Half a year ago, due to the high training costs of large models, including R&D personnel, large amounts of data required for training, powerful computing resource support, and time costs, I thought chatGPT might really become the standard for general-purpose large language models in the future. Coupled with its current highest usage, it can feed back model optimization or may form a positive feedback loop, turning it into a moat that other competitors cannot surpass.
Now I have greatly changed my previous thoughts. Although chatGPT has demonstrated good general conversation capabilities and knowledge reserve levels, the development of large models in the future may still be diversified. Large companies and research institutions may continue to develop and launch large models in general domains, but industries may also train and deploy their own large models in specific domains. Different industries and fields will choose a model training strategy suitable for their own goals and business based on their specific needs, data and resource conditions. Such diversity helps meet domain-specific needs and drives innovation and growth across industries.
Let me directly provide some areas/industries that may require you to train large models for analysis:
-
Financial field: The financial field has very high requirements for data security and compliance. Training and using third-party large models may involve sensitive financial data, so financial institutions may be tempted to train customized models themselves to meet specific financial tasks, such as risk assessment, credit risk management, and investment decisions.
-
Legal field: Tasks such as processing legal documents, answering legal questions, and analyzing legal cases require the understanding and application of a large amount of legal knowledge and rules. Self-trained large models can be pre-trained on domain-specific legal data to better understand things like legal terminology, statutory provisions, and precedents. By training large models on its own, the legal industry can achieve more accurate and efficient legal document processing, legal consulting, and legal research.
-
Medical field: Self-trained large models can be applied to tasks such as clinical diagnosis, disease prediction, drug development, and medical image analysis. Medical data has a wide range of diversity and complexity, including medical record data, medical imaging, genomic data, etc. Self-trained large models can be fine-tuned using domain-specific medical data to improve the accuracy and adaptability of the model in diagnosis and prediction. This helps physicians perform more precise disease screening, personalized treatment and drug dosing optimization to improve patient outcomes.
-
Retail and e-commerce: The retail and e-commerce industry has huge product catalogs and user behavior data. In order to improve the effectiveness of recommendation systems, personalized marketing, inventory management, etc., these industries may train large models themselves to better understand and predict user needs and behaviors.
-
Automobile industry: The automobile industry tends to develop in the fields of intelligent driving and vehicle safety. In order to achieve more accurate and reliable autonomous driving and intelligent assisted driving systems, car manufacturers may train large-scale vision models themselves to recognize and understand roads, traffic signs, pedestrians, other cars, etc.
-
Pharmaceuticals and Life Sciences: The pharmaceutical and life sciences fields require processing large amounts of biochemical data and medical images. Customized large models can provide more precise and accurate analysis and predictions for tasks such as drug development, bioinformatics analysis, genomics and more.
-
Aerospace: The aerospace industry handles complex aviation data and flight control systems. Self-trained large models can provide more accurate flight predictions, flight safety analysis and other functions to support flight safety and aviation management.
-
Military industry: Self-trained large models can also play an important role, especially in aircraft combat control. Aircraft combat control requires processing a large amount of complex data, including sensor data, flight parameters, target tracking and weapon systems. The self-trained large model can be fine-tuned using large amounts of combat data to improve the aircraft's automated control, target recognition, flight decision-making and weapon system optimization. By training large models on its own, the military industry can improve the efficiency, accuracy and responsiveness of aircraft in air combat.
The above introduction also includes some multi-module scenarios. Additionally, training large models yourself is particularly important for data privacy and compliance. For example, it follows strict legal and ethical requirements and attaches great importance to the protection of patient and customer data. Training large models on their own enables organizations to better control and protect sensitive data, reducing data sharing and compliance risks.
Overall, self-trained large models have great potential in various industries to promote more accurate, efficient and personalized services. 2023 is the first year of LLM. Now it is like the mobile Internet in 2011. "Standing on the wind, pigs can fly." Best wishes.