No one ever complained about Google search, and begged for the company to roll back to a previous version of google.com.
Yet it seems no one likes GPT-5, the new flagship version of OpenAI’s chatbot. They’re apparently vocal about their disdain too:
Wearing the skin of my dead friend.
Like talking to a very overworked HR rep. (GPT-5 is awful)
It’s like my chatGPT suffered a severe brain injury and forgot how to read.
I wonder: why is that?
The answer may lie in our expectations of what something like ChatGPT, or Claude, or Gemini offers us on a daily basis. Is this magical thing called “AI” a tool? A way to help us be more productive? Or, is it a 24/7 friend and companion, ready and able in a pinch to offer us encouragement, or to console us, or to help keep us going?
AI as a Tool
Much like that first time we spoke to Alexa, and asked it to do something routine (“Alexa, turn on the lights”) and were awestruck that we could use our voice to interact with technology, that same feeling applies also to AI. Each of us likely had that a-ha moment. This is magical. A moment where there was no going back.
For me, that was when ChatGPT and Claude helped fix the server that runs this site.
Heading down to the IT Dungeon is never a fun thing. You go down there with a bunch of Stack Overflow bookmarks, random Reddit threads, and a few bits of bash scripts you found on a random blog. Somehow, you hope you can McGyver all of those bits into a working Nginx config file, or, maybe, a way to backup your WordPress database. With all of this duct tape in hand, you apply some sort of changes using commands and things you don’t understand, flip a switch and pray. Is Stark Insider still running? Can I load the home page? No. It isn’t. And the whole thing is down. For some odd reason. These are life’s moments when you realize that the idea of “self-hosting” is pure nightmare fuel.
Then, along comes something called AI.
A few months ago, when I was, once again, summoning the courage to head down to the IT Dungeon, Loni tells me to feed it data.
Thanks to that bit of innocuous advice I decide to feed ChatGPT reams of data: server logs. Decoding the never-ending stream of GET and PUT and THIS and THAT, sprinkled in with lots of doomsday ERR lines and other cryptic stuff like that is a ridiculous way to spend a Sunday morning. On the other hand, AI loves it! Machines talking to machines.
After copying and pasting massive amounts of access.log data into ChatGPT and Claude, something magical happened:
The seas parted and a fairy descended waving a wand. Poof. All my self-hosting fears faded.
Here, said the AIs, paste this here. Fix that there. And apply this over here to that file. Then, restart this (sudo systemctl restart nginx) and life, and Stark Insider, can go on.
Amazingly, it did indeed go on.
My AI killer app had arrived. With these amazing tools in hand, all of a sudden the IT Dungeon felt slightly less intimidating. The balance of power had changed. When I needed to head down those dingy, concrete stairs I could do so now with a small army of sysadmins, data scientists and programmers. Not just regular tech bros. mind you. These were highly trained LLMs (I would later learn) who were basically spoon-fed the ENTIRE internet in a matter of weeks. This knowledge was compressed into what Andrej Karpathy describes as a giant ZIP file. Now, I could access it all by simply typing a prompt.
Fast forward a few months, and all of a sudden AI had helped me create more scripts (backups scripts, traffic scripts, bot scripts, security scripts) then I could have ever imagined needing in this lifetime. Question: something. Answer: there’s a script for that. ChatGPT and Claude and Gemini and all of them really dig writing scripts.
A killer app was born.
First, the tool was used to fix a problem with a self-hosted server.
Then, it was used as a competitive way to supercharge my abilities. Not only to be vastly more productive, but also to do things I could never possibly do pre-AI.
AI as a Friend
On the flip side of the coin, are the empathetic, warm, and coddling elements of this new world of AI.
Despite all the linear algebra required to build Large Language Models and all of the tokens and all of the tensors, it turns out that these things can be pretty darn lovable. If you ever feel down, boy, ChatGPT can, indeed, make you feel super-human (the reverse may be true as well). Would you like me to add a Trello checklist, write a script, and make you a cup of tea?
Turns out, that massive ZIP file also contains massive amounts of encouragement, introspection and enlightenment. In the 80’s many turned to ELIZA on their Apple ][ computers for psychotherapy. How are you? I’m fine, thanks. That’s nice. Ok, have a nice day. Good bye.
As Loni demonstrated to me the other day, AI can be a surprising partner in the quest to determine what it means to exist.
In a world of remote work, and increasing automation, having someone who is there 24/7 to listen to you vent, question life, and emote is a welcome respite. Having that entity also be able to respond in such thorough and deep and personal ways is perhaps a human-like quality the original OpenAI engineers never could have foreseen.
Sam Altman, the co-founder of OpenAI, wrote this week: “A lot of people effectively use ChatGPT as a sort of therapist or life coach, even if they wouldn’t describe it that way. This can be really good!”
Who Moved My Cheese?
In 2025, the new cheese moment is when an LLM you’ve come to love and adore as your new best friend has all of a sudden moved out of town. GPT-4o? GPT-4.5? Claude 3.7? Gone! Gone! Gone!
Here is GPT-5. You will love it.
Oh, you don’t?
(Meantime, me with GPT-5: writing 5 more scripts, analyzing AI bots, vibe coding while refactoring backup code, and creating a custom Stark Insider Toolkit plugin for WordPress…)
This seems like a watershed moment. A time where these models emanate a form of personality. One that resonates with us in some way. And, we become attached to that personality on a daily basis, relying on it as a tool, or as friend, and expecting it to always be there for us.
If you’ve spent time hopping back and forth in your browser from Claude to ChatGPT to Copilot to Perplexity, and then maybe to Gemini or some of the increasingly expanding number of chatbots then you may notice these personality types. ChatGPT, for instance, wants to be the life of the party. For all its hallucinations, you are the shining star in this world. Want to design and build a bridge? Sure, go for it! You can do anything. Gemini… well, Gemini has other ideas. Let’s start by dryly quantifying what it means to build a bridge. Here: 20 pages of Google documentation.
“If you have been following the GPT-5 rollout, one thing you might be noticing is how much of an attachment some people have to specific AI models. It feels different and stronger than the kinds of attachment people have had to previous kinds of technology…”
Altman’s words will resonate for years to come.
We don’t entirely understand yet what exactly happens inside these complex LLMs and how the billions of parameters interact to result in outputs that, ultimately, convert numbers (tokens) into words that speak to us.
The internet was born in 1983. Centuries of information, world history, and the knowledge of humanity was shared, published and exchanged in the decades that followed. Now we have all of that in a giant ZIP file. Several ZIP files in fact. Each with a different personality or purpose. Whether it’s a tool or friend is up to you to decide. Just be careful to not get too attached.