Open-Source vs Closed-Source Artificial Intelligence LLMs
Large language models (LLMs) like GPT-4 grabbed headlines in 2023, now there is a trend in 2024 towards smaller, customized open-source models trained on private enterprise data.
These models could potentially match the performance of GPT-4 while being more cost-effective, secure, and optimized for specific enterprise use cases.
As one industry expert stated: “That’s the fork in the road right now, whether you take Llama or Mistral or one of the many other smaller models that are easily available on-prem or put your trust in one of these big boys and use their enterprise solution. There’s still not enough trust in these bigger companies to do better data stewardship.”
Enterprises may shift away from closed-source models like GPT due to cost, security and customization limitations. However, the appropriate model choice will depend on each enterprise’s specific Generative AI (GenAI) requirements.
RAG Techniques and Vector Databases
To improve accuracy and reduce hallucinations, enterprises are exploring techniques like retrieval-augmented generation (RAG) that combine language models with external data sources and vector databases.
RAG allows models to search relevant data before generating an output, enabling verification through embedded citations. Companies like Salesforce and Databricks are deploying vector databases and RAG capabilities into their AI offerings.
As one expert noted: “Salesforce also recently announced an important update to Data Cloud and its Einstein Platform, adding the Data Cloud Vector Database and Einstein Copilot Search… This marks a critical development in Salesforce’s ability to bring quality genAI capabilities to customers.”
These emerging architectures involving RAG and vector databases could help differentiate enterprise AI products in areas like accuracy and data integration.
Addressing Cybersecurity Threats
As GenAI expands in the enterprise, new cybersecurity risks are emerging around model security, insecure code generation, and data governance. Enterprises will need specialized security tools and rigorous model monitoring.
One expert warned: “You’ve got less trained software engineers using code from AI systems, trained on insecure examples, that’s going to create a significant amount of more risk. You’re generating more insecure code from people with less training in the systems.”
Leading cybersecurity platforms like CrowdStrike define AI security as a major new opportunity. Traditional vendors may acquire AI security startups to enhance their offerings in this area.
Private Investment Fueling Innovation
Private investment, especially from large tech companies, is fueling rapid innovation across the GenAI stack beyond foundation models. Areas like cloud security and enterprise AI applications are attracting funding.
As one expert predicted: “Palo Alto and CrowdStrike, two players are in inorganic or acquisition mode. There is a high probability that when a niche player becomes successful, they will go ahead and pick that niche player and integrate them into their solution suite.”
Enterprises will closely watch tech startup innovation around AI security, data integration, customized models and other key areas of enterprise AI enablement.
Safety Protocols and Regulation
High-profile issues like hallucinations could accelerate the need for industry safety standards and potential regulation around enterprise AI deployments. The recent New York Times lawsuit highlighted risks to brand reputation from hallucinations.
As the legal expert stated: “Trademark law is a lot more sound than copyright law, especially when it comes to litigating about it [hallucinations].” Enterprises will scrutinize safety capabilities as they evaluate AI vendors and partners.
In summary, while enterprise adoption of generative AI is still nascent in 2024, several key trends are emerging around model choices, enabling technologies like vector databases, cybersecurity requirements, private investment areas and the need for safety guardrails.
Enterprises have many strategic decisions ahead to securely and effectively unlock AI’s potential. Therefore, I’ll continue to monitor new developments in this emerging AI technology sector.
This article is a synopsis of the report “Generative AI in the Enterprise Software Sector: A Look at Adoption in 2024” from AlphaSense research.
