Thema

主题

• 批准号: 10064969
• 金额: $ 5.78万
• 依托单位: THEMA AI LIMITED
• 依托单位国家: 英国
• 项目类别: Collaborative R&D
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=10064969
• 关键词: Thema

Recent advancements in vector embeddings and large language models (LLM) have opened-up huge new opportunities to power next-generation market discovery and search. Thema are developing game-changing trustworthy AI to unearth deep-insights from online text sources, enabling SME private market investors to significantly improve market discovery/company evaluation, ROI and productivity.Current market solutions have not evolved in the last decade. Market-mapping currently relies on expensive/unscalable/static consulting services (c£500k for one-off 6-week project), is inherently human biased/constrained, and leads to missed opportunities ?as themes evolve rapidly. Existing software/tooling (analyst-driven company intelligence tools) is built on legacy search technology which reinforces bias/lacks contextual understanding, thus stifles discovery. Public search engines (Google/Bing) serve only general information discovery, whilst the nearest state of the art have poor results and rely heavily on augmenting their tools with biased consulting services.Thema have built a highly innovative MVP that offers novel market discovery engine, proving the core technological assumptions leveraging pre-trained open-sourced large language model and tested the MVP with top PE/VC investors, validating the need/demand for the innovation. In order to develop a fully transparent and trustworthy scalable solution, Thema need to increase the breath of data and advance the technology to (1) Add additional company description sources to ingest a variety of lenses that describe what a company does (2) train and evaluate a proprietary custom embedding model instead of using black-box pre-trained models to improve transparency of results to end-users (3) create and evaluate active learning pipeline to optimise unsupervised learning results and capture subjectivity of individual customers views on markets and competition.Success with the project will give us first-to-market advantage with a groundbreaking AI market intelligence tool. PE/VC/Hedge Funds will adopt the innovation due to its unique value-proposition: enables investors to invest earlier (increased returns); find better quality companies (higher exit rate/returns on exit); increased productivity: 35% in time and deliver 25% increase in productivity: (Cipher 2020, study on competitor intelligence tooling); deliver ROI of 1.5x better exits based on 4x discovery and 2x evaluation process improvement.Innovation is the leading driver of economic-growth (_OECD-2017_). By transforming the market and deal discovery we will enable private market investors to discover a broader pool of entrepreneurs/companies, directly impacting overall economic-growth at every industry-scale.

向量嵌入和大型语言模型（LLM）的最新进展为下一代市场发现和搜索提供了巨大的新机会。Thema正在开发改变游戏规则的值得信赖的人工智能，从在线文本来源中挖掘深刻的见解，使中小企业私募市场投资者能够显著提高市场发现/公司评估，投资回报率和生产力。市场地图目前依赖于昂贵的/不可扩展的/静态的咨询服务（一次性6周项目约50万英镑），本质上是人为偏见/限制，并导致错过机会？因为主题发展迅速。现有的软件/工具（分析师驱动的公司智能工具）是建立在传统的搜索技术上的，这种技术强化了偏见/缺乏上下文理解，从而扼杀了发现。公共搜索引擎（Google/Bing）只提供一般的信息发现，而最先进的技术效果不佳，严重依赖于通过有偏见的咨询服务来增强他们的工具。Thema建立了一个高度创新的MVP，提供新颖的市场发现引擎，利用预先训练的开源大语言模型证明了核心技术假设，并与顶级PE/VC投资者测试了MVP，确认创新的需求。为了开发完全透明和值得信赖的可扩展解决方案，Thema需要增加数据的呼吸并推进技术，以（1）添加额外的公司描述源，以摄取描述公司所做的各种镜头（2）训练和评估专有的自定义嵌入模型，而不是使用黑盒预先训练的模型，以提高最终用户的结果透明度（3）创建和评估主动学习管道，以优化无监督学习结果，并捕捉个人客户对市场和竞争的主观看法。该项目的成功将为我们提供开创性的人工智能市场情报工具，使我们具有率先进入市场的优势。PE/VC/对冲基金将采用创新，因为其独特的价值主张：使投资者能够更早地投资（增加回报）;寻找更优质的公司（更高的退出率/退出时的回报）;提高生产率：35%的时间，并提供25%的生产率提高：（Cipher 2020，竞争对手情报工具研究）;基于4倍的发现和2倍的评估流程改进，实现1.5倍的投资回报率。创新是经济增长的主要驱动力（_OECD-2017_）。通过改变市场和交易发现，我们将使私人市场投资者能够发现更广泛的企业家/公司，直接影响每个行业规模的整体经济增长。