A Brewing Storm in Generative AI

May 17, 2024
Generative AI | 7 min READ
Generative AI has sparked intense debate amongst technology leaders since the introduction of ChatGPT by OpenAI at the end of 2022. While some dismissed it as mere hype that would be of little use to enterprises, others considered it revolutionary, akin to the internet in the 1990s.
Sudhir Parwal
Sudhir Parwal

Global Client Partner


Nitin Jindal
Nitin Jindal

Global Digital Partner for Life Sciences Business


Recent forecasts from IDC suggest that generative AI (GenAI) is unarguably more than hype, with projected spending reaching $143bn by 2027.1 What's exciting is how GenAI is evolving for rapid adoption in the enterprise environment.
New announcements from OpenAI, like the GPT Store and Assistants API, suggest a significant shift, indicating that the entry barrier for scaling high-value use cases is diminishing. This will leave a decisive mark on the competitive landscape. Technology leaders want to become the first movers to gain a decisive advantage.
As the strategy for fostering enterprise wide GenAI adoption changes, let us explore how this shift can be exploited in your organization's favor.
Revisiting the traditional GPT application framework
While the first generation of GenAI solutions frameworks offered powerful capabilities, building practical use cases for enterprise environments proved challenging.
The traditional Generative Pre-trained Transformer (GPT) use case development pattern needed to be simplified. It required extensive development expertise to build custom use cases beyond applying the same core large language model for various application scenarios. Typically, documents were parsed and chunked to precompute embeddings, then stored in a database. This database was a foundation for semantic models that powered generative or semantic search applications.
Stay Ahead
Visit our Generative AI page
These GPTs were typically utilized for use cases like Q&A chatbots, translation, summarization, content creation, and code generation.
Limitations of traditional GPT application framework
  • Customization Complexity: Building a custom GPT that was finetuned for domains required considerable AI engineering expertise, specialized knowledge, and time and resource investment.
  • Response Variability: Traditional GPTs often delivered variable responses to the same queries, rendering them unreliable for implementation in enterprise apps and workflows.
  • Prompt Engineering Challenges: The efficacy of a GPT in handling end-user queries strongly relied on prompt engineering proficiency, which was a clunky and time-consuming process.
  • Hallucination Risks: Traditional GPTs were prone to hallucination, which further degraded their applicability in sensitive or critical scenarios.
However, traditional GPT development paved the way for realizing the immense value that GenAI could bring to enterprise applications.
A paradigm shift in GenAI: key advances shaping the future.
The latest developments introduced by OpenAI have eliminated the above challenges and redefined the rulebook for prototyping and scaling GenAI use cases. In addition to reduced time-to-value, these innovations have also de-risked GenAI experimentation and implementation initiatives.
Moreover, low-code approaches to building custom GPTs are rapidly democratizing access to the technology, priming the digital ecosystem for a knowledge and creativity revolution.
The following latest announcements from OpenAI sum up the key elements that are driving this paradigm shift in GenAI technology:
GPT-4 Turbo
GPT-4 Turbo is a multi-modal model that succeeds GPT-3.5 and significantly outperforms the latter on nearly every metric. Due to increased context length, it can take more nuanced instructions from users and deliver more consistent responses to the same queries. It offers better control over its output, and with API customization, it can be embedded across a broader spectrum of use cases. It seamlessly integrates with existing workflows through APIs , facilitating deployment across diverse applications.
GPTs and GPT Store
In addition to the acronym, GPT refers to customized LLM versions. Users with natural language inputs can program these customizations with nearly zero coding skills. Custom GPTs open Gen AI for more targeted applications where general-purpose models fall short.
These GPTs can be published on the GPT Store, where they can be monetized by creators and consumed by the users. Experimenting with this capability early on, Birlasoft developed a PubMed GPT, which can synthesize information from millions of biomedical papers to deliver insightful responses to users.
Assistants API
The Assistants API is a powerful tool that allows developers to create customized AI assistants that make it possible to manage the state of conversational threads with users, respond to their queries using a GPT, and trigger downstream actions with function-calling capabilities. It enables users to embed GenAI capabilities using APIs and offers a low-code approach to building such use cases. Further, introducing Persistence within conversation threads is a game-changer, as its absence severely limited previous ChatGPT versions. This allows for better contextual understanding, continuity, complex interactions, and personalization.
Implications for enterprises
So, how do these developments influence the value proposition of GenAI in the enterprise landscape? Here are a few ways:
  • GPT-4 Turbo is more reliable and consistent, making it viable to implement in enterprise mission-critical processes. This represents a cost-saving opportunity by reducing manual intervention and automating repetitive tasks. Some use cases will significantly elevate the employee and customer experience.
  • The ability to build custom GPTs opens the technology up to domain-specific applications. The ability to program GPTs without extensive coding expertise and reduce the effort required to make such capabilities truly democratizes access to AI technology.
  • Finally, the Assistants API will enable faster prototyping and solution development. It will enhance the productivity of GenAI developers and allow enterprises to embed GenAI capabilities into existing apps without having to train, run, and manage the underlying models powering them.
Lead the change: fostering GenAI adoption at the enterprise level.
Over the last two years, enterprises have increasingly focused on identifying adoption opportunities for GenAI. While stabilizing and optimizing existing models was once imperative, the advent of GPT-4 Turbo has streamlined these aspects.
As enterprises embrace Gen AI, it is high time to devise a deliberate strategy for widespread adoption. OpenAI's low-code GPT-building capability empowers business users. While the idea of empowering users to build and implement GenAI directly may be alluring, it will be advisable for technology leaders to be prudent and proceed with caution.
Generative AI is a big data technology, and as a result, it remains subject to all the considerations that must be made when leveraging such technologies at the enterprise level. While OpenAI's recent innovations significantly reduce the effort that goes into building, adapting, and scaling GenAI use cases, enterprises must still address the following key factors:
  • Infrastructure considerations: When implementing a GPT-powered use case in production, enterprises must still determine the ideal deployment infrastructure. Compliance requirements may restrict some organizations in highly regulated industries from deploying GPTs on the cloud, and identifying the most optimal price-performance options will be essential to minimize running costs.
  • Securing GPT deployments: LLMs, including GPTs, remain susceptible to various security risks. Even if some applications leverage OpenAI's GPT API, several responsibilities will stay with users, such as meticulously configuring environment variables, as misconfiguration can lead to vulnerabilities, storing API keys securely, and precisely defining API key permissions. Moreover, securing the servers interacting with the API will also rest with the users.
  • Preventing context leakage: Custom GPTs have been reported to leak context data, which has been used to train it if the user queries it. In some cases, custom GPTs have also readily made training files available for download. These files are likely to contain confidential information such as identities or financial data, making mitigating context leakage risks essential. It is, therefore, imperative to ensure proactive data sanitization, finetuning, access control, and implementation of security protocols and guardrails.
  • Running and maintenance: When organizations implement their GPTs, they must maintain their deployments. This will involve setting up the necessary infrastructure and considering factors like scalability, server availability & resource allocation. This also means monitoring & addressing any issues that may crop up in the GPT's behaviour and performance. Finetuning and continuous adaptation as real-world data accumulates is also a crucial need to improve the accuracy & relevance of the model.
Business users may need more skills to address the above considerations. Thus, if your organization is undertaking GPT use-case development and deployment without support from technology partners, driving close collaboration between technical teams and business users will remain a crucial consideration to prevent GenAI disasters.
That said, most organizations will benefit from teaming up – with a technology partner deeply invested in the GenAI ecosystem and has already delivered high-ROI use cases to their clients. Such a partner will prove valuable in accelerating adoption and bring invaluable insights that your internal technical teams may lack. Significantly, technology partners can help you align technical and business teams to run and maintain GPT deployments across the enterprise.
What is coming next?
With the latest announcements from OpenAI, the GenAI landscape has taken on a transformative turn. Customization is much easier and more accessible, and custom GPTs, along with the Assistants API, offer enterprises new opportunities and deployment options. One thing can be said with certainty – a knowledge revolution is in the making, and new GenAI use cases will have a marked impact on competitive factors.
While these capabilities may have democratized access to GenAI, successful enterprise adoption will still be determined by crucial considerations like infrastructure and deployment, security strategies, and ongoing upkeep. To score a definitive win, businesses should strongly consider collaborating with technology partners that have already delivered success with GenAI and those who bring invaluable insights and robust practices.
Was this article helpful?