Suggestions
Edy Liongosari
Chief Research Scientist | Fortune Connect Fellow | Speaker at Accenture Labs
Edy S. Liongosari is a prominent figure in the field of technology and innovation, currently serving as the Chief Research Scientist and Managing Director at Accenture. He holds significant roles in various organizations, including being the Chair of the Thought Leadership Task Group and the Lead Editor of the Journal of Innovation at the Industrial Internet Consortium (IIC).
Liongosari has a strong academic background, having studied at Indiana University Bloomington. He is recognized for his contributions to the field, including his involvement with the World Economic Forum on topics related to the business of data. His expertise spans across various domains, particularly in the intersection of technology and business innovation, making him a key thought leader in the industry.12
Highlights
Friday Posting on #GenerativeAI: The Rise of Specialized AI Models
While my previous posting (1) focused on the synergistic relationship between humans and multiple AI models, we are also observing the rise of highly specialized pre-trained AI models. Much like humans with niche expertise, these domain-specific models excel in their specific domains, standing in stark contrast to the earlier belief that domain-agnostic AI models, such as GPT, could serve as jacks-of-all-trades. While domain-agnostic AI models work well in many circumstances, they may lack the nuance needed for specialized tasks.
A case in point: those who used GPT for stock selection and investment advice have reported that the outcomes were mixed at best (2). Successful financial investing requires a nuanced understanding of historical and current market conditions, sentiment, and timing—elements that generalized AI models are not equipped to address. Specialized models like BloombergGPT (3,4) and FinGPT (5) are designed for these particular tasks, incorporating investment-relevant data, sentiment analysis, and temporal sensitivity to offer actionable investment insights. It's hardly surprising that financial institutions such as JPMorgan Chase are also investing in these specialized models (6).
The application of specialized models extends beyond the financial services industry. For instance, BioGPT (7) and PubMedBERT (8) are tailored specifically to support biomedical research. Numerous other models focusing on areas like industrial systems and supply chain management are also in development. As people gain a clearer understanding of AI capabilities and how to leverage them, the demand for such specialized AI models is expected to grow.
This trend towards specialization is not just promising in terms of performance; it may also help make progress in addressing ethical concerns. Tailored models allow for the incorporation of industry-specific ethical guidelines and can undergo more rigorous vetting. For example, financial advice generated by specialized models can be better ensured to comply with investment regulations. Specialization thus allows companies to tackle ethical issues more effectively, aligning with societal values, mitigating bias and discrimination, and enhancing the user experience.
#ai #genai #chatgpt #opportunity #ChatGPT #GPT #bard #llm #llms #foundationmodels #workforce All views expressed in this article are the author's own.
References: (1) https://lnkd.in/gFMzUa3S (2) https://lnkd.in/g--Bx93u (3) https://lnkd.in/gEBmeZCU (4) https://lnkd.in/gkF-fGXq (5) https://lnkd.in/gkTbtdbG (6) https://lnkd.in/gajG5Dyu (7) https://lnkd.in/gDwQfS8h (8) https://lnkd.in/gN8CB7uB
Friday Posting on #GenerativeAI: Human + Multiple AI Models to form Agile of Workforce
When ChatGPT made its debut late last year, it was met with high expectations: a highly capable AI model that would be articulate, accurate, fair, creative, and all-encompassing - essentially, a perfect model. However, the reality that emerged in the subsequent months was quite different. This situation did not change after the introduction of subsequent models like BARD and Llama.
Many reports, including one from Stanford (see https://lnkd.in/gPGyGNm6) underscore that no single AI model, including GPT, embodies all of the desired characteristics. The key, then, lies in a keen understanding of each model's strengths and weaknesses. A recent report by AI performance company Arthur AI (see https://lnkd.in/gDN5S2pq) drives this point home. It compares several commonly used models and provide simple characterization of their strengths and weaknesses to offer guidance on which model to use when.
The report highlights that different tasks may require different models, much like how different people have different competencies and talents. For instance, it points out that GPT-4 excels at math, Anthropic's Claude 2 is superior in knowing the limits of its knowledge, and Cohere AI stands out in terms of creativity. Admittedly, such characterization is a gross simplification but valuable nonetheless for shifting the mindset toward using multiple models depending on the task at hand. For instance, a highly creative model could be chosen for writing fiction, while a model with greater accuracy might be preferred for drafting legal documents. Such characterization may also help model developers in strengthening the models in a specific fashion.
Furthermore, these models can be used in tandem, leveraging their unique strengths (e.g., one to create and another one to validate). Human intervention remains crucial in orchestrating the interactions between these models and bridging the gaps where they fall short. This blended approach, combining the capabilities of AI models and human expertise, could shape the workforce of the future, offering an agile, adaptable, and efficient way to tackle a broad spectrum of tasks.
#ai #genai #chatgpt #opportunity #ChatGPT #GPT4 #bard #llm #llms #foundationmodels #workforce All views expressed in this article are the author's own.
