A couple of questions on CHAT GTP/4:

[El Presidente]

I have been using GTP-4 since release and have been impressed. I use it very much as a personal assistant/information gatherer. While sceptical, it is proving to be an excellent personal productivity tool. I will roll it out through the rest of the team over the coming month. 

My two questions. 

1. How are you using CHAT GTP4 or similar AI?  Let us know your experience to date. 
2. Will  GTP future versions (or similar) be able to solve the issue of verifying authentic Habanos? How do you see that working? 

 

[rolaand]

I have been using GTP-4 since release and have been impressed. I use it very much as a personal assistant/information gatherer. While sceptical, it is proving to be an excellent personal productivity tool. I will roll it out through the rest of the team over the coming month. 
My two questions. 

1. How are you using CHAT GTP4 or similar AI?  Let us know your experience to date. 
2. Will  GTP future versions (or similar) be able to solve the issue of verifying authentic Habanos? How do you see that working? 

 

Also, counterfeiters could use it to possibly identify pitfalls and learn very quickly. I think many of the key benefits of large language models are exactly what you specify: summarizing things and correlating harder to find/obscure data.

[El Presidente]

37 minutes ago, rolaand said:

Also, counterfeiters could use it to possibly identify pitfalls and learn very quickly. I think many of the key benefits of large language models are exactly what you specify: summarizing things and correlating harder to find/obscure data.

True.

I can see a time in the not too distant future where you can scan your box or upload relevant pictures of seals, bands, serial number, box code etal. 

Not foolproof by any means ....but a nice tool.

[whaleyboy]

I voted that I would use it for counterfeit spotting but I am not completely sure but I am open to the possibility.

It is really interesting as a vehicle or mechanism to use natural language to dig into information that is out there so I can easily imagine it being useful for all kinds of collectable spotting (I was thinking about records and their provenance which is another area of interest for me).

[BoliDan]

Sometimes I ask it to help me with some python code. Its not good at anything too big, but if you ask it the right way with the right details it can spit out a piece of an algorithm or something I'm working on. Its pretty crazy. 

I don't know what I'd use it for cigar wise at the moment 

[chris12381]

I have the paid version.  I've been playing around with it a lot.  I've been doing a few things with it, primarily.  A few of the top of my head:

*Engaging it in intellectual discussions.  Challenging it to provide supporting evidence for it's psoitions and consider different ideas.  

*Ask it to put together an outline of a complex topic (such as the historic interpretation of the 4th Amendment in the US).  Then have it modify it's answer to expand upon x and y and discuss Terry v Ohio as it relates to it's previous answer.  

*Creating prompts that actually generate progressively better and more concise prompts based on an initial prompt.  That final prompt is then fed into the engine to generate a better response from the engine. 

*Translate from English to Spanish (it does a much better job than Google).  You can specify specific national dialects.  

 

[iggy_marley]

I tried using it to pair cigars and spirits. Often times i got pretty comprehensive answers to proper pairings and other times i got some basic generic responses. It seems that you either need to know how to ask the question properly or it needs to be more polished to become more useful. The potential is there for sure though, just a matter if time before it sucks up the universe

[MrBirdman]

ChatGPT will be the ultimate first-draft machine. Eventually we’ll all be editors - let the AI do the grunt work and then the big brain takes over. As someone who hates the first-draft but loves editing, I’m thrilled with the prospect. It’s long term effects on creativity and how we think are more concerning. 

[Duxnutz]

I want nothing to do with skynet. 

[ha_banos]

1 hour ago, Duxnutz said:

I want nothing to do with skynet. 

You've used the interwebz and have a mobile phone? Too late josé!

[gormag38]

8 hours ago, MrBirdman said:

It’s long term effects on creativity and how we think are more concerning. 

You hit the nail on the head in my humble opinion. The research has shown that reading and writing will often times lead to better critical thinking and problem solving skills. Colleges and high schools are going to find more and more students taking advantage of this software; I truly believe that there will be a generation where AI (and similar) is the norm. How will that change our young persons' brain development? Creativity? Problem solving? At first glance I would think it could have a negative effect. 

14 hours ago, El Presidente said:

My two questions. 

1. How are you using CHAT GTP4 or similar AI?  Let us know your experience to date. 
2. Will  GTP future versions (or similar) be able to solve the issue of verifying authentic Habanos? How do you see that working? 

To answer the questions above. I currently do not and have not used CHAT GTP4. It does seem to have the possibility to be an incredibly useful tool I will admit.

As for verifying authentic Habanos, I couldn't really see it working in that regard (or at least until system advances are made). Seems to me that with the level of fakes we've been seeing recently a counterfeiter could basically match everything that could be uploaded in a photo. If provenance could be included in the equation that would help with the ability to determine fakes but third party/auction/private sales would make that pretty tricky. 

[Bijan]

6 hours ago, gormag38 said:

You hit the nail on the head in my humble opinion. The research has shown that reading and writing will often times lead to better critical thinking and problem solving skills. Colleges and high schools are going to find more and more students taking advantage of this software; I truly believe that there will be a generation where AI (and similar) is the norm. How will that change our young persons' brain development? Creativity? Problem solving? At first glance I would think it could have a negative effect. 

Reading and writing abilities likely peaked right before the radio and telephone era (not to mention television) and are probably doing a bit better now that we have email/instant messaging.

[BoliDan]

I just asked it about the existence of plume. Nope, time to kill it. 

Quote

Yes, plume is a natural phenomenon that can occur on cigars under certain conditions. Plume, also known as bloom, is a thin layer of white, powdery substance that can develop on the surface of a cigar when the oils in the tobacco leaves interact with the air and create tiny crystals.

Plume is often considered a sign of a well-aged and well-maintained cigar, and it is generally not harmful to the smoking experience. However, it is important to note that plume can sometimes be confused with mold, which can be harmful to both the cigar and the smoker. If you are unsure whether the substance on your cigar is plume or mold, it is best to consult with a cigar expert or discard the cigar to avoid any potential health risks.

 

[Corylax18]

57 minutes ago, BoliDan said:

I just asked it about the existence of plume. Nope, time to kill it. 

 

This takes me back to my first "computers" class circa 1998 or so. The first thing the teacher said: Garbage In, Garbage Out.

I've still yet to be impressed by anything I've seen from this software. I have no doubt A.I. will eventually take over the world, but this is just a google search bar that can "type" the first few lines of the first few links it finds. Great, its faster than its ever been, but I still don't see anything that even closely resembles any form of "intelligence." 

[Bijan]

25 minutes ago, Corylax18 said:

but this is just a google search bar that can "type" the first few lines of the first few links it finds.

Except ChatGPT doesn't use the internet. It is entirely self contained. (some chatbots aren't but most are offline, and only trained on what was found on the internet)

You could take a computer with the software on a submarine and ask it the questions and it would answer just as well.

And the answers (while sometimes/often tragically wrong), will be better written than most people could manage.

[rolaand]

This is correct ^^. For example the LLaMa model can currently be downloaded from huggingface and run on your computer at home (theoretically). No internet should be required after the initial download since the model and parameters are contained in the code base.

[Bijan]

1 minute ago, rolaand said:

This is correct ^^. For example the LLaMa model can currently be downloaded from huggingface and run on your computer at home (theoretically). No internet should be required after the initial download since the model and parameters are contained in the code base.

I have run the 20GB size model on my home computer. It requires a video card with 24GB to run that size. Which is pretty much the NVidia 4090. But you can run the smaller models on 12GB or 16GB cards. Also you can run it with somewhat slower performance and somewhat worse output entirely on CPU.

Also the next thing I want to play with is that there are image generators that are very very good, that can also be run offline at home, as well as text to speech and speech to text.

[rolaand]

Yeah I am familiar with stable diffusion but i havent played with too many others. I am actually pretty lucky this came up as a research direction at work because I find it very interesting.

[Corylax18]

9 minutes ago, Bijan said:

And the answers (while sometimes tragically wrong), will be better written than most people could manage.

As others have alluded to, that speaks more to the general lack of reading/comprehension skills than anything and this will only exacerbate the issue. Computers have been better at spelling, grammar and word selection than most humans for 20 maybe 25 years now. Microsoft Word's spellcheck worked without an internet connection in the mid 90s, was that A.I.?

11 minutes ago, Bijan said:

Except ChatGPT doesn't use the internet. It is entirely self contained. (some chatbots aren't but most are offline, and only trained on what was found on the internet)

So, once it has everything it needs from the internet, it doesn't need the internet anymore? Hmm. Until the next version comes out, that has more information, from the internet? 

Again, I'm not saying A.I. wont eventually be a very powerful thing. But what we have so far is not intelligent, even by the drastically reduced standard we have for that word today. 

[Bijan]

1 hour ago, Corylax18 said:

As others have alluded to, that speaks more to the general lack of reading/comprehension skills than anything and this will only exacerbate the issue. Computers have been better at spelling, grammar and word selection than most humans for 20 maybe 25 years now. Microsoft Word's spellcheck worked without an internet connection in the mid 90s, was that A.I.?

No, they haven't. Computers have not been able to write decent text until recently. Machine translation was pretty shoddy 10 years ago.

The spellcheck was horrible. You'd put in a page of Mark Twain and it would correct every other word.

You can ask chatGPT to generate text in the style of the King James bible or Gangster rap.

 

1 hour ago, Corylax18 said:

Again, I'm not saying A.I. wont eventually be a very powerful thing. But what we have so far is not intelligent, even by the drastically reduced standard we have for that word today. 

So you're saying on a scale of 0 = ant, to mouse, to dog, to monkey, to 2 year old, to average human being = 10, you'd say it's what 0?

If you asked it a question where would you rate it?

I think it's maybe a 5.

Computers were not that 10 years ago, nowhere near there 25 years ago.

Could you walk up to a computer with no internet 25 years ago, ask it a question and have a 50/50 chance of getting a useful answer in proper English?

Edit: and note ChatGPT 4 will handle dozens of languages.

[Chibearsv]

So how does it gain experience?  Isn’t that a key to intelligence?  If the box at home has no way to act and analyze it’s own experiences, isn’t just a fancy encyclopedia?  A self driving car that crashes and then learns not to make the same mistake again is more impressive to me than a reference tool that can answer questions in a text conversation. 

[Bijan]

10 minutes ago, Chibearsv said:

So how does it gain experience?  Isn’t that a key to intelligence?  If the box at home has no way to act and analyze it’s own experiences, isn’t just a fancy encyclopedia?  A self driving car that crashes and then learns not to make the same mistake again is more impressive to me than a reference tool that can answer questions in a text conversation. 

I think we're moving the goal posts.

This is a question of artificial intelligence. Whether or not a computer can do something that was previously entirely within the domain and capabilities of human intelligence.

If a chess computer can beat any human, is it intelligent? I guess in that specific domain of playing chess, that's the metric.

If a self driving car can drive with much fewer accidents than a human, is it intelligent, even if the individual car can't learn from it's mistakes? Could be, depends.

Sure there's always something or someone that is more impressive. Maybe Einstein is more impressive than Newton.

ChatGPT is about what 1950s films and television thought of and showed as a computer, and what computers have not achieved until very recently.

[rolaand]

I mean there are many different kinds of learning within AI and the most recent LLMs and GPTs are extremely impressive imo

[Bijan]

Also I will say that ChatGPT is really sold and aimed at working as a reference librarian, and refuses to just be a text generator or chat bot.

But the facebook llama model in chat mode, is pretty passable as a regular human being chatting.

You just give it a script of who to be, and it does a very believable job as chatting as that persona.

And it is entirely more argumentative than ChatGPT (which basically always agrees with you).

All this to say, the llama chatbot is more "intelligent" (or at least definitely appears so) than the average FB troll.

 

Again 25 years ago Microsoft added clippy, the annoying paperclip 📎 assistant to office/word.

I would bet it didn't help anywhere near 1% of users.

I'm sure not a single sensible person felt that it understood them and responded sensibly to their questions.

People in technical fields are now using ChatGPT (and similar tools) to get real work done. It's definitely wrong a lot (this is where @MrBirdman's idea of first drafts and editing comes in). But it's also usually more useful than asking random people off the street.

[Rhinoww]

I just started playing around w an ai program. It reminded me of when I started my first online legal research project. It’s a tool, but currently only as good as the operator and their search terms. Was with a colleague last night who had some fun with the prompts answers in Haiku written in the voice of a particular writer. His paid version came back with something passable. If the prompt had just been for a poem different product. 
it’s a tool that I’ll need to figure out when, where and how to use. Glad I’m not writing code though …