AI vs. Marshy

AI versus Marshy #43: first-look at Instagram AI, simulated photon chips, and boosting old clients

Published about 1 month ago • 4 min read

Greetings and welcome to the next edition of AI versus Marshy.

The newsletter that demystifies AI and helps you learn a thing or two.

A warm welcome to the new faces we’ve got this week - thanks for jumping on board 🏴‍☠️

This week looks at:

  • A first-look Meta releasing AI into Instagram
  • What cutting edge looks like - designing new data chips with Google X
  • How to use AI to be useful to old clients

A meaty edition today, so let’s make like a butcher and chop on with it 🔪


What AI inside Instagram looks like

Via my Instagram account and news.

I never quite got into Instagram.

I was a heavy Facebook user. Writing posts and getting engagement was something I enjoyed and having an instantaneous response platform was good while it lasted.

FB newsfeed isn’t what it used to be and we use ours mainly for FB spam.

My Instagram used to be foodie snaps, snippets of drawings and handwriting, and is currently swamped with my slow-motion basketball 3-pointers.

So imagine my surprise when this appeared on my Insta.

I vaguely remembered reading something about an AI test in Instagram but didn’t really pay attention to it.

I even ignored it for a day or so because how does this help me drain more slow-motion 3-pointers? 🏀💨

Playing with it, it talks a bit of a game - claiming to be able to integrate with other platforms with its API key.

(Pro-tip: This is usually the first thing a no-coder looks for when they see a new tool).

I haven’t been able to do much with it yet, but shared it with the Ben’s Bites community.

One nugget I learned is someone has been able to bypass WiFi in planes (where Whatsapp is free in some countries?) to get free news!

I’m enjoying the early vibes of Ben’s Bites 2.0 and if you’d like to get involved and join the community you can get some early bird pricing via this special link (heads-up I get a finder’s fee).

It’s a good investment in my opinion.

The trend is only going to keep growing and getting in early will give you access to a lot of sweet exclusives as well as only paying once for something that will go up in value.

Using AI to create computer parts that will power even better AI

Via X company.

Brace yourself - I’m going to try and translate some really complicated concepts for you.

Data travels faster down one pipe.

But at a certain point (say closer to being used) the data needs to be separated into different information components.

This is done with a demux and is housed in a data centre (think where the cloud is).

Google X (aka the moonshot factory) has built a new kind of demux with AI.

The way they built it is with a different technology than usual called integrated silicon photonics.

Here’s a picture - see how it’s much smaller than a standard design?

µm = micrometres.

A length of bacterium is 1-10µm

A fog droplet is ~10µm

Super chip - 14x16µm

So how is this possible - simulation.

I’ve talked about digital twinning before - where you can copy data from the real world onto your computer, and predict what’s going to happen in the simulation with terrifying levels of accuracy.

Well Google built an AI design simulator, then:

  • created an array of chips with the simulator
  • Put the chips through simulation and testing
  • identified a design that passed the simulation
  • was able to then use that design to create a real-world chip at a semi-conductor foundry

Put another way:

They created something with a computer that’s impossible to design as a human, and then printed it out to be used by humans in the real world.

I was going to include this news in last week’s newsletter but was glad I didn’t.

Because I got a WhatsApp from my old colleague Patricia!

She ran point on this project and it doesn’t surprise me in the slightest.

Well done!

Like with all deep tech it will be a while before this is realised by use mere mortals - but don’t try and tell me AI isn’t doing groundbreaking things.

Using AI for old client relationships.

Via Sydney Startups FB Group.

As part of my mission to identify more ways to help my dear readers, I put a shoutout in a local FB group basically seeking to share ways you can use AI for yourself - today.

I didn’t get a huge response, but one question was about learning strategies to communicate to clients about AI.

So I reworked the ask to:

How can I be of more service to my clients with AI?

Here’s a way to get started:

  1. Get a list of your current clients (from invoices or your CRM)
  2. Use an automation tool like Pipedream or Buildship to gather generally client company information - website, industry, who their customers are, PR and news appearances, etc.
  3. Connect this to a database or Google Sheet, and use it to prompt some ways you can service your clients better
  4. For example, “I’m an accounting business that works with all of these customers, using this spreadsheet - can create 5 repeatable ways I can service these clients better” or “As a content marketing agency - are there any obvious things I should be talking to my clients about in 2024?”
  5. You can ask follow-up questions and grade the answers you get with a some specific feedback. e.g. 42, make it less about accounting and more about them etc.

The end result will be some actionable strategies you can adapt and run with.

Let me know what you think by replying to this email (or if you’ve got some further challenges!)

I’ll try testing this with some of my previous clients, and also have an older client I’m trying to help more specifically soon.

That’s it for this week - thanks for reading if you made it all of the way.

I’ve read that 300-400 word emails are best for email newsletters but honestly I want to give you more thought-out writing and this works for me.

Remember - learning how to work with this stuff is always going to better than relying on machines alone.

We’ve got this 💪


AI vs. Marshy

by Luke "Marshy" Marshall

Growth marketer meets biggest technological advancement in our lives. Learn about AI in a way that doesn't overwhelm. Add a splash of strap yourself in and be prepared.

Read more from AI vs. Marshy

Hello and welcome to another edition of AI versus Marshy. This is the newsletter that strives to make sense of AI hype and gives you ways to apply lessons today. This week goes a bit differently because we’re going deep on a topic - emotion detection. I started investigating and it demanded its own edition, so let’s move on - like soothing a tantrum - and get on with it. Thanks 🙏 -Marshy Emotions triggered? Via LinkedIn. This story starts with an assertion from my friend Dave. “Emotional...

3 days ago • 4 min read

Hello and ahoy to another edition of AI versus Marshy! It’s been a challenging few weeks - so I’m pleased to get this newsletter to you at a reasonable time. This newsletter charts AI’s trajectory curve and our 1 year anniversary is next year - if you like the newsletter can you forward it someone you think might be interested and ask them to visit: So they can subscribe themselves. Thanks! 🙏 This week covers: The best is yet to come - a novel look at...

10 days ago • 7 min read

Hello and welcome to another edition of AI versus Marshy! This is the newsletter that demystifies AI and gives you things you can think about and action TODAY. Straight-up, I have TWO apologies I want to dish out. Last week I tried uploading images into a new backend to make the uploading images process easier in ConvertKit. This looked fine in preview, but I skipped test email, and some of you got an email with no images. 2/3 of you got a second email with the images fixed. My bad! Running...

15 days ago • 5 min read
Share this post