I enjoyed the latest LA Tech Week here in Los Angeles this past week.

It essentially amounts to a week of events hosted by VCs and startups to bring together the local tech ecosystem, which continues to thrive.

I had some great discussions on topics such as AI, fundraising, as well as the ever-changing media/entertainment landscape. And, of course, lots of product demos (e.g., Wonder, an on-demand desk research platform powered by AI.)

It always comes down to trade-offs and opportunity costs in terms of what to attend. That said, I was able to find a nice balance — both IRL and virtual — and learned some things 🤓.

One particularly good discussion was AI Pioneers hosted by USC’s Information Sciences Institute, which tends to do more applied work vs. straight-up research.

A few of the more interesting topics raised included:

• AI was likened to a bit of an “alien intelligence” 👽 in that we don’t fully understand it. Engage, collaborate, but also doubt the results seemed to be the party line.

• The large models have already ingested vast amounts of online community data and have begun assessing human behavior at macro levels.

  • AI will change social sciences and potentially become a social science as this massive social experiment is already underway.

  • That said, as practitioners and researchers, we still need to talk to communities about the biases they actually experience — and plug that into the models to clearly understand biases.

• There were varying opinions on training AI with AI. Although training with synthetic data is already happening, it was likened to a snake eating its own tail 🐍 — and there was general consensus that there need to be human signals in the loop.

• The need for safety engineering was brought up several times given how value systems differ between people. Think of the oversight associated with say, the construction of a bridge; or, the clinical trials necessary to bring pharmaceuticals to market to understand side effects, etc.

  • We don’t have clear historical examples to learn from. Not even nuclear power, necessarily. We also used to burn down cities until we understood how to harness and properly distribute the power of electricity safely.

  • At the moment, given the sheer compute dollars necessary to train these massive models, it is only the largest tech companies that can afford it. Traditionally, it has been governments that have funded innovation efforts like this (e.g., the internet). The thesis seemed to be that only government regulation can offer a proper counterweight.

Overall, it was great to hear from experts that have been working in this problem space for years, if not decades.

As the saying goes, you become less wrong by talking to experts.

Marc