2024 LA First Look Retreat — Inside AI: Predictions, Use Cases, and The View from LA
Fika hosted its 9th annual LA First Look Retreat at the iconic Huntington Library, Art Museum, and Botanical Gardens in sunny San Marino, CA. The event’s historic setting served as the backdrop to bring together distinguished guests from Los Angeles’ community along with innovators and leaders in tech and investing. The breadth and diversity of First Look has established the event as one of Los Angeles’ premier tech gatherings with attendees flying in from across the country.
The day kicked off with a series of keynote presentations from industry leaders, addressing a variety of noteworthy topics.
Fortune Senior Finance Reporter, Allie Garfinkle, moderated a lively panel discussion with AI experts Sara Hooker (VP of Research and Head of Cohere For AI), Andy Konwinski (Co-Founder, Databricks and Perplexity), and Yan-David Erlich, (CRO, Weights & Biases) to discuss the latest trends and developments in AI — including the proliferation of open-source AI models and agentic tools, and potential risks and benefits of GenAI.
Summary
Advancements in AI have led to the development of foundation models for non-text, unstructured data. These models are starting to be utilized in various sectors such as robotics and pharma, as they can process input in the form of observations which might not be language based. These models are also capable of making real-time decisions about the actions to take based on the observations from the environment.
Universal models are now being used for a variety of specialized use cases, leading to efficient “personalization” across industries. However, these models are massive and require significant resources to operate. It’s also important to consider training costs associated with work for specific use cases, especially as semiconductor technology continues to evolve. On the near-term horizon of model capabilities, Konwinski commented, “there’s a lot of agent frameworks that add [workflow] tooling and allow multi-modal capabilities and create this long-term memory system around the LLM.”
AI has the potential to solve complex real-world problems and has already begun to make an impact in areas such as healthcare and finance. Beyond the B2B software and white collar workforce spheres, the panelists also touched on personal experiences — such as virtual therapists for managing familial cancer care—to underscore the human-level implications of this technology.
On AI’s potential in healthcare applications, Hooker pointed out, “the minute we see a drug for a disease that might have been plaguing humanity for some time where human approach and navigating the state space of like whatever protein you needed to fit to the thing was going to be really hard and somehow AI accelerated that.”
While AI has numerous benefits, the technology also presents potential risks and ethical considerations. Misinformation and hallucinations are real concerns that need to be addressed formally. Panelists discussed the importance of raising the threshold for misuse and finding ways to make it more inconvenient for bad actors to exploit the technology.
“Most importantly on my mind is these models are super performant and generate very realistic outputs and so we have to be articulating what are reasonable ways to raise the threshold for misuse. What can we do to make it much more inconvenient for bad actors and to reward good actors?” Erlich shared.
