AI in your business 1

[LOTE]

With the rise of AI, what are some ways you are using AI, or other automation, in your business?

I have been using chatGBT to draft response letters, ask it preliminary legal questions (to save time with my lawyer, not as a replacement), and learn how to improve things on my website. I have used automation for when a new client fills out an intake form on my website, it automatically adds their information into Quickbooks and sends me a CSV file so I can copy and paste into a project information spreadsheet that I hand to my drafter.

 

[debodine]

I am an ignoramus with the AI of today, but I wonder if everything you do through AI is stored/reported somewhere. I recall folks who know much more than I do say everything we do online is recorded somewhere.

 

[LOTE]

debodine, I hate to break it to you, but everything you do on a computer or phone period is recorded online somehow. Microsoft and Apple log every keystroke on your computer. Apple and Google record your whereabouts at all times through your phone. Even the engineering software we use nowadays sends back information to the company. I know you are giving up some privacy through AI and automation, but you are giving it up anyway (unless you don't use a computer or only use Linux over Tor browser). But I believe those of us that don't adopt new technologies will be obsolete if we don't adapt.

 

[CWB1]

Does yelling at various IT help and telecom chatbots count? The later almost make miss the old "Press 1 for...." menus.

Not being a computer scientist I'm honestly not sure where "AI" is used vs old-fashioned logical programming. No doubt that AI is becoming an overused buzzword for the sales folks. I ass-u-me there is AI behind some of the newer generative-design and analysis tools in recent years as I've found many inferred and surprisingly intelligent relationships/assumptions. My employer has taken a hard stance against using ChatGPT and the various other online bots for anything work-related due to security concerns, so I've only spent a few odd minutes on them after-hours.

 

[GregLocock]

During one of the AI winters they realised they needed some hits, so co-opted genetic algorithms and gradient descent into ML, which then became part of AI. Since I do gradient descent and a bit of GA, quite a lot. GD was invented by Cauchy in 1857, and GA by Turing in 1950.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[IlliniEI]

I have been wondering the same lately. Would it be possible to have chatGPT or another AI software read through a building code and provide a design based on inputs? Would it be able to consider exceptions to simplify design? The thought gets me excited, but I do not know how to "program" it to do what I need it to do.

 

[271828]

I asked ChatGPT for the tensile strength of A36 steel (58 ksi) and it gave the yield stress (36 ksi).

I also asked it for the weight of a 10 ft long 2x10, with all of the relevant info, and it came back with a calc and a result that was off by 50%.

It will have to improve a lot to be helpful. I've filed this under "check back later."

 

[IlliniEI]

I asked ChatGPT for the tensile strength of A36 steel (58 ksi) and it gave the yield stress (36 ksi).

I also asked it for the weight of a 10 ft long 2x10, with all of the relevant info, and it came back with a calc and a result that was off by 50%.

I wonder if you had put "ultimate" if it would know better. I asked it to give me wind pressures based on the 2012 IBC and it calculated them perfectly. Saved me some time looking up the coefficients from the tables.

Is there possibly another software that you could upload a pdf of the code to though? Then it would have exact knowledge instead of what it randomly stumbled upon on the internet.

 

[IRstuff]

I asked ChatGPT for the tensile strength of A36 steel (58 ksi) and it gave the yield stress (36 ksi).

I also asked it for the weight of a 10 ft long 2x10, with all of the relevant info, and it came back with a calc and a result that was off by 50%.

It will have to improve a lot to be helpful. I've filed this under "check back later."

It's a CHAT bot, not a material science bot, and it's not even supposed to do math correctly, since it's programmed and trained specifically for conversation. I don't think the Eliza test required the respondent to do math. We're getting closer to something that can seriously help with math and physics, but not there yet.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[271828]

I wonder if you had put "ultimate" if it would know better.

I don't know. I doubt it.

Google didn't have any trouble with this, in 0.40 seconds!

 

[271828]

It's a CHAT bot, not a material science bot, and it's not even supposed to do math correctly, since it's programmed and trained specifically for conversation.

There are certainly mixed messages. I've read statements similar to what you typed many times. However, look at what IlliniEI typed -- it was apparently able to compute wind pressures per the IBC.

 

[Sparweb]

It's a CHAT bot, not a material science bot, and it's not even supposed to do math correctly, since it's programmed and trained specifically for conversation.

I've met many people like that.
There seems to be a lot of them.
Why is anyone making artificial ones, when they are so many that occur naturally?

 

[IRstuff]

The current generation of ChatGPT doesn't actually "compute" things, it predicts the answer based on the stuff it trained on, so sometimes it gets lucky, particularly with calculations involving pi. It can get within 15 decimal places for some calculations with pi, but if the number is obscured, then it can fail within the first 4 digits.

User: what is 23*339.78522855738065442
ChatGPT: 23 * 339.78522855738065442 equals approximately 7,802.211265775176.

but the actual answer is 7815.0602568197550517. So ChatGPT is not a good computor; it's barely as accurate as a slide rule.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[LOTE]

I have no intention of using it for math/engineering. I think we are still many years from that being a thing. But I think the productivity and business optimizations can be endless.

 

[phamENG]

As a sort of enhanced search engine, sure. But it's writing is terrible. Maybe better than the average college freshman (and better than your average engineering graduate), but it's still very mechanical and clunky. I can pick usually pick out a GPT generated message or document in the first few sentences.

 

[geesaman.d]

I haven't done anything with chatgpt because it regurgitates ideas already written somewhere on the internet.

My needs for writing generally don't come from that body of knowledge so, no advantage there.

I will keep an eye out for other AI platforms going public. I'm told there are some shockingly good ones behind closed doors. ChatGPT is deliberately dumbed-down and narrowed to avoid scaring the public or getting into legal trouble. It would be interesting to see AI advise on component or structure optimization, or perform CAD conversions by actually operating the CAD software like a person would.

 

[IRstuff]

It would be interesting to see AI advise on component or structure optimization, or perform CAD conversions by actually operating the CAD software like a person would.

I think we are a long way from that. ChatGPT and other "AIs" are trained on training data and the whole "Chinese Box" issue comes up; do these AIs really "understand" anything at all? If there's no example to go from, the AI will be stuck. To the first order ChatGPT is a prose/conversation extrapolation algorithm, essentially "curve-fitting" a prompt to existing data and then extrapolating the result, which it's good at, but as with all arbitrary curve fits, the extrapolations can easily go into the weeds.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[GregLocock]

Is it really much more than a Markov chain algorithm, better trained? I must admit it wrote an adequate script to find every instance of a defined string in a text file correctly on the first attempt, but I imagine there are hundreds of examples of that on the web.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[IRstuff]

The analogy is plausible, but I think the distinction is the Markovian notion of "memorylessness" is not there in ChatGPT, since it's the very nature of the "memory" of the prompt and the subsequent response predictions that allows it to come to a conclusive response. ChatGPT seems to agree with me, more or less

Certainly, I'd be happy to explain the difference between a Markov process and ChatGPT's algorithm.

Markov Process:
A Markov process, also known as a Markov chain, is a mathematical model used to describe a sequence of events where the future state depends only on the current state, and not on the sequence of events that preceded it. In other words, the probability of transitioning from one state to another depends solely on the current state and the transition probabilities, and it doesn't matter how the system arrived at that current state. This property is called the "Markov property."

For example, consider a weather model with two states: "Sunny" and "Rainy." The next day's weather depends only on the current day's weather, and not on previous days' weather. If it's sunny today, the probability of it being sunny tomorrow might be 70%, and the probability of rain might be 30%, regardless of whether it rained or was sunny yesterday.

ChatGPT's Algorithm:
ChatGPT is built upon the GPT (Generative Pre-trained Transformer) architecture, specifically GPT-3.5 in this case. GPT-3.5 is a state-of-the-art language model that uses a deep neural network to generate human-like text based on the input it receives. It's pre-trained on a massive amount of text data and learns to predict the next word in a sentence given the previous words. This enables it to understand context and generate coherent and contextually relevant text.

ChatGPT's algorithm utilizes a variant of a Transformer architecture, which is known for its attention mechanisms that allow the model to weigh the importance of different words in a sentence. This enables the model to capture long-range dependencies and relationships between words.

Unlike a Markov process, which only considers the current state to determine the next state, ChatGPT's algorithm considers a broader context by taking into account a variable-length history of text to generate responses. It generates text word by word, using the probabilities learned during training to decide the most likely next word given the preceding words.

In summary, the key difference between a Markov process and ChatGPT's algorithm is that the former only depends on the current state to determine the next state, while the latter relies on a much richer and broader context to generate coherent and contextually appropriate responses.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[GregLocock]

Thanks, and that bolded bit is a useful explanation of the dangers of relying on it. Super Markov I'd call it!

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?