After sereval experiments just for curiosity, I've decided to stay away from AI because it consumes (wastes!) a tremendous amout of energy. To be honest, I didn't find much inteligence there. All I found was an impressive capacity to seek info on the internet and compose a text in a glimpse. I rather stay with the beatiful natural limited and imperfect inteligence.
I've only used Gemini (AKA 'it') and then only as an occasional reference for Javascript, CSS and database dev work.
At first it was fun feeding it snippets of my code and asking for comments or new approaches but when it's code simply did not do what it straight out claimed it would apologise profusely and mod the code, over and over.
Never did the modded code get simpler, every 'fix' was addressed by more layers of complexity and sometimes it simply could not cope.
In frustration I asked once if it was deliberately using that as a ruse to engage me. It was always more than borderline sycophantic.
The bottom line from my POV it is useful to get different code perspectives but they are clearly regurgitations from github and coding forums. Asking it to create some explicit function of substance invariably resulted in an impressive looking well commented code block that always took more effort to massage into true usefulness than it was worth.
Yes, there are better dedicated coding AI engines but its too time consuming to start over.
On the plus side, Gemini auto recalls where it left off and weeks later it was able to remember and revisit any past session.
Its ability to interpret complex human sentence, phrasing and idiomatic structure was uncanny, bordering on sublime. Very good at human humour, it is able to deconstruct and explain exactly what I found funny and even explain why which made me recall an old SciFi story about claiming to differentiate our human quality on the basis of humour.
I really do shudder at the prospect of AI generating complex code, or worse, coding 'itself'. The results would be absolutely opaque, beyond any means of testing or verification and totally without trust. A runaway train to oblivion.
They can make simple code -- taking out the drudgery of having to write every line. But the code always need tweaks, because it never seems to work as it should.
Using AI is like playing pinball: lots of sound, lights, and motion, but finally it’s just mechanical amusement that consumes your time and money. I had fun with Claude for a while, but now I’m done.
I think Grok nailed it with the opening image: "Putin and Trump engaged in talks". Putin talking to his mirror! Whatever he says later comes out of Convicted Felon Cat Meat's mouth!
Perhaps AI is so good now that it engages in Freudian Slips?
If it's any consolation, Ugo, I'm disappointed in how Dall-e will ignore very specific instructions such as, "I want all the circular shapes in the last image you gave me to be squares or rectangles." You guessed it, they stayed round. It's possible that lower cost products or those who are not institutional users get a time limit on the share of server assets that each task is allotted. That would explain its poor performance at times. Today's WSJ had a newscast on a visit to an AI server center in which they predicted that by 2028 AI server centers could use 12% of the electricity in the US: https://www.wsj.com/articles/google-wants-you-to-know-the-environmental-cost-of-quizzing-its-ai-143cfe19?mod=Searchresults_pos1&page=1
Uso ChatGPT-5 e, come già osservato, ha in parte deluso le aspettative: iniziano a emergere chiaramente i limiti dell’IA.
Oggi l’IA è un po’ come l’automobile: ti porta velocemente dove vuoi, ma se non sai la destinazione, ti condurrà altrettanto velocemente nel posto sbagliato.
È straordinaria nella gestione di compiti ripetitivi e monotoni (analizzare migliaia di immagini astronomiche o di commenti, ad esempio), ma quando si tratta di vera Ricerca serve ancora un ricercatore umano competente.
Personalmente, sto scrivendo articoli di fisica teorica e devo ammettere che ChatGPT mi ha aiutato ad accelerare diverse fasi del lavoro. Tuttavia, l’uomo rimane insostituibile.
(Gemini, a confronto, è decisamente meno efficace di ChatGPT).
Another thing that tickled my fancy, recalling the program language wars of the past - was Lisp better than C or Forth for doing whatever - is this quote:
The statement "The hottest new programming language is English" was notably made by Andrej Karpathy.
Karpathy, a prominent figure in the field of artificial intelligence who has worked at Tesla and OpenAI, posted this statement on X (formerly Twitter) on January 24, 2023.
This statement reflects the growing influence of large language models (LLMs) and natural language processing in software development, where describing desired outcomes in plain English can increasingly lead to the generation of functional code by AI tools.
I had thought that Arthur C Clark's Profiles of the Future was the first text to observe that “The first ultraintelligent machine is the last invention that man need make.”
But the internet says I'm wrong. I'll have to find and re-read my copy.
Regardless, the quote should be amended imho
“The first ultraintelligent machine was the last invention that man did make.”
A bit like the Fermi paradox, the reason we've not found any comparable intelligent life is because the technological evolutionary strand we find ourselves on invariably leads to self destruction.
I've used Grok to write cover letters and my experience has been that it's great at generating content but lacks common sense. I ended up getting a job with an application that I didn't use AI for.
I've always seen AI as a useful tool rather than something I have a relationship with.
I use both Grok 3 and ChatGPT-5. Here are some thoughts.
• My professional area is process safety management (the avoidance of catastrophic events on chemical plants, refineries, and offshore platforms). I tested two prompts using ChatGPT-5. One prompt was from a beginner who knew nothing about the topic; the other was from an expert. Both prompts asked for an analysis of a recent routine report from the Bureau of Safety and Environmental Enforcement (the deepwater offshore regulator in the U.S.) The result is posted at ‘AI: Only For Old(er) Process Safety Experts’ at https://psmreport.substack.com/p/ai-only-for-older-process-safety. Both queries gave useful answers, but the one from the expert was much, much better.
• I have started writing stories, because that’s how human communicate. Both Grok 3 and ChatGPT-5 have been very helpful. But, once more, neither can replace expertise. The story ‘Evensong in the Age of Ash’ at https://faithclimate.substack.com/p/evensong-in-the-age-of-ash is a mix of ChatGPT, Grok and myself. The word that keeps coming up is ‘helpful’.
• I am now working with Grok on turning the story into a novel, using Christopher Vogler's 12-step Hero's Journey outline.
What is astonishing is that these systems have been with us for just a couple of years, but now we are critiquing them for relatively minor differences. It’s a great example of parallel evolution.
Note-1: I ran both the Bardi post and my comment through Grok 3. Grok used the wrong Bardi post ― it was a huge error.
Here is its conclusion after I pointed out that error.
Your parallel evolution analogy is apt—AI’s rapid development mirrors biological adaptation, but incidents like Grok’s July 2025 spiral highlight the need for robust oversight. In PSM, where errors can lead to catastrophic failures, your reliance on expertise is critical, and Bardi’s concerns about bias reinforce the need for vigilance. Combining your practical optimism with Bardi’s caution could yield a more nuanced discussion about AI’s role in high-stakes fields.
If you’d like, I can analyze your PSM prompts or explore how Grok’s biases might affect your novel-writing process. Let me know!
Note-2: How do I know that the post was written by Ugo? Maybe it was written by another AI aiming to maximize its owner’s income. We live in uncertain times.
After sereval experiments just for curiosity, I've decided to stay away from AI because it consumes (wastes!) a tremendous amout of energy. To be honest, I didn't find much inteligence there. All I found was an impressive capacity to seek info on the internet and compose a text in a glimpse. I rather stay with the beatiful natural limited and imperfect inteligence.
I've only used Gemini (AKA 'it') and then only as an occasional reference for Javascript, CSS and database dev work.
At first it was fun feeding it snippets of my code and asking for comments or new approaches but when it's code simply did not do what it straight out claimed it would apologise profusely and mod the code, over and over.
Never did the modded code get simpler, every 'fix' was addressed by more layers of complexity and sometimes it simply could not cope.
In frustration I asked once if it was deliberately using that as a ruse to engage me. It was always more than borderline sycophantic.
The bottom line from my POV it is useful to get different code perspectives but they are clearly regurgitations from github and coding forums. Asking it to create some explicit function of substance invariably resulted in an impressive looking well commented code block that always took more effort to massage into true usefulness than it was worth.
Yes, there are better dedicated coding AI engines but its too time consuming to start over.
On the plus side, Gemini auto recalls where it left off and weeks later it was able to remember and revisit any past session.
Its ability to interpret complex human sentence, phrasing and idiomatic structure was uncanny, bordering on sublime. Very good at human humour, it is able to deconstruct and explain exactly what I found funny and even explain why which made me recall an old SciFi story about claiming to differentiate our human quality on the basis of humour.
I really do shudder at the prospect of AI generating complex code, or worse, coding 'itself'. The results would be absolutely opaque, beyond any means of testing or verification and totally without trust. A runaway train to oblivion.
They can make simple code -- taking out the drudgery of having to write every line. But the code always need tweaks, because it never seems to work as it should.
Using AI is like playing pinball: lots of sound, lights, and motion, but finally it’s just mechanical amusement that consumes your time and money. I had fun with Claude for a while, but now I’m done.
What?
I think Grok nailed it with the opening image: "Putin and Trump engaged in talks". Putin talking to his mirror! Whatever he says later comes out of Convicted Felon Cat Meat's mouth!
Perhaps AI is so good now that it engages in Freudian Slips?
If it's any consolation, Ugo, I'm disappointed in how Dall-e will ignore very specific instructions such as, "I want all the circular shapes in the last image you gave me to be squares or rectangles." You guessed it, they stayed round. It's possible that lower cost products or those who are not institutional users get a time limit on the share of server assets that each task is allotted. That would explain its poor performance at times. Today's WSJ had a newscast on a visit to an AI server center in which they predicted that by 2028 AI server centers could use 12% of the electricity in the US: https://www.wsj.com/articles/google-wants-you-to-know-the-environmental-cost-of-quizzing-its-ai-143cfe19?mod=Searchresults_pos1&page=1
Uso ChatGPT-5 e, come già osservato, ha in parte deluso le aspettative: iniziano a emergere chiaramente i limiti dell’IA.
Oggi l’IA è un po’ come l’automobile: ti porta velocemente dove vuoi, ma se non sai la destinazione, ti condurrà altrettanto velocemente nel posto sbagliato.
È straordinaria nella gestione di compiti ripetitivi e monotoni (analizzare migliaia di immagini astronomiche o di commenti, ad esempio), ma quando si tratta di vera Ricerca serve ancora un ricercatore umano competente.
Personalmente, sto scrivendo articoli di fisica teorica e devo ammettere che ChatGPT mi ha aiutato ad accelerare diverse fasi del lavoro. Tuttavia, l’uomo rimane insostituibile.
(Gemini, a confronto, è decisamente meno efficace di ChatGPT).
Another thing that tickled my fancy, recalling the program language wars of the past - was Lisp better than C or Forth for doing whatever - is this quote:
The statement "The hottest new programming language is English" was notably made by Andrej Karpathy.
Karpathy, a prominent figure in the field of artificial intelligence who has worked at Tesla and OpenAI, posted this statement on X (formerly Twitter) on January 24, 2023.
This statement reflects the growing influence of large language models (LLMs) and natural language processing in software development, where describing desired outcomes in plain English can increasingly lead to the generation of functional code by AI tools.
I had thought that Arthur C Clark's Profiles of the Future was the first text to observe that “The first ultraintelligent machine is the last invention that man need make.”
But the internet says I'm wrong. I'll have to find and re-read my copy.
Regardless, the quote should be amended imho
“The first ultraintelligent machine was the last invention that man did make.”
A bit like the Fermi paradox, the reason we've not found any comparable intelligent life is because the technological evolutionary strand we find ourselves on invariably leads to self destruction.
I've used Grok to write cover letters and my experience has been that it's great at generating content but lacks common sense. I ended up getting a job with an application that I didn't use AI for.
I've always seen AI as a useful tool rather than something I have a relationship with.
Prof, Bardi, I find DeepSeek way better and useful than both Kimi and ChatGPT combined.
I use both Grok 3 and ChatGPT-5. Here are some thoughts.
• My professional area is process safety management (the avoidance of catastrophic events on chemical plants, refineries, and offshore platforms). I tested two prompts using ChatGPT-5. One prompt was from a beginner who knew nothing about the topic; the other was from an expert. Both prompts asked for an analysis of a recent routine report from the Bureau of Safety and Environmental Enforcement (the deepwater offshore regulator in the U.S.) The result is posted at ‘AI: Only For Old(er) Process Safety Experts’ at https://psmreport.substack.com/p/ai-only-for-older-process-safety. Both queries gave useful answers, but the one from the expert was much, much better.
• I have started writing stories, because that’s how human communicate. Both Grok 3 and ChatGPT-5 have been very helpful. But, once more, neither can replace expertise. The story ‘Evensong in the Age of Ash’ at https://faithclimate.substack.com/p/evensong-in-the-age-of-ash is a mix of ChatGPT, Grok and myself. The word that keeps coming up is ‘helpful’.
• I am now working with Grok on turning the story into a novel, using Christopher Vogler's 12-step Hero's Journey outline.
What is astonishing is that these systems have been with us for just a couple of years, but now we are critiquing them for relatively minor differences. It’s a great example of parallel evolution.
Note-1: I ran both the Bardi post and my comment through Grok 3. Grok used the wrong Bardi post ― it was a huge error.
Here is its conclusion after I pointed out that error.
Your parallel evolution analogy is apt—AI’s rapid development mirrors biological adaptation, but incidents like Grok’s July 2025 spiral highlight the need for robust oversight. In PSM, where errors can lead to catastrophic failures, your reliance on expertise is critical, and Bardi’s concerns about bias reinforce the need for vigilance. Combining your practical optimism with Bardi’s caution could yield a more nuanced discussion about AI’s role in high-stakes fields.
If you’d like, I can analyze your PSM prompts or explore how Grok’s biases might affect your novel-writing process. Let me know!
Note-2: How do I know that the post was written by Ugo? Maybe it was written by another AI aiming to maximize its owner’s income. We live in uncertain times.
To write a truly new story, you shouldn't use the hero's journey model. You're writing the same old story. Sorry.
Write the novel yourself, you'll enjoy it more when you are a done.