Hello and welcome to another edition of AI versus Marshy.
There’s more than usual new readers this week, including some new owners of my Start Marketing book.
This newsletter charts and give perspective to the incoming change driven by AI - from someone that’s worked in tech, marketing, and sales for nearly 20 years - that’s me!
Onto this week:
🛠️ ChatGPT got a big upgrade - text-to-image for Pro users
♾️ Tool of the week: Pi, conversational AI that now references recent events
🎤 Meta’s Head AI scientist for Meta makes a bold and silly declaration (again)
I’m writing this newsletter from a plane, I’m making an unexpected day trip to Sydney today to sit on a panel about mental health for SANE.
I’ll include more about that next week - but mental health IS the passion topic I care about most (other than AI of course ;-)
A big upgrade for ChatGPT - image creation on demand
Hot off the press! This week ChatGPT released its image generation capability for DALLE-E3 this week for Pro users.
Here’s what I really like about it: accessibility.
Like it or not, ChatGPT changed the game for LLMs, generative AI, and the world of machine learning not by being the most cutting edge, but by being the most usable. AI had been around for decades in escalating echelons of power. But it was the fact ordinary people could use AI and glimpse an understanding of it that changed the game.
It’s a blip in history now, but OpenAI was on death’s door. Founded by a group of people wanting to take action against “an existential threat” is how it started, and as an NFP no less.
It was only after ChatGPT was released that the validation came - after 6+ years of research and switching into a corporation.
Suddenly - the global population (not just tech boffins) stood up and took notice.
It achieved this not via capability - but usability.
We’ve all seen AI-generated images by now. Some have even “won” competitions.
But they were still kind of crappy to use. Some rely on Discord back and forth, expensive subscriptions, or don’t let you do much after generating.
I pay for the Pro subscription (I do write this newsletter after all, I probably could write it off as a tax expense?) and had a quick play.
It makes this stuff easy.
I bashed out a command from the airport gate, and it coached me on how to get the most out of the prompt when I asked it.
My first prompt:
A hyper-real photo of an athletic middle aged man in a hoodie with brown hair and a short-cut beard hunched over a Macbook Pro in an airport? He’s sitting in a row of seats next tot the gate. More seating and the gate should sit in the background and the decor is dark grey carpet, off-white walls, silver finishing to grey seats, and a mixture of business travellers and other people - it’s early morning.
I gave some feedback prompts and landed at this:
I was (very creatively!) telling it to craft a picture of what I was doing.
Here’s a photo of what I was trying to recreate (it’s hard to take pictures on a solo mission):
I expect this to become easier and easier, and it still boggles the mind that this was nearly science fiction a year ago.
Tool of the week: Easy as Pi
Pi is another chat-based AI tool that focuses on beautiful design and language.
You might scoff at that sentence, but this is a real thing.
Pi hasn’t netted the acclaim that ChatGPT enjoys, but has enjoyed vast amounts of funding. So they’re either onto something, or very good at selling the idea that they’re onto something. The latter happens a lot.
This new feature is pretty cool though.
Pi is now up-to-date.
Meaning you can talk about regular and current events with it.
Other tools are famous for the delays and pushing back on getting current.
So it’s pleasing to see things continue to grow into new areas.
Shooting from the hip - scientist makes bold assertion that you can’t possibly know to be true
I pick on Meta a bit. I used to work for Facebook, have made a lot of friends from my time there some of whom still work there too.
I benefited from being exposed to a hyper-growth mega company, and am privileged to have been able to do that dance twice (having worked in Google before).
But still… I don’t understand the AR obsession thing, and I don’t understand why they give this guy a microphone sometimes.
This is a quote from Yann LeCun saying:
"”In terms of underlying techniques, ChatGPT is not particularly innovative”
They’ve now gone ahead and stated that AI will never threaten humanity.
I think it’s a bit too early to be making those kinds of calls don’t you think?
Here’s some ways it could pose a threat (just off the top of my head).
- Sophisticated hacking attacks
- Mass-machine learning for flying a coordinated army of weaponised drones
- Deep fake spoofing that triggered irrational behaviour in a VIP Head of State
I don’t like thinking about malevolent use - but it’s a terrifying reality that we hope the side of progress can consider, govern, and protect us from.
But to state that it’s never going to be a threat - yikes - step away from the microphone please!
That’s it for this week. Enjoy!
p.s. Thanks to those of you who replied with feedback last week - muchos appreciatos 🙏