AI versus Marshy #29: voice-activated ChatGPT, brain machines, and is it right to copy?


Hey again!

Welcome back to the newsletter that indulges your intrigue about AI and its practical applications.

At least that’s how ChatGPT introduced me.

Read on to discover what I’m talking about:

  • Voice conversations with ChatGPT
  • The case for cyborgs
  • Training models, copying, and copyright

Let’s turn it on like Donkey Kong.

-Marshy

Talking to AI with ChatGPT iOS

Available on App Store for premium customers.

Only a couple of years ago - I broke both of my arms.

A good portion of my job involves typing and I was off work for at least 4-5 weeks.

I had to do something after that and turned to voice dictation software to help me do my job.

I trialled Dragon, open source software, and MS Word dictation and all of them were not quite there.

In the meantime - my arms started healing enough and I was able to get on with it.

It was 2022 - surely there’s a better way.

This week I did find a better way.

It’s performing better than dictation apps.

Before I jump in - you can watch a video of it in action here:

video preview

Link

It does a remarkable job of capturing what I say and do, and picks up my thick Australian accent well - something dictation software struggled with.

It doesn’t handle pausing well - and it went in some weird directions while I was brainstorming some things.

For example, I was asking what’s some good ways to introduce using this technology for people reticent about talking to a machine (you know - all of us).

It suggested asking a question on topics you’re passionate about (good suggestion) but then misheard me and mentioned “the Ukraine situation”.

It was odd trying to find out why that happened, and then hearing it dodge as I tried to get to the bottom of that answer.

The voice will also drop in the occasional “um” and “hmm” - which is mind-boggling from a user-interface perspective - a designer has clearly decided that adding these human-isms makes the voice sound more natural. It works!

Another wild use case was getting it to do a role play as the irate customer and I’ll play the role of customer assistant.

Basically - I wanted to see if I’ve still got it when it comes to client services:

The app errored after this exchange which was good - because I could feel myself getting defensive despite knowing this was a role play 🤣

OpenAI are clearly a leader in this space and I want to know what they know (they’re clearly holding some of the capability back).

But the law might catch-up with them (see further below).

Brain augmentation machine?

A case for cyborgs via every.to.

I’ve covered Neuralink in previous editions, and since reading about them in detail 7 years ago, a lot has happened in the world.

Another entrant in this space is re:collect founded by Alice Albrecht. She argues that have collective access to a tool that improves with human feedback is a far better pathway to superintelligence than simply robots improving on robots.

Singularity hinges on machines improving upon themselves until they learn how to learn and exponential improvements carry it to becoming far smarter than us.

A cyborg approach combines man and machine into something that steers us in a better direction as one combined with the other gets to a superior outcome than either one alone.

This stuff tickles my brain, and there’s more detail in the linked article if you’re interested.

What do you mean AI can’t copy everything?

Covered in The Guardian.

I’ve enjoyed following the Legal world tackle AI as it’s extremely uncharted waters when neither side has a compelling case.

Copyright exists to protect a creator’s intellectual property.

AI is currently trained on vast multitudes of intellectual property on the web (and elsewhere).

Tools like ChatGPT wouldn’t exist if it stuck to the letter of the law and trained on nothing other than non-copyrighted material.

The law is currently playing what role copyright plays in training.

This is an existential problem to solve for AI’s progress.

You can’t just ignore the law.

The law exists to contest ideas in an adversarial system for the betterment of society.

It has many faults - but arguing for copyright isn’t one of them.

I’m not sure how this gets solved and it’s above my pay grade - I do like how Europe in particular stands up for this stuff.

They led the world into implementing GDPR protections, and Mistral is an incumbent from Europe seeking to position itself as an alternative model for Generative AI - their strategic memo is a gripping read on why they’re doing this.

I’ve been playing with ChatGPT (Dall-E) for image generation and combining it with Midjourney.

Here’s a photo of my work set up:

I ask ChatGPT to describe it as if it’s being drawn by someone through text description alone, and then give the prompt to Midjourney:

Imagine you’re drawing a workspace that is designed for productivity and personal comfort, radiating a sense of modern technology blended with personal touches.

Begin with a wide desk situated in front of a large window that offers a view of lush green foliage. The window is framed by two visible bamboo roll-up blinds, adding a natural element to the scene. On the desk, place a dual-monitor setup. The left monitor displays the desktop icons while the right monitor showcases a wallpaper of two smiling people, suggesting a personal connection to the user.
To the left of the monitors, set a ring light on a sleek, adjustable tripod, indicating readiness for video conferencing or content creation. There’s also a webcam perched on top of the left monitor.
In the center of the desk, sketch a closed laptop perched on a thin stand, indicating a preference for ergonomics. Rest atop the laptop a black notebook and a pair of black-rimmed glasses, hinting at work breaks or moments of reflection.
Directly in front of where the user would sit, draw a white wireless keyboard, its minimalistic design suggesting modernity and simplicity. Next to it, portray a black, ergonomically-shaped mouse resting on a large, green gaming mouse pad with an artistic, possibly serpentine design.
On the right side of the desk, include a personal touch with a small, framed photograph featuring three people in sports attire, all smiling, adding a warm, familial atmosphere to the workspace.
Finally, add a cat tree beside the desk, complete with platforms and resting areas, implying the companionship of a pet within the space. This pet accessory also softens the tech-heavy vibe, introducing a sense of home and comfort.
Overall, ensure that the drawing communicates a balance between a high-tech work environment and a space that is personal and lived-in, reflecting the character and interests of its occupant.

Here’s what it churns out:

Midjourney is too expensive a subscription to maintain for long but for now I’m experimenting and seeing what I can learn.

This week is go-go with existing projects and the new venture.

I’ve been fine-tuning my big fast value call process and my goal is to simply deliver free insights calls until there’s so much demand I will need to raise my entry point price for 1:1 coaching.

I’ve done about 10 so far and the feedback has been great.

Have a great week, email me any thoughts/questions you have, and remember: we’ve got this! 💪

-Marshy

AI versus Marshy

I call out big tech company bullsh*t, avoid hype, and show scaling companies how to grow with AI.

Read more from AI versus Marshy
ai chart

Ahoy my friends! Welcome to another edition of AI versus Marshy. For new readers - welcome! And if you have hung around for a long time - I appreciate you giving me your attention (or at least skimming) for a while now, thank you! I’m going to do something a bit different this week and look at three apps I tested earlier on in this newsletter’s history, and check-in on where they are now: Summit AI coach Tome presentation building Audiopen voice note recording I want to get jiving and give...

dad with the boys

Hi Reader, Have you noticed I play with the greeting each week? I like personalisation, but don’t have everyone’s first name in my list because you’ve arrived in different ways. Sometimes I just like calling you all fam - this isn’t a large list, and I just love that there’s a humble cohort of people that welcome and read this email most weeks. Fun quote from a friend other day: “I knew we had this catch-up booked - so I binge read your newsletter in advance” Thanks Ross 🤣 For those of you...

dad and boy

Hello Reader, Great to be here again. This week I’ve been webinarring, discovering new toddler ailments from daycare, and did my first beep test run in a long time. But let’s talk AI. This week I look at: Some cool ways n8n is used for ADHD A newsletter deep dive and building in public Building an app with “vibe coding” tools There’s a lot to rock’n’roll with so let’s make like a shake and get steady 🥤 -Marshy Some cool ways n8n is used for ADHD I follow a lot of the n8n (an automation...