ChatGPT

I feel like sometimes I start a chat and get a "dud" instance. Like it will suck at some basic things. And other times it's great.
What I found curious the other evening is that I asked the same question to both GPT and Gemini and they gave the exact same response, word for word. I thought they were totally different things. :unsure:
 
ChatGPT Search is here to take on Google

The much-anticipated search function of ChatGPT is now live. It's called ChatGPT Search, and searches the web for you. The twist is that it's integrated into ChatGPT, and ChatGPT may choose to search for you for some things you ask it. You can also manually click/tap the Search option after you input anything into the text box.

OpenAI says it "blends the benefits of a natural language interface with the value of up-to-date sports scores, news, stock quotes, and more". The company has partnered with "news and data providers to add up-to-date information and new visual designs for categories like weather, stocks, sports, news, and maps".

Chats with ChatGPT now include links to sources, like news articles and blog posts, which should have been there from day one, but better late than never. You do however need to click the Sources button below the response to then have a sidebar open with all the references.


 
Write unit test using accurate expectation. Get fail result testing python module. Chatgpt: "Let's change your test to make the problem go away."

jerry-seinfeld-nope.gif
 
It’s presented on Apple intelligence supported devices as a standalone app only being queried once once you need to consult the cloud.of AI
 
I bought a month’s upgrade to ChatGPT, thinking I could use it to do some data processing for me, and in the end I cancelled at the end of the month in disgust. ChatGPT is like one of those people who knows what you want it to do, and can rephrase your request in a confirmation of understanding, but the implementation … therein lies the problem. I tried getting it to extract a member directory from a Word document and save the info into Excel, and gave it a large number of prompts over how to complete this simple task that any intern could grudgingly do: and it failed spectacularly (failing to match the correct lines to columns in the Excel). It kept apologising and reprocessing and returning trash data to me … and eventually suggested I should try a different source document. And then it analysed my website and said it would give me a detailed PDF to download with UX advice, complete with graphs and competitor examples like I’d requested … and said that a job of this scale will take it five days and it would give me daily updates (which I requested). Needless to say it gave me zero updates, and when prompted to continue the job, it kept giving me around a 1-page PDF that looked like a kid vomited it out, even though in its confirmations to me it kept promising that the next document would be in-depth and contain over ten pages of feedback. So yeah, great promises … terrible implementation for my use cases.

Maybe one day it will be a bit smarter, but for now I got tired training ChatGPT for free.
 
I tried getting it to extract a member directory from a Word document and save the info into Excel, and gave it a large number of prompts over how to complete this simple task that any intern could grudgingly do: and it failed spectacularly (failing to match the correct lines to columns in the Excel). It kept apologising and reprocessing and returning trash data to me … and eventually suggested I should try a different source document.

I'd love to see how the data was structured in the word file. Mind sharing a snippet?
 
I'd love to see how the data was structured in the word file. Mind sharing a snippet?
I don’t have the docs on-hand now on my cellphone, but here’s the kind of errors it was making (and blaming the document’s OCR from an original PDF scan):

IMG_4238.png

I expect the premium language model to recognise it is creating junk data: if it doesn’t know basic validation (like don’t place parts of physical addresses under a column labelled “Tel. no” and don’t place “Tel. Email” under the email column) then I can’t help it. It just doesn’t understand context correctly, and that’s gonna catch it out if it doesn’t scan for it and correct itself).
 
I don’t have the docs on-hand now on my cellphone, but here’s the kind of errors it was making (and blaming the document’s OCR from an original PDF scan):

View attachment 1770982

I expect the premium language model to recognise it is creating junk data: if it doesn’t know basic validation (like don’t place parts of physical addresses under a column labelled “Tel. no” and don’t place “Tel. Email” under the email column) then I can’t help it. It just doesn’t understand context correctly, and that’s gonna catch it out if it doesn’t scan for it and correct itself).

That's why I want to see the actual data in the Word file. If it's not formatted correctly, even ChatGPT won't be able to clean it up properly. How can it know if a particular word or number belongs to a number or address etc if it's not formatted correctly with delimiters?

Some think chatgpt can produce miracles. It can't. You need to understand how to use it and what its limitations are. Once you do, it's very powerful.

Also, if the data already had delimiters in the Word file, why not just copy it into a spreadsheet and use text to columns?
 
That's why I want to see the actual data in the Word file. If it's not formatted correctly, even ChatGPT won't be able to clean it up properly. How can it know if a particular word or number belongs to a number or address etc if it's not formatted correctly with delimiters?

Some think chatgpt can produce miracles. It can't. You need to understand how to use it and what its limitations are. Once you do, it's very powerful.

Also, if the data already had delimiters in the Word file, why not just copy it into a spreadsheet and use text to columns?

Yeah I've used it to great effect, but you need to understand what its good at and where its useless.

I like to think of it as a new intern to the dev world that has no actual coding experience but for some reason has a photographic memory and has spent time scanning everything on the internet.

Recently I asked it to create a Roslyn C# parser for code sequences for me. It created a great stack implementation that only someone experienced in code parsing would have thought of. But on the other hand it immediately got lost in the smaller details and the code by itself wouldn't have worked. As an experienced dev I could use its code as a base and have it expand on it, while fixing the little idiocies by hand.
 
That's why I want to see the actual data in the Word file. If it's not formatted correctly, even ChatGPT won't be able to clean it up properly. How can it know if a particular word or number belongs to a number or address etc if it's not formatted correctly with delimiters?

Some think chatgpt can produce miracles. It can't. You need to understand how to use it and what its limitations are. Once you do, it's very powerful.

Also, if the data already had delimiters in the Word file, why not just copy it into a spreadsheet and use text to columns?
That’s my exact point. If a junior intern can use their eyes to look at a document and figure out the relationships between data, Chat GPT doesn’t get a free pass to stuff up. AI is meant to make work easier out of the box, I’m not about to chew its food for it. Mostly because I don’t have the time to waste. Maybe one day Chat GPT will have a modicum of actual intelligence, but for now I’ve come to realise it’s just a great parrot.
 
People seem to forget that ChatGPT is first and foremost an LLM - its primary purpose was more fluent communication and MAYBE creative writing. Everything else was tacked on.
 
Top
Sign up to the MyBroadband newsletter
X