Buy Tobacco Leaf Online | Whole Leaf Tobacco

AI recommended pipe tobacco blends

DaleB

Well-Known Member
Joined
Aug 23, 2023
Messages
312
Points
93
Location
Omaha, NE
There's a similar thread in the cigar forum, but I thought I'd start one here so as not to hijack that one. Plus I always just look at all new posts, but someone just looking in this forum might not see it at all. I'd still love to know if @buck ever tried that cigar recipe recommendation, and if so how it tasted.

For my part, I asked ChatGPT to come up with a recipe. The prompt was this: "Assume you're a master pipe tobacco blender. How would you replicate Balkan Sobranie Original Smoking Mixture pipe tobacco from the 1970s?" It came up with a recipe that may or may not result in something similar, but it seemed plausible so I mixed up a small batch> I pressed it for a couple days, then rubbed it out and put it in a jarfor a few more. It's now been about a week total.

I tried a little sample of the ChatGPT-suggested blend last night. It was pretty good. I should have let it dry a little while longer, and of course all of the blends I've tried are much better after getting a few months of age on them, but I'd say it's got some potential. As I get closer to the bottom of my jar of Towers of Antioch, I decided to play around with some possible alternatives. I like ToA, and will probably make more of it, but some days it's a little heavy for my mood. That could possibly be because, as I discovered not long ago, my stock of home grown VA has some unknown percentage of Burley in it from a plant - or maybe two - that turned out to be not what I thought it was. The next batch will be truer to the recipe from Bob's book, and I suspect will be even better because of it.

So, I have some blending to do, and some taste testing, so it looks like I won't be giving up my pipes for a while yet. I mean, it's a tough job, this pipe smoking, but someone has to do it. For science.
 

DaleB

Well-Known Member
Joined
Aug 23, 2023
Messages
312
Points
93
Location
Omaha, NE
I had to depart a bit from what the AI came up with to accommodate what I have on hand. I gave it a list of what I had, and it came up with some suggestions to try. The one I picked out was with a little "tweak" to bring out a little more Oriental "incense/spice" character. Here's the original recommendation:

ComponentType
Weight (g)​
%​
Notes
Cyprian LatakiaFire-cured, smoky
33 g​
33%​
Slightly reduced to keep the Yenidje aromatic space clear.
YenidjeOriental (sun-cured)
15 g​
15%​
Centerpiece: floral, dry, incense-like, delicate.
SamsunOriental
10 g​
10%​
Adds sweetness and body to complement the Yenidje.
BasmaOriental
5 g​
5%​
Woody and faintly sour; adds contrast.
Bright VirginiaFlue-cured (lemon/orange)
20 g​
20%​
Sweetness and high-end sparkle.
Red VirginiaMatured or stoved
17 g​
17%​
Earthy sweetness, smooths the Orientals.
Total Base Weight
100 g

The one I made up as a test batch:

30% Cyprian Latakia
20% Samsun
12% Izmir
16% Bright VA
8% Dark air cured
8% Light fire cured
4% Perique

Yes, there are significant differences - because I have no red VA, Yenidje, or Basma, I have Samsun and Izmir. It seems to be tweaking with the fire- and air-cured and Perique to make up for that.

It gave me a breakdown of what information it considered and where it came from. I suppose I could tell it to ignore certain web sites or sources and give it another think. I haven't spent enough time in the pipe forums to get a good feel for who's full of... themselves... and which ones are better sources of good information.

It also suggested a very light topping "spritz" to add just a hint of floral/citrus note, consisting of a very light spray of rose water and bergamot. I haven't tried that yet, and it's not really needed. I'm not sure I would even be able to detect it, to be honest, I don't have a "connoisseur's palate" for that kind of stuff.

The little 2 gram bowl sample I tried was very nice. It's not quite the Latakia bomb I sometimes prefer, a little more rounded and complex. I'll try some again after it's sat for another week or three, and dry it more so I don't spend as much time re-lighting it. Comparing it to other English blends I've tried, it seems about the same Latakia smokiness as the current Balkan Sobranie, less than ToA or Sutliff's discontinued Sobranie match. It's not as heavy as some of my more recent smokes.

In the end... I have no idea whether the AI recipe would actually come close to 1970s Balkan Sobranie or not, since I have nothing with which to compare it. What I like is its ability to consider what I have on hand and come up with adjustments to create what seems to be a pretty nice blend. I know many people play around with endless combinations to tweak their own recipes; I don't smoke that often and it would probably take more time than I have left on this rock to figure things out.
 

DaleB

Well-Known Member
Joined
Aug 23, 2023
Messages
312
Points
93
Location
Omaha, NE
Thank you!
I wonder if the AI could theorize of an all-oriental blend used in the pre-WW1 cigarettes….
Ha! I asked about replicating Balkan Sobranie cigarettes from the mid 70s, and it said it couldn’t do that due to guardrails regarding smoking materials. I didn’t bother pointing out that it had just done precisely that with the pipe tobacco!! It was quite happy to do it as a question phrased as “research and information”… and missed by a very wide margin. It seemed to conflate the pipe tobacco with the cigarettes. Maybe it would do better with your question phrased. To be fair, there’s a LOT of discussion around Balkan Sobranie pipe tobacco and very, very little about the cigarettes.

Doesn’t cost anything to ask. Anyone can use ChatGPT for free for some number of queries per day.
 

deluxestogie

Administrator
Staff member
Joined
May 25, 2011
Messages
25,711
Points
113
Location
near Blacksburg, VA
This is a Venn diagram. AI trains on everything it can scavenge, without regard for its validity.

AI_PipeBlending_VennDiagram.jpg


Some aspect of each diagram zone is "averaged" into the machine learning result.

Bob
 

DaleB

Well-Known Member
Joined
Aug 23, 2023
Messages
312
Points
93
Location
Omaha, NE
This is a Venn diagram. AI trains on everything it can scavenge, without regard for its validity.
Some aspect of each diagram zone is "averaged" into the machine learning result.
Correct. The AI doesn't have any actual experience with blending, it just has a vast collection of information gleaned from its input sources. Kind of like me, in fact, only it's had the "time" to read a lot more than I have. In this particular case, the results have been interesting and useful. Whether it comes close to what Balkan Sobranie was in the 70s or not, I can't say. It is a pretty enjoyable English blend, and one made with exactly what I have on hand.

My next step will probably be to make up a new batch of Towers of Antioch using a more accurate mix, meaning not using my own home grown VA "mix" of unknown proportions, and see how that stacks up. It's an easier blend to make, to be sure, and what I've made is about 90% of what I want in an English blend. Maybe making it without some random percentage of burley will get it past that last 10%.

In this and in all cases, I think it's important to bear in mind what a general purpose AI like ChatGPT is and isn't, and what it can and can't do. All it "knows" is what it's been fed, of course. I've asked it, for example, to interpret doctor's notes and radiology reports, and it's been spot on every time because the language and facts are well known. On the other hand, I'd never ask it to try and diagnose something from symptoms. You always need to be aware of its tendency, when it doesn't know the answer, to just make stuff up and mix that in with "real" answers. AI developers call it "hallucinating"; in a human we'd refer to that tendency a bit less charitably. If a person does that they're not hallucinating, they're just full of - well, you know.
 
Top