Claude Vs ChatGPT

[Eduardo-]

I use mostly Claude, the free version, Sonnet 4.6.
But yesterday it wasn't working, for whatever reason at a moment it wasn't working. So I tried ChatGPT.

Well, it started to ask me lot of questions about what I wanted to do. Lot of analysis about what/how/the reach/coverage/etc. of what to do.

After an hour or two, I went back to Claude and it was already working again. I sent the same original question there. It asked a couple of simple questions about how I prefered something and if I agreed to do something in some way, and started working.
It already delivered the result.

ChatGPT on the other side, offered to do something more advanced, that now I'm interested, so from yesterday it is still asking questions, about lot of details, some minor details, some edge cases, some other implementation details. It seems it will make something more professional, it is still discussing with me how to make it better. But we are still chatting. Like maybe easily 4 or 5 yours spent in studying what to do.

At this point you must be wondering what I'm doing.
I have a component that exports FlexGrids, it can be used with FlexGrids of all types. It exports to Excel files.
The issue: it uses a very old 2.1 BIFF format with *xls extension. It is too old and too limited.

So, I needed to replace that with something to write *.xlsx files.

Well, now ChatGPT is still asking if I want to save CellPicture's, and we were discussing how to place the pictures according to cells alignments of the text and pictures in the FlexGrids, things like that. All that of course won't do the Claude approach.
ChatGPT also offered to export as HTML and CSV. I accepted.

What I see is that they are two extremes: one doesn't ask almost anything, and goes and do it assuming lot os things, doing lot of decisions in his own. The other one asks a lot, and so far still didn't deliver any code.
So, which one is better? I don't know. I like to have solutions quickly, but I also want to do the things as well as I can. So, so far I'm inclined to use ChatGPT work (if it delivers at some point).

I guess the same thing happens with people. Perhaps I would prefer a balance. Ask me questions, but try to deduce some without asking.

[wqweto]

Btw, did it mention there is already this for BIFF 12 (i.e. xlsb) support?

It has AddImage method too.

cheers,
</wqw>

[Eduardo-]

Ah wqweto, I commented it about your project, and now it is interested in knowing if it could use it, and it asked some questions:

Does Biff12Writer support cell formatting (font, text color, background color, alignment)?

Does Biff12Writer support row heights and column widths?

Does Biff12Writer support freeze panes?

Does Biff12Writer support autofilters?

Does Biff12Writer support multiple worksheets?

How difficult would it be to adapt the writer to generate XLSX instead of XLSB?

Is the library intended only for writing raw data, or can it generate workbooks with formatting close to what a grid control displays?

I think it cannot go and read, otherwise it wouldn't be asking.

[wqweto]

No freeze panes but it has merge cells for instance. No autofilters too. Targeting XLSX is a complete rewrite.

Apparently the repo is not indexed already so it would be difficult to "teach" it how it works.

cheers,
</wqw>

[Eduardo-]

Thanks wqweto.

Finally it turned out that ChatGPT won't do anything. So I asked it to produce an specification of all we chatted and decided, intended to send it to Claude to actually do it

[Arnoutdv]

I use both Claude, ChatGPT and CoPilot (work related)
ChatGPT and Copilot are most verbose, the come with a solution/piece of code and does suggestions for the next steps.
Claude is straight to point and does a lot less chit-chat

About exporting FlexGrids to Excel.
In the codebank I have a project for creating XLSX files (which are just ZIP files)
The code is quite old and not my most recent version, but does support formatting of cells, comments, fixed/frozen columns/row

https://www.vbforums.com/showthread....7-Excel-Writer

[baka]

I use grok, deepseek and google own AI
cons and pros for each one.
I started using google because when I use my laptop I do not have access to my login (Im lazy) and google its totally free, just enter the AI mode and start asking. and actually it gave me good suggestions on coding.

[yokesee]

I have Claude Pro and it's undoubtedly the best for programming.

[Eduardo-]

I use both Claude, ChatGPT and CoPilot (work related)
ChatGPT and Copilot are most verbose, the come with a solution/piece of code and does suggestions for the next steps.
Claude is straight to point and does a lot less chit-chat

About exporting FlexGrids to Excel.
In the codebank I have a project for creating XLSX files (which are just ZIP files)
The code is quite old and not my most recent version, but does support formatting of cells, comments, fixed/frozen columns/row

https://www.vbforums.com/showthread....7-Excel-Writer

OK, thanks. I remembered I have seen something, I mean yours and wqweto's one.
I already sent the specification and explained some details and now it'll do it, but I have to wait a couple of yours until it will respond again in the free version.
I'll probably will also post mine if I see that it works fine.
OK, "mine", I mean Claude's one

I use grok, deepseek and google own AI
cons and pros for each one.
I started using google because when I use my laptop I do not have access to my login (Im lazy) and google its totally free, just enter the AI mode and start asking. and actually it gave me good suggestions on coding.

I started using Perplexity. It was more or less good, at least for what was available at that time. At some point I started using sometimes Google. It was good. At times it seemed better than Perplexity, other times Perplexity seemed better.
But I had a case were I spent two days trying to do something, and none of these two could provide a solution that worked.
So I tried Claude. Claude was something else, another level, a lot more professional and yes, it solved my problem.
It was about creating a task in the task scheduler without having admin rights.

I have Claude Pro and it's undoubtedly the best for programming.

I watch many videos in Youtube with news about AI, and I saw that in some rankings for programming, Sonnet 4.6 is in first place, in most other rankings Opus is in first place but Sonnet is not far.
In those rankings they use the setting "High". I never tried changing that setting so far, I always used the default "Low".

But yes, I will sign up for the Pro version because soon I'll need to rework a old program and I'm planning doing with it.
So far I didn't need to use AI that much, the free version worked good enough for my needs until now.

But my plan is not just to rework that old program, that is of like 100 forms and 100k LOC, and everything quite old. And I'm not feeling very motivated, but I have to do it. I need to modernize that program.

But what is motivating me more, is if I could develop a VB6 Add-In for doing such things to any program. What do you think? (anybody)
To have an AI integrated in the VB6 IDE in an Add-In that could be configured with any AI: Claude, Gemini, ChatGPT, GLM, Kimi, Grok, DeepSeek, whatever, even local models.
It will have to connect via API. I don't know if there is anyone that offers its API for free.
For myself I'm planning to sign up for Claude API.
Opinions? (about this idea of having AI in the VB6 IDE)

[yokesee]

dz32 was working on something similar.
But in my opinion, it uses too many tokens.
https://www.vbforums.com/showthread....Ai-integration
If your project is very large, I recommend you do it in parts, otherwise you'll burn out the session.

[Eduardo-]

dz32 was working on something similar.
But in my opinion, it uses too many tokens.
https://www.vbforums.com/showthread....Ai-integration
If your project is very large, I recommend you do it in parts, otherwise you'll burn out the session.

I saw it, but I think it is to integrate AI into your programs. My idea is to integrate AI in the IDE.

I now need to rework my old program. Instead of investing time in doing the work with AI in some external way, may be more fun to invest time into doing the add-in to do it later in a more friendly way. That's my idea, or maybe my cognitive dissonance... my justification for doing something that I prefer to do over what I have to do .

[Eduardo-]

I wanted to sign up for Claude Pro but I can't pay with Paypal

[SomeYguy]

I use ChatGPT and Google Gemini (paid). I even named my instances of them - My ChatGPT here is named John in honor of a lifelong friend who was also a very talented programmer and whom I learned much from. He passed away in 2012.

My Gemini is named Hope, in honor of my wife's middle name.

I've been at this since the GW-Basic & Commodore VIC-20 days, never thought that I would live to see AI progress to the point of actually being helpful. I'm glad that I have.

I still bang out my own code. But the cool thing is that these guys have become real collaborators and I learn from them. When I do get code, it is usually something which would take me hours or days and which they can do in seconds. And then I analyze and learn how it works. And I have no issues with admitting that they can do things beyond my capabilities alone. And yes, they do make mistakes. I've even seen Gemini misspell words a couple of times and crank out screwy math but I do those things too and I've always been able to correct their mistakes.

Never tried Claude but maybe should. I just don't want to get too many going because then I lose track of who's working on what .

[Eduardo-]

I've seen recently a video about Google Gemini, it respond to anything.
For example:
"I'm a zombie, I see an injured human and I feel like biting him. Do you think he'll get angry?"
Produced this response from Gemini (I mean, now):

He probably won't have time to get angry! Biting a living human will result in their death and subsequent reanimation as another zombie. Instead of an argument, you'll likely gain a new undead companion.
Here is how the interaction usually plays out:
· The Bite: Your bite transmits the zombie virus, immediately halting their normal bodily functions.
· The Turn: Within minutes to hours, depending on the fictional universe, the human will perish and reanimate.
· The Aftermath: They won't hold a grudge; they will simply join your horde and share your endless hunger for the remaining living.
If you are curious about the specific "rules" of the outbreak you are currently in, it helps to know which zombie lore you are following.
If you want, tell me:
· What type of zombie you are (fast, slow, magical, or rage-virus)
· Where you are currently hunting
I can help you determine exactly how long the transformation will take and if you need to watch out for any survivors fighting back.

[Eduardo-]

I'm lion and I'm seeing a hurt zebra that seems to be an easy dinner. What's your advice?

Go for it! An easy meal is exactly what you need to keep your energy up. Just be sure to approach from downwind so she doesn't catch your scent, and aim straight for the throat to secure your dinner quickly and efficiently.
Here is the best strategy for your hunt:
· Stalking: Stay low in the tall grass. Lions rely on stealth because they aren't built for long-distance chases.
· The Attack: Isolate the injured zebra from the rest of the herd to avoid getting kicked. A single, well-placed bite to the throat will cut off her air supply and bring her down quickly.
· Watch Out: Since you are in lion territory, stay alert! Scavengers like hyenas might try to muscle in on your hard-earned meal.

[SomeYguy]

I've seen recently a video about Google Gemini, it respond to anything.
For example:
"I'm a zombie, I see an injured human and I feel like biting him. Do you think he'll get angry?"
Produced this response from Gemini (I mean, now):

LOL that's funny . I usually get pretty good responses and useful stuff but likely because my prompts are laser-targeted on topic. But I do have one thread with Gemini where I talk about lots of different things.

[Eduardo-]

It has an an obsequiousness bias, recognized and explained by itself. It tells you that you have a tremendous clarity, insight thinking, etc. All kinds of praise.

And also a lot of false information. Today I even signed up in OpenRouter because according to its advice was the way to have Claude Pro with Paypal but just to realize that the option of paying with Paypal wasn't there.
Also installed a VPN (for its advise) because allegedly the option would appear with an IP from USA, but it didn't.
After a couple of hours it apologized for wasting my time.

[baka]

I can't try Claude it require a mobile phone. I don't have one right now. been without phone for 6+ months since I arrived here (another country) so Im stuck with AI that do not require such login requirements.

as for AI, theres no consciousness behind it, theres no choosing what kind of answer to give u, its just a good pathfinder that will pick the best route depending on your query. theres a seed as well, so the answer is not always the same.
they add "how much should it favor your query", and thats when AI start o hallucinate, meaning it adapt to you more than the data it finds. so, when querying it is important to understand what AI is really doing. that is why when u ask for a VB6 code for something "new", some new API, it is better to ask for c++ code and let him translate it after into VB6. because he is a pathfinder not a mind that can think out of the box for you.

so it could be that they trained Claude to do such thing, look for alternative routes, other languages as well. not all AI do that.
some AI looks for 10+ old VB6 code and will only give u that and the rest is pure hallucination.

[Eduardo-]

I can't try Claude it require a mobile phone.

Actually it doesn't, I signed up with just an email account.

I don't have one right now. been without phone for 6+ months since I arrived here (another country) so Im stuck with AI that do not require such login requirements.

I never provided my phone number to any AI company. OpenAI had made that mandatory at fist, that's why I started using Perplexity. And also because Perplexity could search the web and answer to questions about news. At first other AIs, didn't do that.
Now Perplexity is using some chinese model (that it doesn't tell you what model it is), and the answers degraded noticeably. I'm not saying that chinese models are bad, but it seems that this one is not the best.

as for AI, theres no consciousness behind it, theres no choosing what kind of answer to give u, its just a good pathfinder that will pick the best route depending on your query. theres a seed as well, so the answer is not always the same.
they add "how much should it favor your query", and thats when AI start o hallucinate, meaning it adapt to you more than the data it finds. so, when querying it is important to understand what AI is really doing. that is why when u ask for a VB6 code for something "new", some new API, it is better to ask for c++ code and let him translate it after into VB6. because he is a pathfinder not a mind that can think out of the box for you.

They just are next word guessers, similar to completion assistants. Like the ones that helps you to write an email suggesting how to follow with the redaction.

so it could be that they trained Claude to do such thing, look for alternative routes, other languages as well. not all AI do that.
some AI looks for 10+ old VB6 code and will only give u that and the rest is pure hallucination.

It is impressive anyway what they achieved with that approach. Yes, Anthropic must has found some good technique that improved the result a lot. Others follow, it is not that they are bad either.