What to Watch in AI | The Generalist

What to Watch in AI | The Generalist

Artificial intelligence (AI) has the potential to revolutionize various industries and professions, democratizing facility and efficiency in unprecedented ways. As we witness the evolution of AI tools, it becomes clear that there will be an AI amplifying tool for every major profession within the next five years. These tools have the power to enhance human excellence across different occupations, unleashing creativity, strategic thinking, and innovation. While viral consumer-grade products like text-to-image/video models capture our imagination, the true impact of AI lies in its intersection with science, particularly in the life sciences.

In the realm of life sciences, AI is making significant strides. Companies like Valence Discovery are developing generative design methods that create new classes of potent drug molecules previously deemed too complex to design. These innovations have the potential to revolutionize the pharmaceutical industry, bringing about breakthroughs in drug discovery and development. However, a fundamental problem arises when it comes to human intent and interaction with AI models. The current interface of a simple, static text box limits the exploration and iterative nature of human intent. There is a product opportunity to build more dynamic and interactive interfaces that allow users to navigate the latent space of AI models.

Synthesia, a video creation platform, demonstrates how to build an exciting business on top of generative AI. Its users can simply type a few lines of text and generate professional videos with human avatars narrating the content in multiple languages. Synthesia's success is attributed to its focus on workflow design and its proprietary AI technology. This highlights the importance of intuitive workflows and fine-tuning models based on user feedback in the next generation of AI startups.

As the AI industry matures, the question of value accrual arises. Will the majority of the value go to AI infrastructure providers or the AI applications themselves? Companies like Grit are leveraging AI models to automate developer tasks and create self-maintaining software. Others are building models from scratch or fine-tuning open-source models. The future will reveal where the value truly lies.

Another promising application of AI is the use of digital twins in clinical trials. Clinical trials often face delays and terminations due to insufficient patient enrollment. Unlearn, a company working with major pharma companies, is deploying digital twin technology to accelerate clinical trials. By creating digital twins for trial participants, researchers can simulate how individuals would have performed in different trial arms. This technology has the potential to revolutionize clinical trial enrollment and improve the efficiency of drug development.

To achieve widespread adoption of AI, intuitive workflows will be key. AI models need to be accessible to users who are less familiar with AI, allowing them to quickly see the value. Startups should focus on designing comprehensive workflows that accelerate common use cases while providing flexibility for uncommon edge cases. The data generated by user interactions should feed back into the system, driving personalization and retention.

Chaining prompts together and utilizing different types of models will unlock massive ROI. This approach allows for multi-step actions and helps overcome limitations of existing API-driven models. The consumer side will likely see these use cases as wedges for more horizontal plays, while the enterprise side will focus on industry or context-specific applications.

The future of AI lies in multimodal models that combine multiple senses, such as computer vision and audio, to interpret the world. By integrating language models into other AI models, we can enable robots to understand their surroundings through visual and language cues. Reinforcement learning can further enhance AI models' autonomy and interaction in different environments.

However, it is important to address the biases and limitations of AI. Tools like Bot Sentinel track and analyze inauthentic behavior on social media platforms, shedding light on the potential biases present in AI image generators. Researchers are developing tools to test how AI models produce biased outcomes for certain word combinations. It is crucial to ensure fairness and accountability in AI systems.

In conclusion, the future of AI is filled with possibilities. By focusing on intuitive workflows, fine-tuning models based on user feedback, and addressing biases, AI has the potential to revolutionize various industries and professions. Here are three actionable pieces of advice:

1. Design intuitive workflows that allow users to quickly see the value of AI models.
2. Continuously fine-tune models based on user feedback to improve performance and personalization.
3. Address biases and ensure fairness in AI systems through rigorous testing and accountability measures.

As AI continues to advance, it is important to stay informed and explore the countless opportunities it brings. The intersection of AI and science, particularly in the life sciences, holds immense potential for groundbreaking innovations that can transform the world as we know it.