Global Software Oracle Corp Rating Outperform Price Target ORCL 225.00 USD The other parts of the announcement that were likely missed by investors is that a) there are multiple other large Cloud contracts already signed in the quarter that are not in RPO or revenue; and b) the multicloud database is expected to grow over100% YoY which will add further growth and help with offsetting margin pressure from the rest of OCI. The contract is likely the much anticipated Stargate contract, but it could be a sovere ...

[Music] Like any driver, autonomous vehicles operate in a world full of unpredictable and potentially safety critical scenarios. NVIDIA Drive, built on the Halo safety system, lets developers build safe autonomous vehicles with diverse software stacks and sensors and redundant computers. It starts with training.Safe AVs need massive amounts of diverse data to be able to address edge cases, but real world data is limited. Developers use NVIDIA Omniverse and Cosmos to reconstruct the real world and generate r ...

IDEAL团队 投稿 量子位 | 公众号 QbitAI 大幅缓解LLM偏科，只需调整SFT训练集的组成。 本来不擅长coding的Llama 3.1-8B，代码能力明显提升。 上海交大&上海AI Lab联合团队提出创新方法 IDEAL ，可显著提升LLM在多种不同领域上的综合性能。 此外，研究还有一些重要发现，比如： 具体来看—— SFT后LLM部分能力甚至退化 大型语言模型 (LLM) 凭借其强大的理解和逻辑推理能力，在多个领域展现了惊人的能力。除了模型参数量的增大， 高质量的数据是公认的LLM性能提升最关键的影响因素。 当对模型进行监督微调（SFT）时，研究人员发现 LLM在多任务场景下常出现"偏科"现象 ——部分能力突出而部分 能力并未涨进，甚至退化。这种不平衡的现象导致大模型在不同的领域上能力不同，进而影响用户体验。 上海交大和上海AI Lab的研究者迅速将目光聚焦到SFT训练的训练集上，是否可以通过调整训练集的组成来缓解LLM 偏科的情况？直觉上来看，直接将LLM的弱势科目的训练数据增加一倍，就可以让最后的结果发生变化。但是，由于 训练数据之间的耦合关系，研究者通过建模量化每个领域数据对于最终结果的 ...

Core Points - Reddit is suing Anthropic for allegedly using its data to train AI models without a proper licensing agreement, claiming this use was unlawful and violated Reddit's user agreement [1][2] - This lawsuit marks Reddit as the first major tech company to legally challenge an AI model provider regarding its training data practices, joining other publishers in similar legal actions [2][3] Company Actions - Reddit has previously established agreements with AI model providers like OpenAI and Google, allowing them to train AI models on Reddit's data under specific terms that protect user interests and privacy [5] - Reddit's legal complaint states that Anthropic's scraper bots ignored the site's robots.txt files, which are intended to prevent automated systems from crawling websites [8] Legal Claims - Reddit is seeking compensatory damages and restitution for the enrichment Anthropic gained from scraping its content, along with an injunction to stop Anthropic from using Reddit's content [9] - Reddit's chief legal officer emphasized the company's stance against profit-seeking entities exploiting Reddit content without compensation or respect for user privacy [4][6]