OpenAI Launches GPT-4o Fine-Tuning; Qualcomm’s New AI-Focused Mid-Range Chip; Kioxia Prepares for IPO; Nvidia Unveils Mistral-NeMo-Minitron 8B; SK hynix and Samsung Focus on Customized HBM4 Solutions

AI Brief Headlines

• OpenAI Launches GPT-4o Fine-Tuning
• Qualcomm Unveils Snapdragon 7s Gen 3 with Enhanced AI Capabilities
• Kioxia Holdings Plans October IPO Amid Strong Chip Demand
• Chinese Entities Use AWS Cloud to Access Restricted U.S. AI Chips
• Nvidia and Mistral AI Launch High-Accuracy, Low-Footprint Language Model
• Customized HBM to Become a Reality in the HBM4 Generation

OpenAI Launches GPT-4o Fine-Tuning

OpenAI now allows third-party developers to fine-tune GPT-4o, its advanced large multimodal model (LMM), enabling custom modifications for specific application needs. Fine-tuning enhances performance, especially in AI tasks like coding and creative writing, and is available with free tokens until September 23, 2024. This move aims to make AI more accessible and customizable for businesses, while emphasizing safety and data privacy. The initiative follows successful tests with partners, highlighting the potential of fine-tuned models across various industries.

[embed]OpenAI brings fine-tuning to GPT-4o with 1M free tokens per day through Sept. 23 The company notes that developers can achieve strong results with as few as a few dozen examples in their training…venturebeat.com

Qualcomm Unveils Snapdragon 7s Gen 3 with Enhanced AI Capabilities

Qualcomm introduces the Snapdragon 7s Gen 3, a mid-range chip promising significant CPU and AI performance improvements, targeting smartphone users with on-device AI needs. The chip enhances graphics, supports up to 200 MP camera sensors, and facilitates 4K HDR video capture. Xiaomi will debut the chip in its upcoming release, followed by Samsung and Realme. This move aligns with Qualcomm’s strategy to bring AI capabilities to affordable smartphones globally

[embed]The Qualcomm Snapdragon 7s Gen 3 Is All About On-Device AI The chip will debut first overseas on a new Xiaomi phone next month.gizmodo.com

Kioxia Holdings Plans October IPO Amid Strong Chip Demand

Japanese chipmaker Kioxia Holdings Corp. has applied to list its shares on the Tokyo Stock Exchange, aiming for a market capitalization of 1.5 trillion yen ($10.3 billion) in what could be 2024’s largest IPO. The company, which saw record sales driven by AI-related chip demand, had previously postponed its listing due to U.S.-China tensions. Kioxia may resume merger talks with Western Digital after the listing, which could reduce the influence of major investor SK Hynix. The global semiconductor market is expected to continue growing in 2025.

[embed]Japanese chipmaker Kioxia aims to list in October amid AI boom TOKYO -- Japanese chipmaker Kioxia Holdings filed an application to list on the Tokyo Stock Exchange on Friday, Nikkei…asia.nikkei.com

Chinese Entities Use AWS Cloud to Access Restricted U.S. AI Chips

Chinese entities are leveraging cloud services, including Amazon Web Services (AWS), to access advanced U.S. AI chips and models restricted by export laws. Despite U.S. regulations, accessing these technologies via the cloud is not prohibited. At least 11 Chinese organizations have sought such services, often through intermediaries. The U.S. government is considering tightening regulations to address this loophole.

[embed]Exclusive: Chinese entities turn to Amazon cloud and its rivals to access high-end US chips, AI State-linked Chinese entities are using cloud services provided by Amazon or its rivals to access advanced U.S. chips…www.reuters.com

Nvidia and Mistral AI Launch High-Accuracy, Low-Footprint Language Model

Nvidia and Mistral AI have introduced Mistral-NeMo-Minitron 8B, a new small language model with 8 billion parameters, optimized through pruning and distillation. This model offers “state-of-the-art” accuracy while being compact enough to run on consumer hardware like laptops and PCs, reducing costs and compute power by up to 40x

[embed]Nvidia and Mistral AI's super-accurate small language model works on laptops and PCs Nvidia and Mistral AI have developed a highly optimized language model that can work locally on laptops and PCs while…www.tomshardware.com

Customized HBM to Become a Reality in the HBM4 Generation

SK hynix announced its plan to develop next-generation High Bandwidth Memory (HBM) products with 20 to 30 times the performance of current offerings. These advancements, targeted for AI applications, are being pursued in collaboration with TSMC for the upcoming HBM4 generation, expected by 2026. The move reflects a growing trend towards customized HBM solutions, with major tech companies like Apple, Microsoft, and Nvidia seeking tailored memory products. This shift is expected to reshape the HBM market towards more specialized production.

[embed][News] HBM Technological Battle to be on the Next Level | TrendForce Insights Per a report by BusinessKorea, SK hynix Vice President Ryu Seong-su announced the company's strategic plan in the HBM…www.trendforce.com