Old dogs, new cuts
How age can shield you from AI's sharpest edges
The other day I was making a cup of tea in the kitchen. As if by magic, my wife appeared. “Has something happened to Claude?” she asked. “It seems to have, well… broken.”
An intriguing question. To me at least.
It turned out she had been fighting with Claude over holiday planning. And in the middle of updating her plan part of Claude’s internal system prompt appeared to have leaked out…
But while my wife was bemused, Claude seemed none the wiser and continued as if nothing had happened…
Maybe it’s just me, but I’ve noticed an increasing number of these, err, quirks over the past few months. For example, my Spanish, remarkably, is now sufficiently good to catch Claude out on a translation error:
Or DeepResearch asking me how soon I wanted the research to complete. That question might make sense if it were a human, but for a machine I’d like the answers as soon as possible. And even if I said I didn’t want the answers before next Tuesday, DeepResearch doesn’t have the ability to procrastinate.
And then there’s real-time voice. I love it. It’s transformed my morning walks. But it has many rough edges. Occasionally instead of talking, it’ll play random snippets of music. Then last week it got stranger. It tagged a whooshing sound onto the end of an answer. Which it then repeated. Continuously. Until, several minutes, later, I concluded it wasn’t going to stop unless I did something.
Real-time voice also offers transcripts. For example:
Note the transcript fails to transcribe my voice correctly. I’m pretty certain I was repeating “¿Puedo ayudarte?” and not asking “Why do I use ASCII?”
And the UIs are full of bugs. Requests to Claude frequently fail…
Or the UI gets confused. Here Claude displays the title of “Episode 9”, but the content is actually for “Episode 8”.
Or you may find the Claude desktop app slow to reset after you exceed the usage limit. Here you can see it’s after 3pm but the UI is still locked out. If I were cynical I might suspect there’s a reason why it is slow to reset…
And my wife’s experience with holiday planning? Sometimes Claude is great. It has found several hidden gems we’d otherwise have missed. But other times it is an exercise in frustration. It forgets we need to eat lunch, is surprised when we are unable to teleport instantly between venues, or hallucinates amazing sounding hotels.
Ouch!
The frontier remains jagged. It’s very easy to get hurt. AI tools are simultaneously brilliant and broken, advanced and awkward. But despite the cuts and scars, AI remains remarkable, enabling things which were unimaginable a year ago. Take a recent example.
For several months I’ve been trying to find a good Spanish podcast series. There are many options, but none of them completely cover the set of topics I’m interested in. In the past I’d have made do with a mix of not-quite right podcasts and some books. And accepting there would be some things I just might not understand.
But today is different. Today I can create my own, custom, podcast. So I did. I discussed my plan with real-time voice and got a rough list of episodes together. Then I gave that list to Claude which built them out into a comprehensive set of episodes (45 of them!). Then Claude wrote DeepResearch prompts for each episode.
Those prompts went to OpenAI & Google DeepResearch which did their research magic and produced ~30 page reports for each episode. A few hours later I had ~650,000 words of source material to turn into podcasts. That’s a lot of text. To put it into context, it’d take a typical human ~2 years to write that much and ~45 hours to read it. Wow.
The docs got fed into Google Notebook LM, which created the audio podcasts. A bit of artwork from ChatGPT, Audacity to convert the wave files to MP3s and ta-da. I had ~30 hours of custom Spanish podcasts. They’re not perfect - there is the occasional descent into gobbledygook. But, importantly, they cover the subjects I care about. And they are not filled with adverts, or sponsor segments, or random letters from listeners. There is much to like about this new world.
And it’s not just Spanish. I’ve now got podcasts about music theory, construction, the future of the telephony industry. Even the odd one about AI.
Thriving
But it made me wonder. What do you need to thrive in this new world?
Increasingly I think there are three things:
Firstly you need to enjoy - and prioritize - experimenting. There are few paved paths. Those which do exist keep changing. My podcast creation workflow will be irrelevant in a year’s time. And probably sooner. Continual experimenting is required. That requires time and energy.
Secondly it helps if you’ve managed people. Why? AIs need to be briefed. They need to be monitored and corrected. Those are skills you get from managing people.
Finally you need a sense of smell. To spot when something is wrong. To figure out how to get around the AI limitations. Sometimes you will spot this. You can’t turn over every stone, so you need to develop a gut feel for when something feels ‘off’ and just turn over the relevant ones.
Not everyone has all these skills. Learning new things requires energy. Effort. Time. As we get older it’s easy to get set in our ways. After all, following well-trodden paved paths is a great way to save time. But it makes it harder to adapt to change.
Nor has everyone had the chance to - or wanted to - manage people. And having been a people manager doesn’t automatically make you great at briefing and guiding others. Sure, it increases the chance, but it’s not a 1:1 mapping.
And a sense of smell? That comes from experience. From having worked in a field for an extended period of time.
You’ll note that the last two are strongly correlated with age. The older you are, the more likely you’ll have had people management experience. The better your sense of smell.
Sure, young people will be more likely to experiment (just look at the remarkable uptake of AI amongst school and university age groups). But proficiency with AI is not the preserve of the young. The older generations are well positioned to be able to use it effectively.
Bizarrely, without realising it, the skills I learnt from twenty years of people management turn out to be surprisingly useful in helping me make the most of AI. And, as we ride this wave, it is likely to be the people who have this mix of skills who thrive. You don’t need to be the most technically adept. Rather you need to be able to balance experimentation with judgement - and have a sense of smell.
Those are the skills that will prevent you getting cut by the jagged frontier.
