The Data Dilemma in AI with Innospark

Thank you to the Innospark team (especially Lily Zarrella & Eon Mattis) and our talented speakers who assembled last week to discuss The Data Dilemma in AI. Innospark invests in early stage healthcare & B2B software startups, envisioning a world where AI is ubiquitous. It was our final event of 2024 and we went out with a bang! …discussing all things AI, data, and the terabytes in between.

Look at this lineup!

Ok, what did we learn? Here are 5 takeaways for all the readers at home racing into the holidays:

  • It’s Early: We are REALLY early in the AI wave. All of our panelists were focused on considerations in both the development & deployment phase because Native AI (for the most part, outside of the LLMs) has yet to be deployed at scale
  • Testing Techniques: Gauri & Kevin from MIT talked about the tooling & modeling techniques they are exploring to get better results from both synthetic and source data in their research. This includes understanding key considerations like causal inference. There definitely isn’t consensus there and “transformers are still the best thing we have”
  • Sequencing & Risk: The Rob’s (from Dell & DataRobot) talked about all of the considerations that Enterprises must take into account to adopt AI and the challenges of nailing that AI vs. human in the loop “value chain” plus the visualization considerations of seeing where success & failure outcomes occur
  • Startups Have Advantages: Startups have a LOT of net new opportunities helping Enterprises navigate the AI wave
    • Rob Lincourt made sure to call out not forgetting about “good old fashioned AI and ML” outside of the large language models. There’s a lot more to be done improving product performance and helping Enterprises navigate these new tools
    • Data portability is a key part of this. It’s REALLY expensive to move data so helping Enterprises navigate these expensive AI models through innovative governance and access methods is something to watch
  • Agents are Next: Our panelists believe “agents” or agentic solutions – think workflow software, particularly at the infrastructure level, are really interesting 2025 use cases
    • Infrastructure use cases make a lot of sense for agents because it’s pretty much all code. Less customer facing hallucination type problems to contend with and Enterprises are grappling with “re-platforming” their technology stacks
    • Kevin Dunnell, who has worked in both industry settings and has done a lot of research at MIT, made sure to highlight that “chatbots” and text prompts will not create enduring value. Native AI companies who win will need to go 2+ steps beyond

We’ll see you in 2025! Thanks Innospark!

Leave a Comment

Your email address will not be published. Required fields are marked *