TrainAI® by RWS has officially launched a complete suite of AI data services to help organizations build ethical, accurate and reliable generative AI (GenAI) applications, mitigating the risk of bias or hallucinations in their GenAI innovations. The new service follows a successful year for TrainAI, which now works with four of the world’s top five technology companies and continues to innovate and develop AI data training services for clients.
“Developing and implementing GenAI requires vast volumes of training data, as well as significant time and human expertise to fine-tune them – resources that companies often don’t have readily available in-house,” commented Vasagi Kothandapani, President of Enterprise Services at RWS. “TrainAI’s suite of services is designed to fill this gap by providing the deep expertise required to prepare AI training and fine-tuning data in a responsible and ethical manner.”
The TrainAI GenAI service suite harnesses the knowledge of vetted, qualified, locale-specific domain experts from the TrainAI community to engineer prompts, refine responses, verify facts and perform red teaming. TrainAI’s GenAI services include:
Domain expertise: Projects are triaged to a 100k+ strong TrainAI community, which includes subject-matter experts at all educational levels with experience across a broad range of topics and industries.
Prompt engineering: Experienced professionals from the TrainAI community provide services including prompt-based learning, prompt design, prompt tuning, advanced prompting and safety alignment to mitigate the risks of generating harmful, offensive or inappropriate outputs.
Reinforcement learning from human feedback (RLHF): A broad range of RLHF services, including response rating and editing, and fact extraction and verification, ensure that GenAI can understand what humans want and provide relevant, reliable, safe and accurate responses.
Red teaming and jailbreaking: Risk mitigation services, such as red teaming or jailbreaking, involve intentionally creating prompts that can cause the model to hallucinate or generate potentially harmful content, helping to uncover and address potential vulnerabilities in GenAI.
TrainAI is already helping the world’s largest organizations train and fine-tune their GenAI. One client – a US multinational technology conglomerate – turned to TrainAI to help differentiate its open-source large language model (LLM) by fine-tuning it with content prepared by domain experts from the TrainAI community. TrainAI rapidly onboarded, trained and deployed more than 200 domain experts who completed over 32,000 hours of work within the first three months of the project. As a result, the client was able to successfully launch the latest version of its LLM and TrainAI’s work on the project continues today.
The TrainAI GenAI suite of services enables companies to respond to several challenging themes identified in RWS’s recent ‘Genuine Intelligence™, the future of machine-human collaboration’ report. As Melanie Peterson, Programme Director at TrainAI, explains, “Most problems with AI, such as hallucinations and biased results, result from the lack of explainability of many opaque methodologies. The responsible, rigorous and ethical approach to AI development we build through Genuine Intelligence provides far more solid AI foundations.”
