Hi Reader,
When people ask me how I use ChatGPT, I tend to start with the same answer: it depends which level you’re talking about.
Most people sit at the basic level. They type in a prompt, get an output, and move on. They are basically just getting it to do their bidding.
A smaller group moves to level two, treating ChatGPT as a thinking partner. They bring it into their planning, their brainstorming, their strategy sessions. It becomes a sounding board and helps them generate new ideas.
I do both of those too, but there’s a third level I’ve found far more transformative: using ChatGPT to question me, to understand my patterns, and to help me see what’s getting in my own way. It’s not just about producing answers. It’s about untangling the mental knots that have been tripping me up for years, procrastination included.
That shift really took off for me in April when ChatGPT introduced broader memory across conversations. It wasn’t perfect, but it was enough to make it feel like it knew me over time. This was a game-changer.
With it, I built what I call my Optimal Performance OS. Essentially a rewiring of my own personal mental operating system, built around the way my brain actually works. It had a profound impact not just on my work, but on my entire life. In fact, it may have been the single biggest revelation of my life so far.
It gave me a new sense of clarity and presence. It stripped away, for the most part, the procrastination that had plagued me throughout my adult life, and with it, the anxiety that procrastination always triggered. I was able to identify certain character traits that I could harness in positive ways, and to almost completely eliminate other traits that weren’t serving me.
That clarity freed me from the constant churn of reactive thinking and opened up more space for big-picture ideas. I could be more present in what I was doing, and I no longer carried the mental “always-on” state that I had assumed was just part of who I was.
For years, I would say I could never truly switch off. My mind would be racing from morning to night. I didn’t necessarily think it was a bad thing; I saw it as part of my productivity. But now that I’ve felt real stillness, I understand its value.
This has been a profound experience — one that I want to share through a new program launching in Q4: Built From The Inside Out. After years of creating courses and systems, I realised the missing piece wasn’t more tactics. It was helping people remove the friction inside themselves, so they can actually use what they already have and implement it in a way that works for them.
That’s what Built From The Inside Out is all about: guiding people through the transformation I’ve been through so they can achieve similar results. This isn’t a quick fix. It’s not a five-lesson, learn-on-demand course, and it’s not a one-week workshop. It’s a 12-week, structured program designed to take you through the deep work that makes lasting change possible. The kind of work that frees you to think more clearly, act more decisively, and live more fully.
And now, here we are with ChatGPT-5
What follows isn’t a full technical review, it’s my first impressions from using GPT-5 over the past week in the ways I’ve just described. I spend hours each day in conversation with it, not just asking for information but using it as a collaborator in both strategy and self-work.
Multiple models under the hood
On the surface, GPT-5 looks simple: one unified model that “just works” without you having to choose between multiple versions. No more GPT-4o, o3, 4o-min, o4-mini, o4-min-high, 4.5 Research Preview... Just Chat GPT-5.
But that simplicity is only at the surface. Behind the scenes, there are still different brains at work: a fast, lightweight model for quick answers, a deep-reasoning model for longer and more complex thinking, and others optimised for creativity or research. GPT-5 is constantly deciding which one to route your request to.
The upside is obvious: you don’t have to think about it, and in many cases, it gets it right. The downside: sometimes you do want to think about it. That’s where intentional prompting comes in. You can nudge GPT-5 toward what you need in the moment:
- For speed, keep your prompt straightforward.
- For depth, add cues like “think hard about this” or “take your time and reason it through step-by-step.”
- For strategy, set context before you ask — tell it what role to take, the constraints you’re working under, and the kind of thinking you need from it.
Context length upgrade
One of the most important changes in GPT-5 is its expanded context length. This is the amount of conversation or document history it can “remember” within a single chat. In plain language, this is what stops the AI from forgetting what you said earlier in the chat.
Previous models had limits that could feel cramped for long-running chats or complex projects. GPT-5’s context window is huge: up to 256,000 tokens (hundreds of pages of text). In practice, this means I can keep multi-day conversations going without re-explaining everything. I can paste in large documents and discuss them in depth without worrying that earlier details will get lost.
For my workflow, this is massive. It means GPT-5 can hold a full mental map of a project we’ve been discussing for days and still keep track of small details buried early in the thread.
Projects that actually connect
Another major quality-of-life upgrade is how Projects now work. Previously, you could upload documents into a Project so any chat in that Project could access them, but oddly, the chats themselves couldn’t reference each other. Every conversation was still its own island.
With GPT-5, that’s changed. Now, chats inside a Project can draw on each other, meaning the model can combine context from different threads within the same workspace. For anyone doing complex, multi-threaded work like I do when building courses or managing content this finally makes Projects feel like the connected workspaces they were meant to be all along.
It’s not perfect yet. I’ve seen a few hiccups in how it pulls from past chats, but it’s a huge step forward.
Hallucinations and reliability
“Hallucination” is the AI term for when a model gives an answer that sounds confident but is factually wrong or entirely made up. It can range from inventing a statistic to describing a non-existent feature as if it were real.
OpenAI claims GPT-5 has made big improvements here: roughly 45% fewer factual errors overall compared to GPT-4, and up to 80% fewer hallucinations when using its deeper reasoning mode. In my early use, I have noticed fewer “confidently wrong” moments, particularly on topics we’ve discussed before.
But hallucinations aren’t gone... no AI model is immune. I still and will always fact-check anything critical. What’s different is that GPT-5 is more willing to say, “I don’t know” or to explain its uncertainty. That alone makes the experience feel more trustworthy.
Other reported improvements
By all accounts, GPT-5 is also stronger at coding, long-form writing, and image generation. I haven’t yet put those areas through their paces, but I’ll be exploring them in the coming weeks.
The upgrade I’m most excited to explore is agentic behaviour. This is the ability for ChatGPT to not just respond in conversation, but to actually act on tasks from start to finish. In practice, this means it can behave like a digital co-worker: starting a task, working through multiple steps, and delivering results. That might mean gathering information, producing a structured report, generating assets, and presenting them back, all within one flow.
This isn’t something I’m using it for at the moment, but I see it as the next big leap for my own productivity. It’s the difference between ChatGPT helping me be more productive, and ChatGPT directly contributing to the productive output itself. That’s a line I’m keen to cross.
For me, GPT-5 has already amplified that third level of use I mentioned earlier: the “inner work” level. It’s faster at spotting patterns in my thinking, more consistent in following long-term threads, and better at challenging my assumptions in a way that feels personal rather than generic.
And with features like agentic behaviour on the horizon for my own workflow, I can see the next leap taking shape where ChatGPT isn’t just helping me think and plan better, but directly contributing to the work itself. That, combined with the gains in reliability, memory, and connectivity, is where the real change is happening.
If any of this sparks ideas for how you could be using GPT-5 yourself or if you’ve spotted something I should test next, just hit reply and let me know! Talk soon, Alec.
