GPT-5 just launched, but the bigger picture is how AI is changing our lives. Cleo Abram’s interview with Sam Altman digs into his perspective on what this means for us all.
Cleo’s Huge If True podcast is an optimistic show about using science and technology to make the future better.
She’s one of my favourite interviewers and journalists. She comes to every conversation with real curiosity and warmth, great preparation, and extreme intelligence.
No gotcha questions or sycophancy that seem to be the two primary modes for most “journalism” at the moment. She just seeks to deeply understand what’s going on and how people are thinking about it, the decisions they’ve made, and the rationale for those decisions.
Sam Altman fully engages in this conversation. Thoughtful and open with a lot of humility. He also lights up talking about some of the possibilities and what’s already been achieved. He and the OpenAI team are doing science. They’re experimenting, learning, and refining.
You can decide for yourself if he’s genuine and if you believe him… but I mostly do. I’ve been watching his interviews and talks since the early 2010s when he was at Y Combinator. I think he’s stayed amazingly consistent and grounded over the years – through to being a household name now.
If you’ve only seen the media portrayal, it’s worth watching this. An antidote for the rage-bait and fear mongering. This is a conversation between two really smart people.
The Huge If True podcast is explicitly framed through an optimistic lens. I don’t think Cleo does this in a way that’s naïve and ignores risk though. She focuses on the desired futures and what’s possible, what would need to be true to realise it, and the downside risk that needs to be managed.
Ultimately, what the future looks like is really up to us – individually and collectively.
For me, if we don’t view the future optimistically and work to make it a reality, we’ve got no chance of an optimistic future.
Some Highlights from the Interview
AI as a Tool for Enhanced Creativity
There’s plenty of talk about AI replacing human thought, but Sam Altman sees it as a way to unlock creativity. He talks about the speed you can now bring ideas to life with GPT-5 allowing for rapid iteration and experimentation – and that enables a new kind of creative process.
A key focus for this iteration is increased accuracy with fewer hallucinations – and to tone down the sycophancy it’s become notorious for over recent months. I think these are critical steps in improving the creative process. I hope it works.
Collaborative Intelligence
The future is a human-AI partnership, where machines amplify human creativity, insight, and problem-solving skills rather than replacing them. This blend of strengths can unlock breakthroughs neither could achieve alone.
GPT-5 can answer complex technical and scientific questions, create advanced software in seconds, and produce more natural writing, but we will quickly recalibrate expectations and demand more.
Profound Societal Implications
Alongside the excitement, there’s also a sense of loss. Sam hints that traditional markers of intelligence are changing. What it means to be smart, capable, or creative has changed. “A kid born today will never be smarter than AI…” but will be far more capable by using AI.
AI’s growing ability to adapt culturally, with personalisation and memory features that reflect individual and cultural contexts, makes it feel more meaningful to users.
As we reimagine the role of work in our lives, AI carries profound consequences for our sense of purpose, belonging, and wellbeing… as well as for our economies and society.
All of this is both an opportunity and a responsibility that demands we engage deeply with the changes ahead – not just reactively.
Where I’m At
A quick reflection on where I’m at with the AI revolution.
AI vs GenAI
I’ve accepted the semantic reality that when most people say AI now, they’re specifically referring to genAI version of it. Traditional predictive rules based AI and machine learning is still very powerful and relevant but has been folded in under the broader AI umbrella – and I fear forgotten by many.
As part of this we’re losing (or possibly lost) the explainability expectation we had – and mostly got – from traditional AI. We’ve pretty much accepted genAI as a black box tech with a neural network that’s too complex for explainability.
Personally I think it’s important to differentiate between these two technologies but for ease of communication I’ve decided to fall in line with the norm and just refer to it all as AI now.
My Experiments
While I follow AI progress closely and experiment with aspects of it, I don’t have time to invest in playing with the more technical building blocks because things are evolving so fast.
My focus is applying AI’s current capabilities (that grow daily) and embedding them in my workflow – without getting into complex workflows and integrations. The models are leapfrogging each other so rapidly, I’m not willing to go all in on one ecosystem to build that level of sophistication yet.
Deep research with synthesis is my primary use, so what I get most value from at the moment is context engineering. This is where I invest my time to increase the quality of output I get from models. I’ve invested a fair bit of time to make this portable so I can get a better baseline output from any of the models I’m using.
Perplexity Pro is my only paid model at the moment because it’s the most optimised for my primary use case. But I frequently push it too hard and break it. Most days I use Recall, ChatGPT, Claude, Gemini, and Copilot too.
I really hope options are introduced soon for the way context and memory are handled by different models. I won’t dive into the technical comparisons or definitions now, but what I’d love is the choice of having context and/or memory restricted to a thread, a space/project, or to be global.
UX/UI Frustrations
The capabilities of the models are evolving extremely fast but I don’t think the UX/UI is keeping pace. The chat interaction style was fine for simple question/answer and key to ChatGPT’s breakthrough and mainstream adoption in November 2022. We’ve blown past that simple interaction since.
Most of the models still use that same interaction pattern. Even when doing more complex things like deep research and image generation. This means the only information hierarchy is recency – what’s important or refined output easily gets lost in long threads.
ChatGPT’s and Gemini’s canvas, Claude’s artifact pane, and Perplexity’s page and lab features are all attempts at this but work differently and lean toward (impressive) automated output rather than prioritised content to iterate and refine. Gemini is closest to what I’m hoping for on this.
I switched to using Comet (by Perplexity) as my primary browser a few weeks ago. It’s built on Chromium but as an AI-first browser with deep integration with the Perplexity Assistant. A lot of it is extremely impressive but suboptimal UX/UI in places.
Switching Task Modes – Focused vs Discovery
I’d previously experimented with using Perplexity as my default search engine but there’s a mental model transition and different task focus.
When I’m in focused mode and just want answers, Perplexity is far superior… but when I want options surfaced in an exploratory/discovery mode, Google is still more visual with Places, maps, Gemini snippets, images, videos etc.
When using Perplexity and Comet, the biggest challenge is keeping threads organised. Again, I’m in two key modes… disposable searches that I’d really rather were tucked away like traditional browsing history.
Focused mode is often stuff I want to reference later and/or build on. I use the Spaces feature for a lot of that but the information architecture gives all threads in history basically the same level of importance.
When doing long (sometimes multiday) focused mode, the thread length gets way out of control and key things are hard to find. This is where I really want something like Gemini’s version of the canvas to pull out things to work on between raw material and finished output.
I hope the models give the UX/UI more love over the coming months. The other things I’m looking forward to are better native integration with core tools and simple agents that can reliable do scheduled tasks.
I think it’s worth celebrating how far things have come in under three years though… and I can’t wait to see what happens next.
