AI in Business: Investments and Opportunities for Value Creation

Key Concepts

• Deep Tech Venture Capital: Investing in companies with significant technological innovation, often hardware-related.
• AI Domain Expertise: Applying AI to specific business sectors rather than developing general AI tools.
• Bioconvergence: The intersection of biology and hardware, particularly in medical diagnostics and drug development.
• Hardware Renaissance: Renewed interest and innovation in hardware due to the demands of AI.
• AI Overhype vs. Transformation: Recognizing AI's transformative potential while acknowledging unrealistic expectations.
• Data as a Limiting Factor: The availability and quality of data as a constraint on AI development.
• Hallucinations: Inaccurate or nonsensical outputs from AI models due to insufficient training data.
• Chiplets: Breaking down large, complex chips into smaller, more manageable components.
• Foundational Models: Pre-trained AI models that can be adapted for various tasks.
• Innovator's Dilemma: The challenge for large companies to embrace disruptive innovations.

Celesta Capital's Investment Focus

Celesta Capital, a global deep tech venture capital firm, focuses on three main areas:

1. Hardware (Approximately 50%): This includes investments in companies involved in the design, development, and manufacturing of hardware components and systems. This area is experiencing a resurgence due to the demands of AI and high-performance computing.
2. AI Domain Expertise (Approximately 25%): Celesta invests in companies that apply AI to specific industries and business functions, such as radiology, legal services, and drug development. They avoid investing in general AI tools like large language models (LLMs).
3. Bioconvergence (Approximately 25%): This area focuses on the intersection of biology and hardware, with a significant emphasis on medical diagnostics and innovative drug development.

Staying on Top of Technology Trends

• Silicon Valley Ecosystem: The concentration of technology expertise, interaction, and accessibility in Silicon Valley provides a unique advantage for staying informed about technology trends.
• Advisors and Portfolio Companies: Celesta leverages a network of expert advisors, including Nobel laureates, and the CEOs of its portfolio companies to evaluate new technologies and assess their potential.

Evaluating Entrepreneurs and Technology

• Technology Assessment: Celesta evaluates the technical feasibility and potential of new technologies.
• Entrepreneurial Assessment: The firm also assesses the capabilities and track record of the entrepreneur and their team, recognizing that business execution is often a greater challenge than technological innovation.
• Replacing Entrepreneurs: In some cases, Celesta will invest in a company only if the entrepreneur agrees to step aside and allow the firm to hire a more experienced CEO.

Intel's Challenges and Lipu's Role

• Intel's Struggles: Intel has faced challenges in processor technology and foundry capabilities, leading to a decline in its market position.
• Lipu's Appointment: Lipu, a partner at Celesta Capital and former CEO of Cadence, was appointed as the CEO of Intel to address these challenges.
• Partnerships: Lipu's experience and relationships with other technology companies, such as Broadcom, Qualcomm, and TSMC, are expected to be crucial for Intel to forge partnerships and regain its competitive edge.
• Integration of Process and Design: The decision of whether to continue integrating process and design within Intel is a critical strategic question that Lipu will need to address.

AI: Overhype and Transformation

• AI's Transformative Potential: AI is recognized as a transformative technology with the potential to revolutionize various industries and aspects of life.
• AI Overhype: There is a tendency to overestimate the short-term capabilities of AI and make unrealistic projections about its impact.
• Drug Development Example: The claim that AI can rapidly develop individualized drugs is considered overly optimistic, with experts suggesting that such advancements are at least three decades away.

AI in Biology: White Rabbit

• Medical Diagnostics: Celesta invests in companies like White Rabbit, which uses AI to improve the accuracy of mammogram readings.
• Mammogram Accuracy: Traditional mammograms have a high rate of false negatives (15-20%), leading to delayed diagnoses and poorer outcomes.
• White Rabbit's AI: White Rabbit's AI system has demonstrated the ability to eliminate false negatives and improve the overall accuracy of mammogram interpretation.
• AI and Radiologists Working Together: The optimal approach involves AI and radiologists working together, with AI handling routine tasks and radiologists focusing on complex cases.

AI and Job Displacement

• AI's Impact on Jobs: AI is expected to automate certain tasks and reduce the demand for some types of jobs, such as radiologists.
• Radiologist Shortage: The shortage of radiologists in the United States may mitigate the negative impact of AI-driven job displacement in this field.
• Evolving Role of Radiologists: AI can free up radiologists to focus on more complex cases and improve the overall quality of care.

Data Acquisition for AI Training

• Partnerships with Medical Institutions: White Rabbit initially partnered with Washington University in St. Louis to obtain a large, annotated dataset of mammograms.
• Acquisition of Data Sets: The company also acquired additional data sets from hospitals and healthcare systems.
• Continuous Upgrading: AI systems require continuous updating with new data to adapt to evolving diseases and improve accuracy.

AI in Semiconductors

• AI's Impact on Hardware: AI is driving significant innovation in hardware, including processors, memory, I/O, and circuit boards.
• Redesign of the Hardware Stack: The demands of AI are forcing a complete redesign of the hardware stack to improve performance, reduce energy consumption, and increase speed.
• Eleon Example: Eleon is a company that has developed a silicon-based fabric that enables the creation of smaller, more cost-effective chips (chiplets).
• Stherra Example: Stherra is a company that has developed a MEMS-based timing chip that uses a fraction of the energy of traditional crystals.

AI in Applications: Precipient

• Visual Image Analysis: Precipient is a company that uses AI to analyze visual images from satellites, drones, and cameras.
• Military Applications: The company's technology is used to identify and track potential threats in real-time, providing a significant improvement over traditional systems.
• Commercial Applications: Precipient's technology is also used for commercial applications, such as security and loss prevention in retail stores.
• Person in the Loop: While Precipient's AI system can automate certain tasks, a person remains in the loop to interpret the data and make decisions.

AI Adoption Across Sectors

• Fastest Adoption: AI adoption is expected to be fastest in sectors with the greatest need, such as healthcare.
• Regulatory Approvals: Regulatory approvals can be a significant barrier to the adoption of AI in regulated industries like healthcare.
• Consumer Adoption: Consumer adoption of AI may be slower due to resistance to change and a lack of awareness of the benefits.

Data as a Limiting Factor for AI

• Data Scarcity: The availability of data is a limiting factor for AI development, particularly for training AI models.
• Hallucinations: Insufficient training data can lead to hallucinations, where AI models produce inaccurate or nonsensical outputs.
• Patent Work Example: Clariflex, a company that uses AI for patent work, faces challenges due to the limited availability of relevant data.
• Synthetic Data: Synthetic data can be used to augment training data, but its effectiveness is still being evaluated.

Ensuring Data Quality

• Garbage In, Garbage Out: The quality of training data is crucial for the performance of AI models.
• Faulty Data: Faulty data is inevitable, and the best way to mitigate its impact is to use a large amount of data and continuously train the system.
• AI Systems for Identifying Suspicious Data: AI systems can be used to identify suspicious data, which can then be reviewed by humans.

Talent Availability for AI Development

• Talent Shortage: There is a shortage of coding and programming talent available to develop AI applications at the speed that AI is progressing.
• Geopolitical Considerations: The availability of AI talent is influenced by geopolitical factors, with China and India being major sources of talent.
• China's AI Talent: China has a large pool of AI talent and strong government support for AI development.

Commoditization of Foundation Models

• Impact on Investment: The commoditization of foundation models is expected to reduce the cost of AI and increase demand for both hardware and applications.
• Memory Analogy: The commoditization of AI is analogous to the commoditization of computer memory, where increased availability and affordability have led to increased usage.

Competing with Tech Giants in AI

• Innovator's Dilemma: Large companies tend not to be the primary innovators, creating opportunities for startups to disrupt the market.
• Data Advantage: Large companies have a significant advantage in terms of data availability, making it difficult for startups to compete in areas where data is crucial.
• Electric Vehicle Example: The vast amount of data collected by Tesla gives it a significant advantage in the electric vehicle market.

Conclusion

The discussion highlights the transformative potential of AI across various sectors, while also acknowledging the challenges and limitations that need to be addressed. Celesta Capital's investment strategy focuses on deep tech companies that are addressing specific industry needs and driving innovation in hardware, AI applications, and bioconvergence. The conversation emphasizes the importance of data quality, talent availability, and strategic partnerships for success in the AI landscape.