How did you miss our Gen AI webinar featuring an Amazon Web Services (AWS) Solutions Architect and Head of Strategy, plus our Co-Founder and CEO, Ami Daniel? Since we know you’re busy, you can watch it here at your convenience, or quickly check out these three key takeaways:
Foundations FIRST
“When you start speaking about gen AI, it’s all about the data,” said Eric Topp, Head of Worldwide Solutions & Strategy, Transportation & Logistics, AWS. “There’s this big rush to do something with gen AI from leadership, but if you don’t have the right foundation and cleanse your data, harmonize it, and hydrate your data lake properly, you can take a really good gen AI application and turn it into something that’s worthless, because your data is not good. You can ask a question, and you’ll get an answer. It just might not be the right answer because the data is incorrect.”
Don’t Worry About Unstructured Data Disarray
Won’t the unstructured data issue prevent Gen AI from truly impacting the maritime and supply chain industries?
“Gen AI models can work with most structured and unstructured data. Structured data is like what you get from a database, and it’s well-defined. Unstructured data is just text documents, contracts…both of them can be used to enhance gen AI and get better results,” said Vadim Tereshchuk, Solutions Architect at AWS.
“If we take unstructured data, we add it to our vector database and integrate this vector database with our gen AI model using embedding models, then we can filter and extract relevant data from RAG (Retrieval Augmented Generation), from those text documents previously unstructured, and augment our prompts with it and get better results from our LLM model. This is a great way to use unstructured data.”
All speakers at the webinar stressed the need for human oversight to manage AI reliability and prevent issues such as data hallucination. They also highlighted the role of partnerships with AI specialists, such as Windward, in effectively navigating these complexities.
Fast or Slow Adoption?
When advising clients, Eric recommends a cautious approach to adopting gen AI. Starting small and scaling gradually allows companies to integrate AI effectively, without overwhelming their existing systems. Despite the excitement around gen AI, many clients are still focused on traditional AI and machine learning, underscoring the importance of a strategic implementation of new technologies and partnering with industry leaders.
Ami detailed Windward’s approach to data curation. It emerged from the realization that automating everything was not only costly but also lengthy, due to numerous edge cases. Initially, the company aimed for complete automation, but practical challenges, such as inconsistent data from carriers, led to customer dissatisfaction and churn.
This epiphany led Windward to emphasize the importance of curated data, combining domain knowledge and advanced technology. Instead of relying solely on raw data, Windward uses labeled datasets, customer data, and quality assurance processes to train our vertical AI models. This method ensures higher accuracy and reliability, meeting the detailed demands of the maritime industry.
Bonus Tip: Here’s What Not to Do…
Vadim explained why effective gen AI results for maritime and supply chain aren’t going to come from a plug-and-play approach during Windward’s webinar: “Gen AI foundation models aren’t a one-size-fits-all solution. Sometimes, just adding your data to the models gives great results. Sometimes, it’s not enough, and we need to fine-tune the models or continuously pre-train them to greatly improve our results.
To train gen AI models, supply chain companies must standardize the multiple use cases pertinent to their organization and access hundreds of unique data sources, some of which are traditional, outdated, and difficult to comprehend.
I Want to Talk to the AI Experts
