Evangelos Simoudis
Evangelos Simoudis, with over 30 years of experience in Silicon Valley, is a seasoned venture investor, senior advisor to global corporations, and a leader in AI, new mobility, and corporate innovation. Prior to his 25-year investing career, he spent close to 20 years in high-technology industries, holding executive roles in operations, marketing, sales, and engineering. Currently, Simoudis serves as the co-founder and managing director at Synapse Partners, where he continues to invest in early and growth-stage companies developing enterprise AI applications and advise corporations and governments on how to apply AI. Simoudis is the author on three best-selling books on AI and new mobility, including the recently published The Flagship Experience: How AI and Software-Defined Vehicles Will Revolutionize the Automotive Customer Experience.
In an exclusive interview with StartupCity Latin America, Simoudis shares his valuable insights on the importance of thorough understanding and strategic planning in developing AI solutions.
What are some of the notable advancements or breakthroughs in generative AI that you find particularly exciting or promising?
Generative AI is having a significant impact on the business world. An area that particularly excites me is the use of generative AI for code generation and code reengineering. Over the years, enterprises have collected vast amounts of legacy code, which often requires updating to modern languages to align with transforming hardware and computing models. We are currently investing in how generative AI, in addition to code conversion can be used to optimize the generated code for the new target computing environment. This involves introducing optimization techniques during translation and showcasing novel applications of generative AI.
Another area we are exploring is the use of generative AI for process reengineering within enterprises. As companies explore new products, markets, and customers, generative AI systems offer fresh insights to enhance and adapt existing processes. Recently, multimodal models like Google’s Gemini have emerged, capable of handling multiple data types simultaneously. This presents a significant opportunity for enterprises, allowing fine-tuning of models using diverse data, making them more applicable across various tasks and domains.
“As part of our work in generative AI, we assist enterprises in analyzing costs, evaluating potential returns on investment, and assessing different cost structures.”
A few areas where generative AI is starting to be applied hold promise for life-changing breakthroughs. For example, the use of generative AI in biotechnology, particularly in drug discovery, is an area that intrigues me greatly. The biotechnology domain shares many of the language-like characteristics we see in text generation and computer programming. This makes it a very good target for foundation models and other specialized Large Language Models. Leveraging these models to accelerate drug discovery and navigate regulatory approval could lead to groundbreaking advancements, similar to those we’ve seen with the rapid development of COVID-19 vaccines, particularly mRNA vaccines. This holds immense promise for addressing various diseases in the future.
What challenges or obstacles do you face when implementing generative AI solutions in business, and what steps do you take to address them?
We see enterprises facing three challenges: the cost of building and using such systems, data ownership, and people.
Inferencing costs of publicly available models are high. Custom or even customized generative AI models are expensive to develop. The investment required to develop a fully custom Large Language Model can be substantial. Making these systems accessible by customers or employees, especially in large-scale deployments, significantly adds to the overall cost. In fact, in several instances, the inferencing phase of a large language model can be more expensive than the training phase. As part of our firm’s corporate advisory work, we assist enterprises in analyzing the costs associated with their generative AI existing and future efforts, evaluating potential returns on investment, and assessing different cost structures.
Data ownership and provenance are critical concerns for enterprises. Given the growing emphasis on data privacy, security, and copyright infringement companies are increasingly cautious about the sources and data used to train AI models. They seek assurance that the data was obtained legitimately and remains untampered. Enterprises also express concern about legal ramifications when proprietary data is utilized to train large language models without adequate authorization. Similarly, content providers are concerned that their proprietary data was used in model training without consent.
Even though over the years several corporations have hired employees with AI backgrounds, many still lack the requisite talent and the critical mass to develop solutions that will impact their operations. Generative AI has made this problem more acute. We are currently helping several corporations formulate their AI strategy and create hiring plans to enable them to implement such strategies without the continuous use of outside consultants.
Addressing these challenges is vital for the continued growth of generative AI in enterprise environments. We actively work to tackle these issues and ensure the responsible and ethical deployment of AI systems within enterprises.
How do you approach ethical considerations and potential risks associated with the use of generative AI in various applications?
As a community, we eagerly embraced large language models in late 2022 without fully understanding their capabilities and the implications of their deployment. This haste has resulted in unexpected challenges from these systems, leading governments to draft regulations without a deep understanding of what they are regulating. Both corporations and governments must invest more time in comprehending the real risks associated with these systems’ capabilities.
There is a tendency, fueled by marketing efforts, to exaggerate the true capabilities of these AI systems. Corporations may overstate these capabilities to promote their products, while governments might either overregulate or overlook critical areas that require attention. To ensure the ethical and lawful deployment of AI systems, it is essential to strike a balance and devote sufficient time to understanding the risks involved.
The elections in India as well as the upcoming elections around the world, such as those in the European Union and the United States, highlight the risks posed by generative AI, particularly regarding deceptive or misleading content generation. Both developers and users in corporations, along with governments, need to prioritize understanding these risks and crafting appropriate regulations. Currently, there is a discrepancy where some areas lack regulation while others are overly regulated, merely to fulfill regulatory obligations without fully addressing the underlying issues.
What advice would you give to aspiring entrepreneurs or developers looking to explore Generative AI and its potential applications?
Over the past 18 months, many entrepreneurs have embraced ideas, prototypes, and systems that utilize large language models and exhibit some form of generative AI. However, several of these endeavors have fallen short. As investors, we have observed instances where we have rushed to invest in these ventures, only to realize later their limited longevity and vulnerability to competition from tech giants like Google, Microsoft, Amazon, and Meta. Therefore, when advising entrepreneurs, we emphasize the importance of developing AI solutions, especially for enterprises, with enduring value and substantial barriers to entry. This requires a deep understanding of enterprise processes, the value proposition of large language models, and the additional capabilities needed to complement them effectively.
One of our portfolio companies initially planned for the large language model embedded into their enterprise application to handle 80 percent of their application’s workload. During the pre-investment collaboration our firm established with the company’s founders we convinced them that for the application to have enduring value additional AI components besides the large language model were necessary. By the time we funded them the model now was accounting for 20 percent of the application’s workload—a testament to the strategic thinking required for such endeavors.
I would also like to mention that we continue to see many startups that want to build foundation or other general-purpose large language models. We advise them to recognize the immense cost and consider whether this aligns with their company’s goals. Alternatively, focusing on developing industry-specific solutions tailored to niche markets can yield more impactful results. We continue to advocate for the development of autonomous agents, i.e., intelligent systems with planning, memory, and learning capabilities. Although challenging, the potential value of such systems is immense, whether for physical or software applications.
