AI May Actually Work Now

If you're unfamiliar with my previous posts on generative AI--in particular ChatGPT and other similar LLMs--now may be a good time to do a quick review.

For those of you who read them before, allow me to make an exciting announcement: Generative AI might actually work now. That's right, it does what I asked. I am just as shocked as you.

After hearing from my tattoo artist about the Grok app, and its ability to search, I was skeptical. After all, most of these AI models that were not created specifically for searching are just garbage at it! And I had heard nothing even remotely positive about Grok previously. But I put on my librarian hat and did the right thing: I downloaded the app, signed in with my work Google account, and did a little testing.

Holy artificial dialogue. 

I was impressed! And I didn't want to be, either. However, I can recognize when something can provide a good answer. The first question I'll skim over--it was nothing more than I've been able to glean from other models, if among the better responses. It was the second question that caused the reaction.

From the start, I have asked all of the models to find the most recent patents granted to Rice University. Without fail, they have all given me a lot of bad answers, ranging from "I can't" to "here's a list of random patents and patent applications that may or may not be recent". Those that tell me it's "hard" to say what is most recent are just annoying, because patents are granted on specific dates. 

It explains it methodology and sources in its response, not requiring any extra clicks or selections

I ran this question on March 31st without having completed my monthly patent search for anything newly granted to Rice, so I couldn't immediately judge the accuracy of the results. But I do know quite a lot about Rice patents, and the two patents it listed took me by surprise. First, the most recent patent had what looked like a feasible patent number (12,251,449) and was granted to Rice and two other institutions that are frequent collaborators. The first inventor name listed is one I'm familiar with because he is on a lot of Rice patents, and the title of the patent matched his area of research. This was all promising.

All plausible, likely information at first glance

Furthermore, it listed a second patent that was granted near the end of  February, which I fully recognized as completely accurate (that one I had uploaded to the R3 repository). This additional patent made me think the first was that much more likely to be accurate.

Information I already knew was accurate

So I hopped on Patent Public Search, and verified. Sure enough, Grok got it right. 

Not only that, Grok listed quite a few details and sources used. This is also something I appreciate as a librarian and something I remind everyone in my generative AI classes to look for. 

Me shoutout! Love seeing that listed

Here's the bad news, though. It required Grok nearly a minute and 88 sources to find two patents.

This is not a request that requires 88 sources

Me? About 2 seconds and one source. Why? Because I am capable of real reasoning and actual strategic searching, with my very real intelligence.

So, no, Grok isn't going to replace me. Or us. It's just getting that tiny bit much better. Insert your appropriate emotion here: relief that you can stay employed, or disappointment that you still need to work.