Hello there!
GPT-3 was released in May 2020, more than 6 years ago. Somehow, it feels like it’s been 2 years only!
Remember, back then, the goal was Artificial General Intelligence. Everybody was laughing, but now, it happened, we definitely reached AGI.
There is a catch though. It’s not good at everything. It’s just good in general, at stuff that most people will work on in their day-to-day life. But today I’m not going to do some lame “AI is bad” post. I am going to make an “AI fucking rocks” post.
Living with ADHD: the DAMN problem
So I have insane ADHD, textbook case, started in high school, and everything has been downhill ever since.
Everything became x10 harder:
- homework (dropped out)
- work (left before the company got acquired for infinite dollars)
- my own company (thank god for my business partner who had my back and still does)
- taxes, administrative stuff
- medical stuff
- life chores: laundry, cleaning, etc.
So. It’s not fun. With ADHD I just never got around to doing the stuff that matters but feels really boring.
(I know this will sound like I’m just lazy and making my personality all about ADHD. I have no argument against that but also no comment section so, …deal with it.)
My point is that I was basically handicapped living in this society. I just couldn’t do all that stuff for the life of me. When it comes to what I’m good at (hyperfocusing on computer stuff), yes OK. For the rest my policy was “oh, I’ll do it tomorrow.”
First remedy: Beeminder
So, the first remedy I tried for like 5 years was Beeminder.
This is an app that makes you pay e.g. 5 dollars if you don’t follow through on some goal.
For example, I had a goal which was to have an empty mailbox, running every 3 days.
This worked very well. But ultimately, I was suffering. When the email says “hey file your taxes right now lol” and you have 1 hour to do it… well you struggle, then you miss the deadline, then you pay…
Then with time it became less powerful, I cheated a little bit more… until it became 0% effective.
After Beeminder: in limbo
Then after that, there was a period of like one year where I stopped using Beeminder.
I stopped answering my texts, emails… I just had to let go of Beeminder, which I did.
But… that meant that my administrative stuff didn’t get done!
… Second remedy: I’d rather not talk about it
If you happen to catch me IRL I can tell you but I don’t feel comfortable putting it on my blog.
I will give a few hints:
- this method has existed for as long as humans have been on the planet
- it’s natural, doesn’t involve technology
- it’s definitely going to raise your cortisol levels above healthy levels
My final technique: a MacBook with Opus 5.5
One day, I had this idea. What if I used Claude Code to just help me with administrative stuff? I could hook it up to a browser and my emails and tell it “please take care of all this bullshit for me”. I know the OpenClaw dude did it before me, but it just didn’t work for me to do it in a chat box, while having oh-my-pi (a harness similar to Claude Code) worked really well for me.
I have been using it on and off, with different harnesses and models… but I wasn’t 100% convinced.
But, once Fable 5 dropped, it started working really well. And now today I am using Opus 5.5 and it works insanely well.
One thing that makes this technique work is that my MacBook is always lying around. When I feel like I don’t want to work, I can easily crack it open, and ask the agent “hey what was I supposed to do like in terms of administrative?” and it will always find something from my insanely long todo list of stuff to do.
Especially at night, when I can’t use my big mechanical keyboard, I tend to open this laptop to at least get something done without waking up everyone in a 5km radius.
Anyway. Here are a few things that Opus 5.5 did for me in the past few weeks.
Signing up for sports (Opus 5.5)
I had to deal with a form for signing up for sports.
I filled it out twice: once for my family members, and then for myself a few months later because I wanted to join too.
The first time, I did it manually. SURE, I could do it, it worked. It was a pain to understand and fill, it was super confusing, I had to check the website like 20 times… But it worked. Well. They got my phone number right at least, because they called me to come to their office to fix it :).
The second time I filled it, for myself, I just asked Claude to do it. Make no mistakes.
You know what. It did no mistakes. It had all the context, addresses, phone numbers, the website… Went through on the first try, no headaches.
Dealing with tax bs (Fable 5)
The tax office couldn’t explain to me in plain French how I was supposed to get back the VAT they made me overpay. I was supposed to fill some bullshit form and send it to some bullshit office on a bullshit date 6 months later.
2025 me could neverrr because of my goldfish memory. So at the time, I was like, okay L, I am totally never going to recover that $2,000. I put a reminder, saw the reminder 6 months later and uh did nothing about it like I predicted.
BUT! A few months ago, I connected Fable 5 to my taxes, told it:
“hey dude I need help, read all the messages, and then please can you list all the stuff I need to fix from a tax standpoint on this account?”
And it did just that. It managed to get the big phat VAT refund (soooo it paid for itself).
It helped me find a replacement international tax lawyer (the previous one was 100% scamming me, taking advantage of the fact I was inexperienced).
Then it did my taxes. Filled some bs forms for me, asked the tax office to adjust my tax residency, etc. It did all the stuff that accountants were either ghosting me on or quoting obscene amounts for.
I verified everything, it’s all good. Well at least it’s better than what I would’ve done and I suspect better than any accountant as well since at least it was answering my emails.
Btw: I made it close my French company because I don’t want to deal with this anymore. Another W for Fable.
Dealing with medical bs (DeepSeek V4 Flash 1017, Kimi K3/K3.1 Pro, GLM5.3, Opus 5.5)
So, when it comes to administrative medical stuff, Opus 5.5 is amazing.
I tried the chinese models on a privacy-oriented provider… but once I tried Opus 5.5, I couldn’t go back.
Of course, I am not using it to diagnose myself. It’s GREAT at keeping everything in a folder, noting down information, transcribing my voice notes using Whisper, hunting for lost documents in emails…
I can just ask AI to dig through my old notes and get the bigger picture, and collaborate with doctors in an easy way.
Then I can ask it to summarize things, help me prepare for appointments. It’s been a literal lifesaver.
What about privacy?
I know this is controversial, but in my personal case, I decided that not dying of a preventable mystery illness was a reasonable reason for me to hand over all my data to rationalist motherfuckers.
If the NSA wants to know what’s up with me, they can already just look into my iCloud inbox… :)
I initially tried to use “privacy” oriented provider Synthetic. But it turns out that the data goes over Cloudflare and AWS, which is the exact same threat model as if I was sending the data to Anthropic (= NSA + a bunch of corporate drones will have access to the data).
So I decided that if I was giving my data out like free candy, at least I should have the best model taking care of things for me…
2026: you don’t have to suffer anymore.
A USD 300 lightweight secondhand laptop, a USD 20/month subscription, and you have a secretary to handle all the boring stuff for you in life. In fact I would argue that you just need the USD 20/month subscription and your phone. I bet it can just save everything to its cloud. Insane times we are living in.
This is going to be the real killer use case for the check notes 5% of humans who can’t deal with administrative stuff.
I know that I sound like an Anthropic shill. I hate them and wish for their IPO to crash as deep as possible. But I have to admit that there is something special about Opus 5.5 that makes it wayy better than other models. Other models just don’t cut it.
no ai was used to make this post