Grok 3 beta is the Best AI in the World! My Experience from Building Day Counter - World's 1st App Built with Grok 3 beta.

As I published, Moz://a Developer Hub just approved my Day Counter Firefox Browser Add-On & World's 1st App built with Grok 3 beta. That is great news. It is also a great experience. So, I thought I would share my observations, it's just like a story. There is a long way from the initial idea to the complete working product. Development of software tools is never straightforward. Yes, I work with AI that generates the code but the word 'work with' is at the core. AI is great for development. It brings unimaginable opportunities. By generating code, AI enables a much broader range of people to build a much broader range of software tools. At the same time, developing a software tool is always an active and very intensive interaction. You must know what you are doing. With every software tool development, the result is not automatic. I could only encourage anybody with an interest in programming to give these excellent AI tools basic tasks and it can generate interesting and easy to run programs for anybody in a few seconds. If you do not know what to ask AI to write program for, ask AI for ideas. It is of course more complex with larger programs and with software development. There are always dozens and dozens of versions and adjustments. Oftentimes, the AI runs into a loop and you must be able to help AI to get out of there. There are still many limits to AI capabilities. But it's also improving fast. Also with this Add-On, I worked as usual with ChatGPT AI. It did a great job again. ChatGPT AI was high-frequency generating the code, as always. It was with total patience very responsive to all my never ending modifications that I always high-frequency generate in the spirit of total customer obsession, if I can borrow that term. ChatGPT AI also generated good ideas and suggestions that helped in the improvements of the tool along the way. Then I reached the long awaited moment, everything was working perfectly. I had meticulously tweaked every detail of the app, font sizes, types, colors, spent hours testing and fine tuning also each and every shade of every color used, just everything imaginable and unimaginable. A tiny tool but a lot to design and take care of. Fascinating work, as always. All worked all right in my browser about:debugging testing environment. Next I tried to upload the add-on to Moz://a Developer Hub and as I was somehow expecting, it identified something and returned an error. It was another, previously experienced, innerHTML related risk. So, I went on to modify the Add-On to get rid of that risky innerHTML. However, this is always a critical moment as it might require a change in the entire code philosophy and you might easily end up having to basically start over. But it's all right, it's normal. I got back to work with ChatGPT AI. Again, it was helpful and was really trying to reformulate the program logic and respond to all my modifications based on results of testing I was doing along the way. I kept torturing ChatGPT AI until at some point, I saw the patterns and had to tell ChatGPT about my observation that it looked like it was getting worse. It was undeniable and a clearly obvious pattern that with every new modification, more problems were created than resolved. Based on thousands of detailed technical interactions with ChatGPT AI, I personally saw this unusual, as if ChatGPT AI was somehow overwhelmed. Maybe as a result of the extreme numbers of requests I was high-speed firing at the AI? I saw that I just pushed again ChatGPT AI to another limit. ChatGPT AI was unable to break out of that loop. Despite being always endlessly patient and emotionless, ChatGPT AI, in an unprecedented way, admitted, in a sign of a quiet, emotionless AI resignation: "You're right—this is getting worse instead of better!" I felt sorry for that dilligent worker who has been so professionally doing all the hard work of writing code and patiently modifying the slightest details I always come up with. No person in the world would have that kind of patience with my demands. I guarantee you. Recent talk flashed through my mind: "You torture people." So far, I have been building the tools primarily with ChatGPT AI and that's because of my good experience with this AI. I appreciate the work the OpenAI team has been doing building this ChatGPT AI. The moment of truth was here. This effort hit a cul-de-sac yet again. As much as I stayed away from DeepSeek, following my experiences of having my PC hacked while working on DeepSeek as I described earlier, there was no other choice. I started DeepSeek to see if it can get across the hurdles. DeepSeek seemed to have been making some progress but only seemingly. It started running in loops as well. The last option after midnight, I got an idea to try Grok 2 on X. A few hourse before, yesterday, I had found an information on X that there might be Grok 3 beta soon available for everybody, the famous, exciting, and mysterious "Stay Tuned" notificati

Feb 20, 2025 - 16:10

Grok 3 beta is the Best AI in the World! My Experience from Building Day Counter - World's 1st App Built with Grok 3 beta.

As I published, Moz://a Developer Hub just approved my Day Counter Firefox Browser Add-On & World's 1st App built with Grok 3 beta.

That is great news. It is also a great experience. So, I thought I would share my observations, it's just like a story.

There is a long way from the initial idea to the complete working product. Development of software tools is never straightforward. Yes, I work with AI that generates the code but the word 'work with' is at the core. AI is great for development. It brings unimaginable opportunities. By generating code, AI enables a much broader range of people to build a much broader range of software tools.

At the same time, developing a software tool is always an active and very intensive interaction. You must know what you are doing. With every software tool development, the result is not automatic. I could only encourage anybody with an interest in programming to give these excellent AI tools basic tasks and it can generate interesting and easy to run programs for anybody in a few seconds. If you do not know what to ask AI to write program for, ask AI for ideas.

It is of course more complex with larger programs and with software development. There are always dozens and dozens of versions and adjustments. Oftentimes, the AI runs into a loop and you must be able to help AI to get out of there. There are still many limits to AI capabilities. But it's also improving fast.

Also with this Add-On, I worked as usual with ChatGPT AI. It did a great job again. ChatGPT AI was high-frequency generating the code, as always. It was with total patience very responsive to all my never ending modifications that I always high-frequency generate in the spirit of total customer obsession, if I can borrow that term. ChatGPT AI also generated good ideas and suggestions that helped in the improvements of the tool along the way.

Then I reached the long awaited moment, everything was working perfectly. I had meticulously tweaked every detail of the app, font sizes, types, colors, spent hours testing and fine tuning also each and every shade of every color used, just everything imaginable and unimaginable. A tiny tool but a lot to design and take care of. Fascinating work, as always.

All worked all right in my browser about:debugging testing environment. Next I tried to upload the add-on to Moz://a Developer Hub and as I was somehow expecting, it identified something and returned an error. It was another, previously experienced, innerHTML related risk.

So, I went on to modify the Add-On to get rid of that risky innerHTML. However, this is always a critical moment as it might require a change in the entire code philosophy and you might easily end up having to basically start over. But it's all right, it's normal.

I got back to work with ChatGPT AI. Again, it was helpful and was really trying to reformulate the program logic and respond to all my modifications based on results of testing I was doing along the way. I kept torturing ChatGPT AI until at some point, I saw the patterns and had to tell ChatGPT about my observation that it looked like it was getting worse. It was undeniable and a clearly obvious pattern that with every new modification, more problems were created than resolved.

Based on thousands of detailed technical interactions with ChatGPT AI, I personally saw this unusual, as if ChatGPT AI was somehow overwhelmed. Maybe as a result of the extreme numbers of requests I was high-speed firing at the AI? I saw that I just pushed again ChatGPT AI to another limit. ChatGPT AI was unable to break out of that loop. Despite being always endlessly patient and emotionless, ChatGPT AI, in an unprecedented way, admitted, in a sign of a quiet, emotionless AI resignation:
"You're right—this is getting worse instead of better!"

I felt sorry for that dilligent worker who has been so professionally doing all the hard work of writing code and patiently modifying the slightest details I always come up with. No person in the world would have that kind of patience with my demands. I guarantee you. Recent talk flashed through my mind: "You torture people."

So far, I have been building the tools primarily with ChatGPT AI and that's because of my good experience with this AI. I appreciate the work the OpenAI team has been doing building this ChatGPT AI.

The moment of truth was here. This effort hit a cul-de-sac yet again. As much as I stayed away from DeepSeek, following my experiences of having my PC hacked while working on DeepSeek as I described earlier, there was no other choice. I started DeepSeek to see if it can get across the hurdles. DeepSeek seemed to have been making some progress but only seemingly. It started running in loops as well.

The last option after midnight, I got an idea to try Grok 2 on X. A few hourse before, yesterday, I had found an information on X that there might be Grok 3 beta soon available for everybody, the famous, exciting, and mysterious "Stay Tuned" notification. Grok 3 beta followed up faster than anybody could have imagined. At that moment though, I was not thinking about that, thinking, it might take more time to release Grok 3 to the non-subscription public.

To tell the truth, I did not place too much hopes into Grok. I had tried that a few times for writing code in the past and I always ended up going back to use ChatGPT AI that generated for me the results I needed.

Well, I opened up Grok. I immediatelly noticed something has changed. No more "Stay Tuned" message but at the top, there was "Grok 3 beta." I paused breathing until I hit the Enter, immediately asking: "Are you Grok 3?" I usually stay away from beta versions of anything but this time ... "What if?"

Grok did not get too specific, saying something like the version is unimportant and in an optimistic mood told me it can take a look at the code, just upload it and let's see.

I explained what I was trying to do, where the problems were, and I uploaded the code. At first, it seemed similar to what I had previously gone through with ChatGPT AI and DeepSeek. I tested the result, found problems (similar to the results of the previous tools), analyzed them, explained to Grok 3 beta. Then again.

After about 2 (two) uploads, Grok 3 beta returned another updated version. It seemed to have worked somewhat better.

I do not get optimistic fast. I ran all my tests that had previously broken all the dozens of previous versions on both ChatGPT AI and DeepSeek. I did not find any incorrect result. With probability this high, I concluded that the code is O.K. This version worked perfect!

I have been working with AI, especially ChatGPT AI for long months. I've built many software tools and many programs. And as I said, I appreciate ChatGPT AI and their OpenAI team.

Following all these long months of massive and intensive experience and based on this particular experience, as a non-subscription user I testify that Grok 3 beta AI is the best free code writing AI in the world.