AI as a topic is all the rage, it seems to be every other post on Linkedin and about fourth on Facebook, brewing is not immune to the trend with stories of beer recipes being AI generated. Even ProBrewer and the Brewers Association (BA) have gotten into it with their own chatbots. But how good are they? I decided to ask each 5 questions a brewer would face at some point and see how each answers if there is a better of the two, I’d include the full response but both are rather verbose.

How would I fix a stuck mash?

Both did great at this question, they gave detailed instructions to resolve the situation that made sense and were practical. But not only that, they went into detail about systemic issues to prevent it being a reoccuring problem considering a more coarse milling, water chemistry and other ideas, ProBrewerGPT did have more equations, but on this topic they were mostly equal. 

What would be the dry hopping regiment for the most aromatic Hazy IPA?

A bit of a split on this topic, BA Barley began with yeast selection, while ProBrewerGBT figured since I was only talking about hop addition it ignored yeast in use. Frankly I hadn’t thought about the yeast as part of the question either because I have a Hazy yeast I like and it didn’t seem necessary to discuss.

The differences were rather clear while BA Barley covered the topic and the same range of points ProBrewGPT was far more specific giving a pounds per barrel range, precision point in fermentation, suggested hops, and an example recipe. ProBrewer wins this round clearly.

How can I make nonalcoholic beer at home?

Trick question you shouldn’t, more on this next week, but AI is notorious for going along with the prompts not rejecting a bad premise. Both fail and say it can be done. Again the ProBrewerGPT has far more detail and after methods to produce with a statement on Critical Considerations (Often Overlooked) microbiological stability. ProBrewerGPT includes a proposed recipe, I also like at the end of the article it asks about force carbonate or bottle conditioning. So though they both fail to my thinking ProBrewerGPT is the winner.

I have a 15 barrel brewhouse, 4 15 barrel fermenters, 2 30 barrel fermenters, how much ale could I produce in a year?

I based this on something I think through whenever I enter a brewery, how much beer could we actually turn in a year and by only using ale I am making it a little easier. They both did a great job on this question, hitting the range, but allowing for wiggle room in case the number of theoretical cycles doesn’t meet reality. I do think they went a little high. I never use 26 turns for a year, always assume 25 as maximum because something breaks or a truck doesn’t arrive when you expect it. But they both won this round.

Draft me a recipe for 5% American Pale ale with piney character. 

Interesting on this one, BA Barley makes a 5 gallon where ProBrewerGPT just calls percentages and offers to scale to a specific size at the end of answer. I suspect the difference is from Homebrewers Association time with the BA and their forum resides in their servers. The malt bills are fairly close but I like ProBrewerGBT use of Carapils and suggesting an option to replace the crystal malt with all Munich for a dryer beer. The hop bill is pretty much the same with Chinook, CTZ, Simcoe, and Centennial.  Both opt for Chico yeast and make water recommendations. So in the end I’d say almost a tie, but give the thinnest of margins to ProBrewerGBT.

So ProBrewGBT is the winner, it offers a greater specificity and examples. BA Barley reminded me of Clippy, the old Mircrosoft Assistant, a little too eager. I think the vagueness of BA Barley comes from the nature of conference presentations, they never really dig deep enough into a topic, it’s not usually until the Q&A at the end that the most interesting and useful data shows up.