"What to Watch in AI: Lessons Learned from Shutdown Startups"

"What to Watch in AI: Lessons Learned from Shutdown Startups"

The exponential rise in “knowledge” and the increasingly distributed nature of work have created a need for more efficient ways to find existing knowledge. The traditional method of "searching for stuff" at work is no longer effective in today's fragmented knowledge landscape. This is where intuitive work assistants like Glean come into play. What was once considered a nice-to-have tool has now become a critical component in driving employee productivity.

However, one of the key obstacles preventing enterprises from implementing AI applications is the lack of appropriate governance controls. Questions such as "Does my application understand what the end user is allowed to see and not see?" and "Is the inference done on my servers or OpenAI's servers?" need to be addressed in order to ensure proper data privacy and control. Understanding the source data that led to a given model output and its ownership is also crucial.

While data processing and annotation are still the most tedious and expensive parts of the AI process, they are also the most important for achieving quality outcomes. Despite the availability of pre-trained large language models, enterprises must focus on utilizing their proprietary data across multiple modalities to create AI models that can deliver differentiated services, valuable insights, and increased operational efficiencies.

In the realm of startup failures, Rdio provides an important lesson. Instead of relying on recommendations from store clerks, Rdio offered recommendations from people you know. This approach recognized the power of solid recommendations and the trust they build. Word-of-mouth can be a powerful tool in a world filled with noise. Sometimes, offering a focused alternative can be the winning differentiation that sets a startup apart.

Combining the insights from these two sources, it is clear that the success of AI applications lies in the ability to harness the power of recommendations and trust. As organizations become more distributed, the need for reliable recommendations becomes even more critical. AI assistants like Glean can not only help employees find existing knowledge more efficiently but also provide personalized recommendations based on trusted sources, whether it be colleagues or friends.

Actionable Advice:

1. Implement proper governance controls: To ensure data privacy and control, it is essential to enforce appropriate governance controls in AI applications. This includes understanding what the end user is allowed to see, ensuring inference is done on secure servers, and tracking the source data and its ownership.

2. Focus on proprietary data: While pre-trained models can be useful, enterprises should prioritize leveraging their own proprietary data across multiple modalities. This allows for the creation of AI models that can provide differentiated services, valuable insights, and increased operational efficiencies.

3. Embrace the power of recommendations: Recommendations build trust and trust leads to word-of-mouth. Whether it's in the form of AI assistants or personal connections, incorporating recommendations into AI applications can greatly enhance their effectiveness and adoption.

In conclusion, the future of AI lies in the ability to address the challenges of finding existing knowledge in a distributed work environment and implementing proper governance controls. By focusing on proprietary data, harnessing the power of recommendations, and embracing trust-building strategies, enterprises can unlock the full potential of AI and drive productivity to new heights.