Do you encourage your team to use ChatGPT?

[Frans VH]

Do you, as a business owner, encourage your employees to utilize ChatGPT? What kind of challenges or issues have you faced? How do you handle the confidentiality of information? And, do you have any system in place to monitor its usage?

 

[fisicx]

Done quite a lot of testing on ChatGPT and Bard and the outputs aren’t that good. You need to do quite a lot of rework to make it decent.

Also found a number of inaccuracies, it’s not very good with facts.

 

[intheTRADE]

I have found Bard useful for a couple of things but overall, both Chat GPT and Bard, as @fisicx says, can be very factually incorrect.

Asking the same questions, just ever so slightly differently worded can throw up polar opposite answers it gives.

It is no where near ready or advanced enough for me to allow the team to use it for 'serious work'

 

[japancool]

I asked ChatGPT a cricket question and it got it wildly wrong.

Plus, anyone who's familiar with it can recognise its output straightaway. Imagine if a client realises you're replying to them with AI generated text.

 

[japancool]

How do you handle the confidentiality of information?

Our system admin vetoed it for exactly this reason.

 

[cjd]

I asked ChatGPT "what is 12345678 multiplied by 43?"
The product of 12345678 multiplied by 43 is 530,864,254.

I then asked it to regenerate the answer
12345678 multiplied by 43 equals 531,440,154

The correct answer according to Excel is 530,864,154

I don't think we have anything to worry about quite yet - unless we try to rely on it for actual answers.
I asked it again and it gave me the same wrong answers so I asked it why it routinely gives wrong answers to maths questions.
As an AI language model, I strive to provide accurate information and perform calculations correctly. However, I may occasionally make mistakes, just like humans do. These errors can arise due to various reasons, including the complexity of the calculation, potential inaccuracies in the training data, or occasional limitations in the model's understanding or processing of certain types of mathematical operations.

OpenAI, the organization behind ChatGPT, continuously works to improve the accuracy and reliability of AI models. Feedback from users like you helps in identifying and addressing any errors or issues. I appreciate your understanding and apologize for any inconvenience caused by any incorrect calculations. If you notice any specific errors, please let me know, and I'll do my best to assist you.

It seems to me that an AI that routinely gives wrong answers is more dangerous than one that doesn't.

It also has a very shallow knowledge of stuff - as soon as I ask it for details of stuff I really know about it collapses into generalisms.

It doesn't seem to be able to generate references to its information either so you can't check anything.

It's both amazing and useless.

 

[Mark T Jones]

An associate company has an AI thingy, which they use to generate lots of web content.

I asked them to run the notes of a blog through it for me, apparently in 'humorous' mode.

The content it generated was turgid, impersonal and sometimes astray from what was written. The only hint of humour was in how stilted it was.

For me, it highlights the value of being overtly non-AI - ie a real person.

It does however raise the question of whether that stilted content actually works in SEO terms?

 

[thetiger2015]

It does however raise the question of whether that stilted content actually works in SEO terms?

I imagine it would work really well. A machine reading the work of another machine.

Human generated content is for humans, it's a big problem that Google have. How do they generate and show content that is relatable by humans but has been analysed by a digital algorithm that is pretty rigid.

They're trying to weed out all the ChatGPT generated content but that must be tough. Millions of pages per day being indexed, they won't be able to catch everything.

 

[fisicx]

And then there is this:

Lawyers used ChatGPT to help with a case - it backfired massively

OpenAI's ChatGPT has wowed users with its ability to write essays and speeches, complete homework, crack computer code and pen poems - but it's also prone to completely making things up.

news.sky.com

 

[japancool]

What I asked it was how often have batsmen been given out obstructing the field in Test cricket. It told me it had happened 10 times, with the most recent being Stuart Broad during Eng vs SA in 2017, the bowler being Kagiso Rabada.

Wikipedia and Cricinfo confirm that this has only actually happened once. I checked the scorecard of the above match (yeah, I'm a cricket nut, so sue me) and it didn't happen. Rabada didn't even play in that match.

The thing is, this isn't a question of interpreting something and writing a piece on it. This is just looking up a fact, that a machine should be good at.

But ChatGPT isn't a database - it's a language model. It learns the most appropriate words to string together, rather than building a knowledge base. So it'll get things wrong.

 

[MBE2017]

Not sure if you guys are using the original or updated version.

My daughters a teacher and at my suggestion she used it to help generate pupil reports. They all had to be amended, but tbf ChatGPT produced perfectly acceptable reports. So much so the whole school has adopted the idea and is using it at the school’s recommendation to help speed things up.

Personally I have used it for web page copy, forms etc, but always check the work over and amend as required. The quality of the results are hugely dependent on the quality of the questions posed.

 

[fisicx]

A LLM is fine for churning out generic copy if you just want to fill a page with junk. What it’s not very good at is writing factual information.

The school report thing is an example of how it can be used. But not if the report is to contain facts about the pupil.

 

[johndon68]

And then there is this:

Lawyers used ChatGPT to help with a case - it backfired massively

OpenAI's ChatGPT has wowed users with its ability to write essays and speeches, complete homework, crack computer code and pen poems - but it's also prone to completely making things up.

news.sky.com

Also covered here, it's a bit of a horror show...

How to Use ChatGPT to Ruin Your Legal Career

Maybe A.I. isn't ready for primetime. ? Check out https://legaleagle.link/80000 for free career advice that will change the world from 80,000 Hours. ⚖️ Nee...

youtu.be

 

[MBE2017]

A LLM is fine for churning out generic copy if you just want to fill a page with junk. What it’s not very good at is writing factual information.

The school report thing is an example of how it can be used. But not if the report is to contain facts about the pupil.

My pages are brief and non factual, my website only exists as an inquiry site, to collect emails and telephones, so the more vague the better IMO.

The factual parts about pupils is hand edited, as mentioned all about using it where it is good, ignoring it where it is bad.

 

[IanSuth]

A LLM is fine for churning out generic copy if you just want to fill a page with junk. What it’s not very good at is writing factual information.

The school report thing is an example of how it can be used. But not if the report is to contain facts about the pupil.

From my experience of my daughters' reports from their secondary school it is obvious they are part auto generated anyway and chatgpt would make little difference - what i want is an actual report on my daughters' progress not a colour coded signal against an impossible benchmark and some meaningless auto generated twaddle

(both daughters did really well in primary SATS, the secondary school set a "challenge target" based on gaining 2 grades from what those would predict, it meant at gcse level both were challenge targeted to get a*/9's in every subject thus making exceeding that impossible and anything below that level coded red and even that level coded yellow)

 

[Frans VH]

Quite a bit of critical replies here in this thread. Just for fun, I copy/pasted the replies and asked ChatGPT for a reaction. (see below). I put interesting sections in bold.

---

As an AI language model developed by OpenAI, I appreciate the feedback and criticism shared by all these users. It's essential to note that ChatGPT is not perfect and indeed can produce inaccurate responses or interpretations, especially when it comes to factual or detailed knowledge on specific subjects. The model does not have direct access to databases, so it does not "look up" facts but generates responses based on patterns it has learned from its training data.

Furthermore, as has been noted, I don't have the ability to provide the original sources for my responses because I generate them based on a mixture of licensed data, data created by human trainers, and publicly available data. Therefore, I don't have the ability to access or retrieve personal data unless it has been shared with me in the course of our conversation. I am designed to respect user privacy and confidentiality.

For math questions or complex calculations, while I strive to be as accurate as possible, errors can happen. This is an area that OpenAI is continually working on improving.

When it comes to content generation, whether humorous or not, I try to adapt to the input given, but the result might not always meet everyone's expectations. As mentioned, the quality of the generated content can greatly depend on the quality and clarity of the input.

The confidentiality of information and security are taken very seriously, and OpenAI implements strict measures to protect users' data. However, as with all digital tools, it is always advisable to avoid sharing sensitive information.

Finally, it's important to stress that I'm a tool intended to assist users by providing information, generating text, and automating certain tasks. I'm not a replacement for professional expertise in areas like legal, medical, financial, or other specialized fields.

The purpose of AI tools like me is not to replace humans but to aid them. That being said, it's important to critically evaluate the information provided by AI and make use of it judiciously.

Thank you for all your valuable feedback. It helps OpenAI in improving my capabilities and making me a better tool for users.

---

This all seems reasonable, except for the part on math questions or complex calculations. As a language model, I don't see how OpenAI can train the model on making calculations. Unless they connect it with something like WolframAlpha.

 

[Mark T Jones]

Quite a bit of critical replies here in this thread. Just for fun, I copy/pasted the replies and asked ChatGPT for a reaction. (see below). I put interesting sections in bold.

---

As an AI language model developed by OpenAI, I appreciate the feedback and criticism shared by all these users. It's essential to note that ChatGPT is not perfect and indeed can produce inaccurate responses or interpretations, especially when it comes to factual or detailed knowledge on specific subjects. The model does not have direct access to databases, so it does not "look up" facts but generates responses based on patterns it has learned from its training data.

Furthermore, as has been noted, I don't have the ability to provide the original sources for my responses because I generate them based on a mixture of licensed data, data created by human trainers, and publicly available data. Therefore, I don't have the ability to access or retrieve personal data unless it has been shared with me in the course of our conversation. I am designed to respect user privacy and confidentiality.

For math questions or complex calculations, while I strive to be as accurate as possible, errors can happen. This is an area that OpenAI is continually working on improving.

When it comes to content generation, whether humorous or not, I try to adapt to the input given, but the result might not always meet everyone's expectations. As mentioned, the quality of the generated content can greatly depend on the quality and clarity of the input.

The confidentiality of information and security are taken very seriously, and OpenAI implements strict measures to protect users' data. However, as with all digital tools, it is always advisable to avoid sharing sensitive information.

Finally, it's important to stress that I'm a tool intended to assist users by providing information, generating text, and automating certain tasks. I'm not a replacement for professional expertise in areas like legal, medical, financial, or other specialized fields.

The purpose of AI tools like me is not to replace humans but to aid them. That being said, it's important to critically evaluate the information provided by AI and make use of it judiciously.

Thank you for all your valuable feedback. It helps OpenAI in improving my capabilities and making me a better tool for users.

---

This all seems reasonable, except for the part on math questions or complex calculations. As a language model, I don't see how OpenAI can train the model on making calculations. Unless they connect it with something like WolframAlpha.

That sounds a bit like something created by AI??

But seriously, there are no doubt multiple applications for AI and it is definitely set to grow.

For my 2 pennorth - It won't grow in quite the manner many are predicting and it is a long way from sounding properly human.

I'm very much in the mode of watching with interest, whilst playing on clearly not being AI.

 

[Frans VH]

Many interesting replies here, but I still have the question: "if and how to put this into the hands of my team".

Many people have asked me if they can use it and I don't want to miss out on the productivity enhancements that it can bring to the team.

 

[johndon68]

I don't want to miss out on the productivity enhancements that it can bring to the team.

But, if you can't rely on the answers it provides, is it going to enhance productivity if you then have to go and verify the answer elsewhere?

 

[Mark T Jones]

Many interesting replies here, but I still have the question: "if and how to put this into the hands of my team".

Many people have asked me if they can use it and I don't want to miss out on the productivity enhancements that it can bring to the team.

Presumably that depends very much on what your team are going to use it for?

My associate is only using it for blog posts, there are unlikely to be any repercussions.

If it's for technical applications, or anything which the end user will need to rely on, then you'll need to be vey sharp with fact checking and probably need to look at your PI cover.

 

[fisicx]

Many people have asked me if they can use it…

Use it for what?

 

[Frans VH]

To answer some of the questions above, here is a list of questions we've experimented with:

• what copyright or ownership notice should I add to my blog to prevent people from stealing/copying my content?
• Please correct any spelling or grammar mistakes in the following article.
• We get a customer complaint that his order was delivered late and contained the wrong items. What is the best way to approach this issue?
• What's the command to convert a KVM virtual machine into a VMWare workstation
• That client sent us a .gz file. How to read the contents of a .gz file
• What are some of the issues when a customer wants you to sign an exclusivity contract

I am aware there are many other ways to answer each of these questions, but all these gave reasonable answers to start from.

 

[Mark T Jones]

To answer some of the questions above, here is a list of questions we've experimented with:

• what copyright or ownership notice should I add to my blog to prevent people from stealing/copying my content?
• Please correct any spelling or grammar mistakes in the following article.
• We get a customer complaint that his order was delivered late and contained the wrong items. What is the best way to approach this issue?
• What's the command to convert a KVM virtual machine into a VMWare workstation
• That client sent us a .gz file. How to read the contents of a .gz file
• What are some of the issues when a customer wants you to sign an exclusivity contract

I am aware there are many other ways to answer each of these questions, but all these gave reasonable answers to start from.

It doesn't really answer our question, which revolves around are you giving professional advice?

If you ae playing/exploring 'reasonable answers' is fine - if you are being paid for advice I'd say it's inadequate or potentially dangerous.

 

[Frans VH]

It doesn't really answer our question, which revolves around are you giving professional advice?

If you ae playing/exploring 'reasonable answers' is fine - if you are being paid for advice I'd say it's inadequate or potentially dangerous.

Oh no, I am not giving paid professional advice.

 

[japancool]

You see...

Encouraging employees to use ChatGPT or any AI language model in the workplace depends on several factors. Here are some considerations:

1. Efficiency and productivity: ChatGPT can assist employees by quickly providing information, answering questions, or generating text. If using the model improves productivity, reduces time spent on repetitive tasks, or enhances workflow, it may be beneficial to encourage its use.
2. Accuracy and reliability: While AI language models like ChatGPT are powerful, they can occasionally provide incorrect or unreliable information. Critical tasks or decisions might require human judgment and expertise. It's essential to evaluate the model's accuracy and limitations to determine when and where it should be used.
3. Training and familiarity: Employees need to understand how to effectively interact with ChatGPT and interpret its responses. Training sessions or guidelines can be provided to help employees maximize the benefits and avoid potential pitfalls or misunderstandings.
4. Security and data privacy: When using AI models, it's crucial to consider data privacy and security. If sensitive or confidential information is involved, precautions should be taken to ensure that data remains protected. Careful handling of data and compliance with relevant regulations are necessary.
5. Human interaction: Relying solely on AI models for communication might lead to a lack of human connection and empathy. It's important to strike a balance and determine when human interaction is necessary or preferred, especially in situations that require emotional support or complex problem-solving.
6. Ethical considerations: AI models are trained on vast amounts of data, which can inadvertently introduce biases. Organizations should be mindful of potential biases and take steps to mitigate them to ensure fair and equitable outcomes.

In summary, encouraging employees to use ChatGPT or any AI language model depends on the specific needs and context of the workplace. Assessing the benefits, limitations, and potential impact on productivity, accuracy, security, and human interaction will help make an informed decision.

?

 

[Nico Albrecht]

12345678 multiplied by 43 equals 531,440,154

It's a bit of an odd one and a very interesting one. Based on your input inquiry, I got 493,827,100. Fair enough and a strange result. Since we use some AI frameworks for testing and teaching, I took a deeper look. I'm not sure why it came up like that, but if you input the inquiry like this:

Calculate: 12,345,678 multiplied by 43, we get the correct answer: 'Therefore, 12,345,678 multiplied by 43 equals 530,864,154.' If you send it in via an API call, it works correctly. It could be a bug in the web version.

However, if I use Edge with GPT-4 support, it does provide the correct response: 'What is 12,345,678 multiplied by 43?'

The product of 12,345,678 and 43 is 530,864,154.

My only concern is that the web-based version of ChatGPT should come with a manual.

 

[cjd]

'fraid not! Close but no coconut


Calculate: 12,345,678 multiplied by 43


The product of multiplying 12,345,678 by 43 is:
12,345,678 * 43 = 530,864,754

I asked it to regenerate

Sure, let's calculate 12,345,678 multiplied by 43:

12,345,678 * 43 = 530,864,554

So, 12,345,678 multiplied by 43 is equal to 530,864,554.

Fatally bugged....

 

[Nico Albrecht]

Fatally bugged....

Are you on GPT-3 or GPT-4 ? Version 4 seems to work just fine.

In GPT-4 The result of 12,345,678 multiplied by 43 is 530,864,154

 

[Frans VH]

'fraid not! Close but no coconut


Calculate: 12,345,678 multiplied by 43


The product of multiplying 12,345,678 by 43 is:
12,345,678 * 43 = 530,864,754

I asked it to regenerate

Sure, let's calculate 12,345,678 multiplied by 43:

12,345,678 * 43 = 530,864,554

So, 12,345,678 multiplied by 43 is equal to 530,864,554.

Fatally bugged....

ChatGPT is a large language model, which means it is not designed for performing calculations. It doesn't "understand" your queries but rather generates responses based on the patterns it has learned from analyzing various sources on the internet.

To draw an analogy, it would be akin to using MS Word for calculations instead of MS Excel, which is more suited for that purpose.

Maybe some precautions should be incorporated into ChatGPT's interface, such as warning users that the model might not always provide fully accurate answers in obvious cases like this one.

 

[fisicx]

ChatGPT is a large language model, which means it is not designed for performing calculations.

MS is baking AI into everything. It will become the de facto tool for all queries. Which means people will ask if to do calculations and many other things for which it’s not suitable.

But back to your question. There is nothing yet we have found ChatGPT to be useful for. Maybe the occasional chunk of text but not ever to answer a customer query (which your examples seems to indicate).

 

[japancool]

I've used it for quickly rewriting badly translated product descriptions into better English.

I *could* do it myself, but I don't have the time.

 

[sarky]

My post was deleted as "We don't allow unquoted AI content without purpose."

To all humans it would be clear that my reply was generated by ChatGPT and the purpose was to show how naff it was.

Perhaps the moderator is an AI bot as well.

 

[fisicx]

Did you make it clear in your post that this was AI generated?

 

[IanSuth]

in my opinion it acts like that weirdo down the pub who listens in to every conversation, then regurgitates the average of what they eavesdropped as if a subject expert.

 

[ecommerce84]

IMO, we’re too early into the AI curve for me to want to implement it in a real business that may have real percussions.

Far from making staff more productive and efficient, I feel it would do the opposite at the moment.

Based on your list of questions you’ve been experimenting with it feels as if you’re trying to shoehorn ChatGPT into the business any old how, even though there are better ways to answer most of those questions.

 

[AlanJ1]

I've used it for quickly rewriting badly translated product descriptions into better English.

I *could* do it myself, but I don't have the time.

We looked at this but really not ideal for our sector yet.

 

[IanSuth]

We looked at this but really not ideal for our sector yet.

My wife (a tech support engineer) has spent the last 2 days proof reading and editing the translation of the manual for their latest product (they are a distributor and it is a Chinese made piece of specialist electronics), it needed both decent English and an engineers understanding of how the product is used to be able to produce an end product that with save users ringing them rather than cause more work.

 

[japancool]

My wife (a tech support engineer) has spent the last 2 days proof reading and editing the translation of the manual for their latest product (they are a distributor and it is a Chinese made piece of specialist electronics), it needed both decent English and an engineers understanding of how the product is used to be able to produce an end product that with save users ringing them rather than cause more work.

I'm not using it for technical manuals or instructions, so it's not an issue for me.

I just need something more readable than "It car have four wheel and go very fast".

 

[Mark T Jones]

A teeny bit off topic, but related.

Talking to a client today who works with the film industry, said that side of things was a bit quiet due to a script-writers' strike

I looked it up, sure enough they are. It started in the US but has come here

Other than pay, the key concern is the advent / threat of AI to their jobs

 

[JEREMY HAWKE]

I have not got my head around it yet but just to save me some time my question is

Will it make me anymore money ££££££?