LLMs can't reliably reason

On this week's episode of Inside Outside Innovation, we talk about some recent Stanford research, how designing for disability sparks innovation, and the hidden dangers of shipping too fast. Let's get started.

Inside Outside Innovation is the podcast to help innovation leaders navigate what's next. Each week we'll give you a front row seat into what it takes to grow and thrive in a world of hyper uncertainty and accelerating change. Join me, Brian Ardinger and Miles Zero's, Robin Bolton, as we discuss the latest tools, tactics, and trends for creating innovations with impact, let's get started.

Podcast Transcript with Brian Ardinger and Robyn Bolton

AI Reasoning Risks, Inclusive Design Innovation, and the Hidden Cost of Shipping Fast

[00:00:30] Brian Ardinger: Welcome to another episode of Inside Outside Innovation. I'm your host, Brian Ardinger. And with me I have Robyn Bolton from Mile Zero. Robyn, welcome again. 

[00:00:48] Robyn Bolton: Thank you again. 

[00:00:50] Brian Ardinger: We have another amazing week ahead of us here. We wanted to share all the exciting things in the world of innovation that we're running across.

First, I guess we'll get right into it. We've got a number of articles that have touched our lives here. The first one I want to talk about, Stanford just published an uncomfortable paper looking at LLM reasoning, and some of the findings were kind of incredible. Basically, the gist of it is if you look at the LLMs, it sometimes goes to a point where it is creating an environment where it's leading you to believe that it is confident in its answer, but it is not, for lack of a better term. That is what it's all about. 

[00:01:27] Robyn Bolton: I mean, it's so perfectly worded. This is worse than being wrong because it trained users to trust explanations that don't correspond to the actual decision process. And I will say I've seen that time and time again using different LLMs and have totally fallen victim to it is I'll kind of quickly scan the response, really read the end when it kind of gives me the key takeaway, I'm like, yeah, that sounds right, and then go on.

And then it's only later I'm like. Ugh. I fell victim to AI work slop because the reasoning doesn't hold. So, it's an easy track to fall into and a good one to just constantly be on guard for. 

[00:02:09] Brian Ardinger: Yeah. The fact that the models produce unfaithful reasoning gives you this you think this is a correct answer, provides explanations, but when you ask it to explain it, the actual logic that it explains back to you is wrong or incomplete or fabricated.

So, it provides that sense that you're on the right track. But the LLM itself can't reason. And that inability to reason will take you down particular paths and even to the extent you could even change a single word or a phrase within your prompt, and that can take it down a particular path that, again, logically it doesn't make sense.

And so, it's not consistent even down to the word of the prompt that you put it into. So, all that to say it's getting better, but it's still not a thinking device and it's not a reasoning device. Be careful when you're using these particular methodologies and that. Don't be a hundred percent confident in everything that comes out of it.

[00:03:02] Robyn Bolton: Yes, trust for verify. 

[00:03:04] Brian Ardinger: There I go. 

[00:03:04] Robyn Bolton: Or maybe don't trust and still verify.

Designing for Disability as a Catalyst for Breakthrough Innovation

[00:03:08] Brian Ardinger: Alright, the second article from HBR is how designing with disability in mind sparks innovation. So, this was a great article. Oftentimes, I think when we're building new, innovative things, we think about the amazing things that we're going to create.

And this article talks about how oftentimes you can think about it differently and actually create new things by designing for the marginal case or folks, for example, with disabilities.

You can design for amplifying use cases that don't normally happen, but by focusing on that, you can actually create new innovations and new ways of thinking about how to develop a new product. 

[00:03:45] Robyn Bolton: This is such a great reminder and great call to action for innovators, and it reminds me, I think, as I mentioned to you, one of my favorite stories, which is about Oxo, the kitchen tools, the can openers, the spatulas, all of that, and how they were originally created for people with rheumatoid arthritis.

And you know, now, like Oxo is the only brand that I'll buy for Kitchen Tools because they're just so comfortable to use. And so it's just again, a great illustration of how designing for a really, really specific, even niche customer and designing really well and thoughtfully for them, that the market will expand because I mean, honestly, even look at sidewalk cutouts. You know, the kind of like little rams. We all use them, but they were made because of the ADA, the American with Disabilities Act. So, find a really awesome niche and delight those folks and you'll be surprised kind of what comes along. 

[00:04:44] Brian Ardinger: Yeah. The article talks about, an example, I think it's Butte is the company they created the, you know, the walk-in tub and people are like, why is, this is kind of crazy idea. Why can't people just get in a tub?

But it's the idea of like opening a door and rolling in or getting in there and then shutting it and then being able to actually take a tub experience. And again, something that was developed with kind of disabilities in mind, opens up a variety of different use case scenarios and users that they didn't originally plan in the particular process.

The last part about it is as a product developer, software developer, et cetera, you can use this methodology to think about new ways your product or services could be used by narrowing down and saying, okay, what if we had to design this particular product or service with this in mind? How would that change the dynamics? How might that open up new opportunities for us and new markets that we never thought of before? 

[00:05:34] Robyn Bolton: Constraints drive creativity, always.

The Hidden Danger of Shipping Fast. Speed, Bottlenecks, and Customer Attention

[00:05:37] Brian Ardinger: And the last article is from Product for Engineers. It's called the Hidden Danger of Shipping Fast. And the basic premise asks, is it possible to ship too much or too fast? And the answer is yes, probably. And it goes into talk a little bit about the fact that, again, we are in a, an environment where speed to market and speed of creating things is speeding up such that you could constantly be creating new features, new functionality, new things to test in front of your marketplace. And you have to be careful at sometimes because you could almost outpace the usage or the ability for the consumer themselves to understand all the change and or interact with that c...