Hi everyone,

I’ve been following recent developments in AI and how they're being integrated into software development workflows. Tools like GitHub Copilot, ChatGPT, and AI-powered testing suites are changing the way developers write, review, and optimize code.

What fascinates me is how AI isn’t just assisting with code completion—it’s also influencing architecture decisions, bug detection, and even team collaboration. In your opinion, how far can we go with AI in the dev lifecycle?

Do you think we’ll ever reach a point where AI handles most of the actual coding, leaving humans to focus mainly on high-level logic and design?

Would love to hear your thoughts or any research you've come across.

Best

43 minutes ago, faisal.webdev said:

Hi everyone,

I’ve been following recent developments in AI and how they're being integrated into software development workflows. Tools like GitHub Copilot, ChatGPT, and AI-powered testing suites are changing the way developers write, review, and optimize code.

What fascinates me is how AI isn’t just assisting with code completion—it’s also influencing architecture decisions, bug detection, and even team collaboration. In your opinion, how far can we go with AI in the dev lifecycle?

Do you think we’ll ever reach a point where AI handles most of the actual coding, leaving humans to focus mainly on high-level logic and design?

Would love to hear your thoughts or any research you've come across.

Best

This is more of a philosophy question, bc ATM AI is demonstrably stupid in terms of intelligence as we recognise it.

Will it evolve to be better? Yes

Will it evolve to be smarter? Yes, no and maybe.

How will we know, if they break the glass ceiling and become a threat? Not before it's too late... 😉

23 hours ago, dimreepr said:

This is more of a philosophy question, bc ATM AI is demonstrably stupid in terms of intelligence as we recognise it.

Will it evolve to be better? Yes

Will it evolve to be smarter? Yes, no and maybe.

How will we know, if they break the glass ceiling and become a threat? Not before it's too late... 😉

That’s a great point — AI, as it stands today, isn’t truly “intelligent” in the human sense. It’s more like an incredibly fast, pattern-recognizing tool than a conscious being.

Will it get better? Definitely — we're already seeing huge leaps in reasoning, memory, and contextual understanding.
Smarter? That depends on how we define "smart." If it’s raw processing and decision-making, sure. But if we’re talking emotional intelligence, self-awareness, or creativity like humans — that’s still up for debate.

And your last point is the kicker — will we recognize the danger before it's too late? That’s the real challenge. The hope is that we build guardrails faster than we build the rocket ship. Let’s see how that goes... 👀

12 minutes ago, faisal.webdev said:

more like an incredibly fast, pattern-recognizing tool than a conscious being.

What are the key differences you see between those two things?

37 minutes ago, iNow said:

What are the key differences you see between those two things?

It's not just any question... 😁

"The new ChatGPTs lie like crazy. Hallucinations in every second answer.

The latest OpenAI artificial intelligence models for inference give false answers more often than older AIs. It's not clear why."

"A few days ago, as we wrote about, OpenAI released new models for inference - including the most important o3 and o4-mini. They perform better than their predecessors in some areas - especially coding and math. Now we have found that, unfortunately, they make up answers much more often than the older OpenAI models. They hallucinate in greater explicit numbers than the company's previous reasoning models - o1, o1-mini and o3-mini - as well as traditional “non-comprehending” OpenAI models such as GPT-4o."

"The response rate of new models with fake content is alarmingly high. OpenAI discovered that o3 showed hallucinations in response to as many as 33 percent of questions. This is a result achieved in a special PersonQA tool used by OpenAI.to measure the accuracy of the model's knowledge."

"The 33 percent response rate with hallucinations is roughly twice the hallucination rate of OpenAI's previous reasoning models, o1 and o3-mini. These get scores of 16 percent and 14.8 percent, respectively, in the tests here. o4-mini performed even worse in the PersonQA test. In its case, hallucinations occurred in as many as 48 percent of cases."

"That new inference models are more likely to lie has also been noticed by Transluce, a laboratory specializing in artificial intelligence research. Its researchers noted, for example, how o3 informed a user that it had run code on a 2021 MacBook Pro. “outside of ChatGPT,” and then copied the numbers into his answer. That's bogus, o3 can't do that."

"There are also reports that links to non-existent Web pages, for example, appear in application codes generated by the new models."

"No one knows why this is happening. Transluce, in an interview with TechCrunch, speculates that the increase in the number of hallucinations of the new models should be linked to the technique of teaching them through so-called reinforcement. In it, not only do their capabilities grow, but their greatest weaknesses are strengthened as well. "

"So now it seems that the AI industry has entered a bit of a dead end. Last year, it focused on developing inference models after techniques to improve traditional AI models began to show declining effectiveness. Reasoning seemed to improve the model's performance on many tasks, without the need for huge calculations and the use of huge amounts of data during training. However, it now seems that reasoning models also have greater hallucination than standard models. As we now know, they are starting to get so big that they often derail the sense of using AI."

(Translated by AI ;) )

All the code generated by the older ChatGPT did not work.. Almost useless for beginners, as they will have no idea where to start in order to get it to compile and work..

On 4/17/2025 at 1:17 PM, faisal.webdev said:

Will it get better? Definitely — we're already seeing huge leaps in reasoning, memory, and contextual understanding.
Smarter? That depends on how we define "smart." If it’s raw processing and decision-making, sure.

It seems like we're in a sort of Heisenberg uncertainty principle with the question (bolded), we have to know where to start the calculation i.e. what is a smart human?

On 4/17/2025 at 1:17 PM, faisal.webdev said:

And your last point is the kicker — will we recognize the danger before it's too late?

Of course not, even the superhero 'Captain hindsight' isn't that powerful...