I have recently started running an experiment: I'm building the habit of talking to an AI.

About two years ago, when ChatGPT broke into the mainstream, I had the intuition that it would be a big thing. I started right away using it for 30-60 minutes a day, forcing it into all kinds of tasks, just to build the habit of going AI-first.

Now, I am doing the same with voice mode, and I've found some interesting things.

Early experiments

My initial thought was, "can I use it to practice languages?". Since voice mode was activated, I thought I could have conversations with an AI to help me improve my English accent.

If you know me, you know I have a pretty solid level of English, but I still get weird looks when I mispronounce certain words, especially words I've read a lot but I don't get to hear very often - like "acquiesce", "myriad" or "suffice" . Some words I've never heard at all, so I don't know how to pronounce them the first time around.

I activated voice mode and started talking to it, but back then, voice mode was sort of fake: it simply transcribed your voice into text and it processed the transcript. Thus, my mission to improve my accent failed miserably.

However, it worked for learning Greek. Duolingo isn't very helpful for speaking practice, especially in less common languages like Greek. Turns out, ChatGPT is very useful for this: activate voice mode, tell it to guess your language level and then start talking to it. Using the transcript was fine.

It's improving

A few months ago, ChatGPT rolled out the GPT-4o model, which processes audio inputs and outputs directly, preserving crucial phonetic features such as intonation, emotion, emphasis, and pacing.

For instance, today I was practising a few words I struggle to pronounce like "throughput", and it noticed I was breathing irregularly and speaking faster than usual. Turns out, it caught me doing squats. Lesson learnt: don't exercise while you practise speaking languages!

While this might sound anecdotal, it isn't: the AI actually told me something like "you're speaking faster now and that has an impact in your pronunciation. Usually, when you speak more slowly, you make fewer mistakes". Interesting.

The possibilities are endless

You can do a lot of things with voice, not just learning languages.

For language learning, you can tweak it however you want: faster, slower, change topics, help me prepare for a job interview, let's talk about business, help me remember complex words in German, etc. The sky is the limit.

These days, I am doing this with voice mode:

• Practising basic Greek to level up.
• Occasionally speaking in Italian and German on specific topics to refresh my vocabulary and fluency - it comes back fast!
• Bouncing off ideas in English about things I might want to write about later on. More on this... now!

But you can use voice for other things: setting reminders, creating tasks, or doing regular braindumps onto the platform so it can turn it from unstructured gibberish into clear thoughts.

The braindump part is very interesting. Generative AI is particularly useful for turning unstructured data into something coherent. I use it a lot to describe something I want to do, and ask it to give me a more accurate and thought-through prompt.

I also use it to debate whether an idea I have in my brain is good or not. In this blog post, I shared that one of my most-used prompts is "tell me why this is a bad idea", so now I basically tell ChatGPT what I want to do and we can discuss it, so I can refine my initial thoughts and iterate on them. However, my ideas still suck in the end 😅

My next thing will be to ask it to help me prepare for difficult conversations: the project will take longer to complete, we have to change the team composition, our initial assumption was wrong, etc. This can be a particularly useful thing to invest time into.

Another experiment I want to try is developing agents using voice inputs to help me become more efficient, but ChatGPT told me that it is a bad idea:

Not just a bad idea. It is The worst.

Recommended content

• 🎙️ As per usual, your weekly dose of My First Million talking about upcoming market trends.
• 🎙️ ReworkFM on how small teams can make more progress than big teams.

Update on my projects

• 👾 We hired two more people at MarsBased and about to hire our second Project Manager. We have hired 5 people this year already.
• 🤝 Happy to have agreed to work again with one of our favourite clients, Localistico, after a year hiatus.
• 🎙️ Foc a Terra: We discussed how we support the elites and how DOGE isn't turning out to be the revolution they promised (in Catalan).
• 📖 The Rework podcast inspired me to write also about how small teams beat goliaths.
• 🎙️ Our marketing guy is turning me into a YouTuber. Check our latest Life on Mars episode where I discuss three common mistakes building SaaS companies (in Spanish).
• 📖 My folks at EALYX are kicking ass and getting lots of press.
• 💸 Another one of my investments, CeleBreak, are raising funds. Get in touch if you're interested.

Asks & gratitude

• Grateful for my team at MarsBased, who I got to see last week in our quarterly get-together.
  • Special kudos to David Garmendia for having spent the last eight years with us. Wishing you the best for your next adventure!
• Grateful for the great conversations about AI and more I have regularly with Leo Ruffini over morning coffee.
• Ask: Still looking for someone to jam with (either bass or guitar).