ChatGPT Absinthe

[PLAYMP]

Not sure who among us is applying ChatGPT to our hobby but I've recently been into a bit a bit for the purposes of selecting botanical bills for gin recipes. I just finished a gin run on my air still earlier today as a birthday present for a buddy. I asked his girlfriend what he was into these days and put her response into ChatGPT and asked it to craft a gin botanical bill around it. I was really impressed, the botanicals were creative and complimentary but not impossible to obtain and the ratios of juniper/coriander/angelica/etc were all totally on the money as far as I was concerned. I think it has the potential to be a really powerful tool for our craft if I'm being honest.

Anyway, I've had good luck with gin but I wanted to throw ChatGPT an absinthe challenge, understanding that there's more than likely far less on the internet to draw on. I asked ChatGPT to suggest a botanical bill, separating out those for maceration and those for coloration, that is rooted in the 19th century french style but also features some modern and complimentary flavors. It suggested the following:

For maceration (g/L of 85% spirit):

17 grams wormwood (Artemisia absinthium)
17 grams anise seeds
17 grams fennel seeds
9 grams peppermint
9 grams coriander seeds
4-5 cardamom pods
1 small piece of star anise
9 grams lemon or orange peels
9 grams chamomile flowers
9 grams angelica root

Coloration botanicals weren't anything special: hyssop, lemon balm, petite wormwood, the usual. So I'm just going to go with the Duplais method for coloration.

I wasn't happy with the amounts suggested for wormwood, anise, and fennel as they are smaller than anything out there I can find in terms of gross amount but also the share of the botanical bill. I want to experiment but I don't want to do anything completely out of nowhere. So I aligned the holy trinity botanicals to fall in line with Duplais' Lyons recipe, which is heaviest on those botanicals. I really want to go hard on those botanicals to leave no flavor on the table and also ensure they aren't overpowered by what I think are fairly sizeable amounts of accent botanicals. Maceration recipe was adjusted to be:

31 grams wormwood (Artemisia absinthium)
84 grams anise seeds
42 grams fennel seeds
9 grams peppermint (I had to use fresh and 9 grams was about all I had)
9 grams coriander seeds
4-5 cardamom pods
1 small piece of star anise
9 grams orange peels
9 grams chamomile flowers
9 grams angelica root

The whole botanical gang minus peppermint, which I hadn't picked yet (d'oh)

I usually do this for gins but I toasted all the botanicals in a cast iron pan on medium low heat for a short period. I would never cook with these without this step and It does a great job at waking up the flavors and aromas:

Then I crushed everything up:

and added all back to the bowl for an aroma check, before giving it a bath in 1.5L of 85% neutral.

Hopefully I'll get around to running this soon but I'm pretty excited about it. I'm definitely getting traditional absinthe aromas but the orange is working really well with the anise and fennel and anybody is share this with is really going to be wondering what the notes the chamomile is giving off comes from. Stay tuned!

[Wildcats]

Nice work putting this all together. Let us know how it goes. Looks good.

[MooseMan]

Very interesting!

It never even occurred to me to try that!

[NZChris]

If you don't use the ChatGPT botanical bill and method, it's a bit cheeky to call it ChatGPT Absinthe.

If you don't like the ChatGPT result, you don't have to throw it away. You can make another batch using more and less of the AI suggested herb bill to blend with it into something that you like better.

[PLAYMP]

If you don't use the ChatGPT botanical bill and method, it's a bit cheeky to call it ChatGPT Absinthe.

Certainly not trying to misrepresent. I guess I could have put a word in there, "ChatGPT Assisted Absinthe" or something. I would have liked to take the recipe as written but with respect to the Wormwood/Fennel/Anise amounts, I tried to probe and understand where it was coming from but it started confusing itself and it was clear this was probably towards the limit of what it was capable of in it's knowledge base for now.

I had a much different experience for gin where the amounts and ratios logically correspond to anything you could easily read about. I think the magic is getting it to recommend botanicals and flavors that aren't on your radar.

[contrahead]

Not sure who among us is applying ChatGPT to our hobby but ................ I think it has the potential to be a really powerful tool for our craft if I'm being honest.

I think you are powerfully mistaken.

I think AI should have no role here. Gin and absinthe products are simple enough concepts that little ole people alone, can create them – unhindered.

As we speak, AI is intruding upon several creative realms simultaneously. I personally resent this threat to the livelihoods of artist and writers worldwide (whether they realize it yet or not).

Someday soon poor human creative intellects (writers / artist) may come to realize that even their best and most cherished intellectual aspirations are ultimately pointless, because some machine can spit out better work and without effort. A depressing notion.

Personally, I am alarmed by the presence of a witless community that is imprudently rushing headlong to embrace all aspects machine intelligence. Their will be repercussions. The benefits be dammed.

If I was to read a story or long piece of text, and then discover that it was composed by some chat-bot, then I'd want to spit it out; regurgitate it metaphorically. How do college professors even spot machine faked essays? Can they?

I'd better stop.

[still_stirrin]

Caution for when “artificial intelligence” takes control of our lives. Wait until AI recommends arsenic as a botanical for your gin. Don’t become sheep blindly following someone’s computer program as you jump off a cliff.
ss

[NormandieStill]

Not sure who among us is applying ChatGPT to our hobby but ................ I think it has the potential to be a really powerful tool for our craft if I'm being honest.

I think you are powerfully mistaken.

I think AI should have no role here. Gin and absinthe products are simple enough concepts that little ole people alone, can create them – unhindered.

For those struggling to find old recipes, or wanting a better translation, ChatGPT might be useful. But could we please stop calling it AI. It's basically the predictive text from your smart phone on steroids. It doesn't "know" what it's answering. Which is why it doesn't know when it gets it wrong. There's no creativity there, it's just spitting out what other people have written on the subject. it could make some interesting suggestions by mistake, but if you want to create a recipe, you'd be better off experimenting with different botanicals yourself... and trying to imagine what's missing... or what would complement the flavours you've already got.

[Stonecutter]

Don’t take this as a dis towards AI but
It’s a fishhook. You didn’t end up using the recipe spit out by the bot.
Kudos for the gin journey but F*€k the headline. The recipe was spit out by whatever the “engineer” fed into it.
It does make you wonder how much of the HD the filthy bot was fed but in the end, YOUR hardwork and dedication to the hobby could easily outpace some programmer.

Let that light shine down PLAYMP. Definitely interested to hear how your recipe turns out

[PLAYMP]

Don’t take this as a dis towards AI…

I mean…haha

I think I misread the room on this post. For where I’m at and how I approach this hobby I’ve found it a pretty useful tool but I’ll always stop short of referring to it as a “solution”.

I should be able to run this tomorrow but may let it sit for a few days before I really get into it, will post updates just for the sake of wrapping this up and on the off chance there are any interested lurkers out there.

[contrahead]

ChatGPT might be useful. But could we please stop calling it AI. It's basically the predictive text from your smart phone on steroids. It doesn't "know" what it's answering. Which is why it doesn't know when it gets it wrong. There's no creativity there, it's just spitting out what other people have written on the subject. it could make some interesting suggestions by mistake, but if

I guess I was led astray by the 'business technology news website' story that I read yesterday.
They (the writer and two magazine reviewers) named ChatGPT an “artificial intelligence (AI) revolution” - in the very first sentence.
https://www.zdnet.com/article/how-to-use-chatgpt/

[NZChris]

I had to do battle with ChatGPT to get it to give me the answer I was looking for, rather than morphing every query into an easier question and answer could have come from a cursory search of HD without checking the quality of the posts and qualifications of the posters it was using.

I had to constantly rephrase the query to cut off every escape route/method it used to avoid doing the actual research needed to find the answer I was looking for. It was like pestering a politician or preacher with a question about their offerings that they didn't want to answer, but who couldn't get away with saying nothing without embarrassing themselves.

I did eventually get an answer that sounds useful, but I haven't tested it in my shed, yet.

[NormandieStill]

ChatGPT might be useful. But could we please stop calling it AI. It's basically the predictive text from your smart phone on steroids. It doesn't "know" what it's answering. Which is why it doesn't know when it gets it wrong. There's no creativity there, it's just spitting out what other people have written on the subject. it could make some interesting suggestions by mistake, but if

I guess I was led astray by the 'business technology news website' story that I read yesterday.
They (the writer and two magazine reviewers) named ChatGPT an “artificial intelligence (AI) revolution” - in the very first sentence.
https://www.zdnet.com/article/how-to-use-chatgpt/

The problem is that "AI" has two meanings. In the AI research community it covers the full range of anything that is not "pre-programmed". So neural networks or other trained systems as applied to various problems including facial recognition, spell-checking, route-planning. This goes all the way up to the "holy grail" of generalised AI (a human-like intelligence for want of a better description).

However in the general public, AI (courtesy of books and films) has come to be synonymous with generalised AI (The self-aware Skynet system, or the Tic-Tac-Toe playing program in Wargames). Since the release of ChatGPT a phenomenal number of articles have been written on the subject, but few of them by people with any prior understanding of the field, and many of them using (ironically) ChatGPT to help with the research. The result is that the distinction between an LLM and GAI has not been successfully explained to the general public a large portion of whom believe that GAI will just be a more powerful version of ChatGPT, as opposed to a fundamental shift in methodology.

ChatGPT passes the Turing Test and it's responses don't read like those of a computer, but those of a human being. Our natural instincts then place it into the category of human-like and so we assign it values which are human-like, including understanding. But it does not "understand" anything that it prints. The answers to any given query are an assembly of probabilities. Given a keyword, it will produce what are the statistically most likely sentences that answer that question. But all it's doing is regurgitating the sentences that it was trained on. If it's training had not included actual recipes for absinthe then it would have been unable to infer the ingredients from tasting notes that it had already seen. It has no understanding of the difference between maceration for distillation and maceration for coloration.

It is a very useful search engine. Able to parse vast amounts of data from myriad sources and potentially make links that would have been hard to do as a human simply because the data is too disparate and we're (mostly) too bad at note-taking.

I think that when the dust settles a (relatively) small number of jobs will be lost. As happens with every technological leap forward. Actual scriptwriters will continue to write scripts, but they may well use ChatGPT (or an equivalent) to research, rather than using a dedicated researcher. Some bodies of text will be auto-generated but re-read and tweaked by an editor. There will be fewer graphic designers but those that remain will produce more in less times by leveraging generative AI programs for the heavy lifting. What scares people now, is that the jobs that will be lost to this new tech revolution are not the "old", "menial", "hard-labour" jobs, but the high-tech creative ones that were supposed to be safe. But for now (and for the near future at least) this is not going to replace creativity. It's just going to make (for better or for worse) a smaller number of creative people more productive.

Which is all more than a little off-topic, so I'll stop there.

None of the above was written with the help of generative AI, so any errors are not "hallucinations", they're just dumb mistakes!

[PLAYMP]

Well I finally got to run this and let me be the first to admit any faults with this spirit is going to be due to my execution. I just got an air still to do small botanical runs and after a few with gin this was my first foray into absinthe and it simply didn't behave the same way. Lost a little product along the way but kept enough for a couple of bottles, just coloring one for now. Since I used the Lyons botanical proportions I just went with that recipe for coloring:

1 part petite wormwood
1 part lemon balm
1/2 part hyssop
1/2 part veronica

This was added to about 40% of what was going to make the contents of this bottle.

I've never used this recipe for coloration and I thought I might have botched it because it came out pretty yellow-brown as opposed to anything remotely emerald green but I did a quick google and it looks like I'm not wildly off for a Lyons-style absinthe: https://images.app.goo.gl/vZxV9WtbGZXDwatL8

Anyway, wife and I both caught an awful bug over the last few days and my palate isn't in near good enough condition for a taste test. Just as well since I always like to let it sit anyway, will post a louche pic and anything of note in the taste test when I get around to it.

[NZChris]

My Absinthe doesn't lose its green for at least a couple of years.

Check your botanicals individually to find out which one to throw away. If none of them are at fault, your coloration method might have done it.

[tommysb]

The answers to any given query are an assembly of probabilities. Given a keyword, it will produce what are the statistically most likely sentences that answer that question.

Absolutely correct Normandie. I wish more people understood this. Unlikely, as probabilities themselves are not intuitive, and difficult to grasp unless you are properly trained to interpret them.

[contrahead]

None of the above was written with the help of generative AI, so any errors are not "hallucinations", they're just dumb mistakes!

--------------------------------------------

"<Phew> the following took a long time to read.

I am not buying all of this professor's arguments either. She starts off by explaining how generated text is easy to spot; but then defeats this premise buy the time she finishes her long winded essay. She is obviously proficient in directing ChatGPT herself also, and I wouldn't be the least bit surprised if she used some of it here. The text is worthwhile reading in some spots.

What AI Teaches Us About Good Writing
By Laura Hartenberger July 25, 2023

https://www.noemamag.com/what-ai-teache ... d-writing/

[contrahead]