AI hallucinations, jackpot moments, and other key insights from the GenAI conference

As AI plays an important role in furthering augmented reality and ushering in the metaverse, I have been taking some time to explore this technology. As part of this journey, I recently attended the GenAI conference hosted by JasperAi in SF which featured speakers from OpenAI , Stability AI , Cohere , Replit , Anthropic, and others. Here are the five things I learned from this event.

1. AI is a powerful horizontal technology that will "rewrite civilization"

Perhaps not surprising from an AI conference, the key takeaway in all talks was that AI will disrupt every aspect of our lives and that now is the time for individuals and businesses to understand how to get ahead of this. Former GitHub CEO, Nat Friedman , went as far as saying AI will “rewrite civilization” and that there is “not a lot it isn’t going to impact”. AI was positioned as a powerful horizontal technology which will usher in a brand new era of computing similar to the internet and eventually will become part of the “fabric of our lives”.?

2. Innovating the user interface is just as important as evolving the model

Cohere founder, Aidan Gomez , broke down the AI tech stack as (1) Model/Prompt engineering (eg. GPT-3); (2) Instruct (eg. InstructGPT); (3) Dialogue (eg. ChatGPT); (4) Retrieval (eg. ChatGPT + Bing) and; (5) Action/Tool (eg. Jasper). Most of the speakers agreed that we should expect significant innovation over the next year across this stack. But it was also emphasized that innovating on the user interface / tool layer was just as important, if not more important, than evolving the model.??

3. Today's AI hallucinates which makes it hard to trust, but it wants to be better

Overall the conference definitely provided a "glass half full" optimistic perspective on the current state of AI. Despite this, a recurring topic was AI's lack of accuracy and tendency to make things up—often referred to as “hallucinations”. It was pointed out that AI doesn’t want to lie, but in its current state it has to resort to either “I don’t know” or fabricate the answer. It is expected that innovation on the “retrieval” layer (mentioned above) will help with this as is next generation models such as GPT-4. The key takeaway was that our inability to “trust” AI right now is impacting broader applications and adoption of this technology. The goal is to get AI to a point where it never hallucinate unless we want them too.?

4. AI’s unreliability creates jackpot moments that make humans feel like badasses

The fact that today’s AI gets things wrong could be used as a feature rather than a bug. In his fireside chat, Friedman highlighted how AI’s inaccuracy makes the user feel like “a badass because you prompted it”. While he noted that being right all the time is better, he stressed that today’s “personality” of AI makes the experience feel more creative and human— referring to the moments AI produces useful results “jackpot moments”. This unreliability however, makes it easier to create demos than build products. Friedman spoke at length on the process the team took to create Github Copilot indicating it was easy to make a mind-blowing demo that you can capture on video but extremely hard to use AI in product reliably for sustained usage. Working with this deficiency, making AI “fun”, and keeping the user in the driver seat were just a few ways to successfully use generative AI in product.???

5. Leaning into what makes us human is the key to thrive in the age of AI

NY Times columnist and author, Kevin Roose , gave a great keynote highlighting three key values you/your business should adopt to either lean into or protect yourself from AI: (1) Surprising (instability, unpredictability); (2) Social (empathy, experiential) and; (3) Scarce (unique, artisanal). He proposed that you could dial these values up to protect yourself from AI or dial them down to allow AI to take over. Leaning into the things that make us intrinsically "human" will be the key differentiator for businesses and individuals during the AI era.?

I recently wrote about how AI relates to augmented reality in my 2023 Augmented Reality trends report. You can read it here.

This article was written as an independent piece. The ideas and opinions expressed in this post are mine alone and do not represent any organization, past, present or future, which I may be affiliated with.

All images were created with Midjourney.