-training strategies like lightweight SFT, online RL, and DPO to balance model intelligence and conversational ability[12][14] - The CoDA method is introduced to mitigate hallucination in large language models (LLMs) by identifying overshadowed knowledge through mutual information calculations and suppressing dominant knowledge biases. This method significantly improves factual accuracy across datasets like MemoTrap, NQ-Swap, and Overshadow[23][25][29] - The KG-SFT framework enhances knowledge manipulation in LLMs by integrating external knowledge graphs. It includes components like Extractor (NER and BM25 for entity and triple extraction), Generator (HITS algorithm for generating explanatory text), and Detector (NLI models for detecting knowledge conflicts). KG-SFT demonstrates superior performance, especially in low-data scenarios, with a 14% accuracy improvement in English datasets[45][47][52] - DeepCoder-14B-Preview, an open-source code reasoning model, achieves competitive performance with only 14 billion parameters. It utilizes GRPO+ for stable training, iterative context length extension, and the verl-pipeline for efficient reinforcement learning. The model achieves a Pass@1 accuracy of 60.6% on LiveCodeBench and a Codeforces score of 1936, placing it in the 95.3rd percentile[53][61][64]