blog

Linux on Macbook Air

2026-03-20

I installed Linux Mint on an old Macbook Air for a relative, here are some notes.

Target machine

MacBook Air 6,1 (~2014)
4GB RAM
128 GB disk
Intel I5 CPU
Intel graphics

Issues

The disk is not visible by default. I needed to enter the option intel_iommu=off in grub as described on super user. The installer did not have good error handling for this and quit with TypeError: "NoneType" object is not subscriptable.

The Broadcom wifi chip needed a proprietary driver. I was happy to find out that the driver can be installed from the install USB, so that was pretty easy. However the driver does not play nice with the current kernel - 6.17¹ - and I had to pin the kernel to 6.14².

The webcam was more of a mess; cheese couldn't find a device. Turns out "Macbook Air Facetime HD camera" needed a driver too. The driver can be found on github but it took quite a bit of finagling to get it to work.

Enhancements

For a long time Mac user, these helped make the new OS feel more familiar.

Albert, a launcher which looks similar to spotlight. I enabled: Applications, files and system plugins. Had to play around a little with the indexing allow and disallow lists.

Moving the taskbar to the top is an easy little win...

It's 2026, so I described everything I did to the user's Claude. There are no issues yet but that should pay dividends.

Happy Surprises

Microphone, bluetooth, closing the lid / suspend and brightness, volume and backlight keys all worked well out of the box. In fact, wifi is working better now than it did on Mac OS where it couldn't connect to hotspot.

Before the install, the laptop barely worked (opening a browser took ~30 seconds) and felt outdated, now it feels snappy and modern. Maybe I'll update in 6 months to see how it's holding up.

I discovered this myself but it's online too ↩
bash sudo apt-mark hold \ linux-generic \ linux-image-generic \ linux-headers-generic \ linux-image-6.14.0-37-generic \ linux-headers-6.14.0-37-generic ↩

Kharg island

2026-03-15

I have a short term prediction in the Iran conflict.

The US sent a group of marines towards Iran which has raised eyebrows. The US also struck over 90 Iranian targets on Kharg island.

I think the US is planning on taking the island. This fits nicely for a few reasons. First it seems that many of the attacks against the straits have come from there and securing the strait is of utmost importance More importantly, the combined force is still hoping for regime change and so it doesn't want to destroy Iran's economic future as evidenced by the American reaction to Israeli strikes on Iranian oil fuel depots. However Trump has threatened to escalate to attacks against Iranian energy if Iran does "anything to interfere with the Free and Safe Passage of Ships through the Strait of Hormuz". Taking and holding the island with marines enables him to do both - escalate, take Iran's oil, while saving the infrastructure for the future.

~30% by the end of the week

Economics of Continual Learning

2026-03-11

Dwarkesh thinks continual learning is a bottleneck to AGI. Regardless of the technical difficulties, I don't understand how it's ever supposed to be viable for a business.

As I understand it, the definition of continual learning is that the model updates its weights in order to preserve new information (as opposed to keeping memories in an external database - whether that's memory.md or RAG). The idea is that every user would have a model that would learn how to work for that user and would get better over time. That implies that every user has to use a model with personalized weights. This seems impossible to me!

First, there are the obvious storage costs. OpenAI has ~1 billion users and 50 million paying users¹. OpenAI would have to store a 1-10 TB model for each one of those users (if they all get access). That comes out to more than $10B a month. Less obviously, all inference providers rely on a trick called batching, which means that instead of running one query through all the layers of the model end-to-end, many queries are processed per layer before moving on to the next layer. This saves time on loading parameters into the GPU. This would be impossible if everyone's models have different weights and would make inference prohibitively expensive.

Yes, a lab can offer this model to a few high paying customers but is that enough of an incentive to spend R&D on? The labs don't even release models RL'd for specific tasks (outside of coding) which would have far more users.

https://openai.com/index/scaling-ai-for-everyone/ ↩

Tmux brackted paste issue

2026-03-02

While perfecting the perfect shade of blue for links, which mostly entailed copy/pasting the values from the default google color picker I came across a (sort of) bug in tmux and simple fix.

When pasting the color hex into nvim inside a tmux window, the hex trailed off with ^[[201~. Weirder was I couldn't do anything anymore, every character I entered was instead represented by the utf code. Even Ctrl-b was hijacked, I couldn't exit (n)vim (that brought back bad memories...), or close the pane. By pure luck, I discovered the copy/pasting something again into nvim gave me back full control.

I found this issue which explains the cause. tmux waits escape-time long after an ESC to decide if the following chars are part of a longer key sequence. Due to lag (or possibly Windows Terminal) and a short escape-time tmux no longer interprets [201~] as "End bracketed paste", it just takes the raw code and passes it through to the open program (nvim).

The solution (as mentioned in the issue comments) is to set a higher escape-time. There is a trade off - esc to normal mode being less snappy - so I'm still experimenting with the right time, but 50ms seems good so far.

I still don't understand why it was messing up my terminal that badly, or why pasting again fixed that...