Time for a February update - moving to Linux, EVE Online tools and AI!

Onwards to Linux!

In my previous post I had said I wasn’t planning to move to Linux as my main OS, but then Greenland happened, so I guess here we are!

The first step of the move was to get more storage, which lead me to discover that RAM wasn’t the only hardware directly affected by AI - local stores were selling SSDs at over 3x of what they cost back in October (the screenshot below is on Amazon, but local prices were pretty much the same):

Amazon price history for a SN5100 2TB NVMe from Western Digital

Fuck AI.

This will come up again later in the post, but I wanted to bring it up now as well. Fuck AI.

Fortunately I actually had plenty of storage available to be reallocated from Windows to Linux, so I just did that.

The migration took around 3 days, on and off - about 2 hours per day - in order to set everything up, migrate the software I wanted, and finally clean up the Windows installation of almost everything that had already been migrated over to Linux. Despite being somewhat familiar with Linux, I still tried to follow the setup instructions from Linux Mint to the letter. The instructions mostly worked, but I still had to drop to the CLI in order to finish installing NVIDIA drivers:

  • I had skipped over the SecureBoot pop-up when rebooting the system after installing the drivers;
  • that blocked the drivers from being loaded during boot, even though they showed up as installed in the OS UI;
  • unfortunately you can’t get that pop-up to trigger again from the UI and need to use a CLI command for that;
  • the documentation doesn’t mention the pop-up in the installation steps and I tried to follow the docs as closely as I could.

I’ll try contacting Linux Mint about potentially getting that SecureBoot pop-up described in the docs.

While the experience of migrating is still a bit too cumbersome for non-technical users, the experience of migrating is so, so much better compared to 10 years ago. And, possibly more importantly than that, the rate of adoption seems to be increasing quite rapidly. Give Trump a few more years and Linux could reach 10% adoption by the end of his term.

The most shocking change compared to the Linux experience of yonder is just how easy gaming has gotten on Linux - not because of increased native support, but because the compatibility layer is fantastic. I think a lot of that has to do with Valve putting their weight behind it and it really shows. Not everything works though - Battlefield 6 for example requires kernel level anticheat that won’t work on Linux - but most games I’ve played so far (EVE Online, Baldur’s Gate 3, older titles from GOG etc.) just worked straight out of Steam. Getting Battle.net to work took a bit more effort (via Lutris), but it does work just fine. Compare that to my experience with gaming on Linux about 15 years ago when I could barely get League of Legends to run after a few days of trying, though with significant visual artifacts.

In any case, after I had migrated I found out about the International Criminal Court ditching Microsoft after their services went down shortly after the conviction of Netanyahu and Trump imposing sanctions on one of the ICC prosecutors (allegedly - Microsoft denies the connection and claims this was just a technical issue). Not that it influenced my decision in any way (as I had already migrated), but it just makes me more confident that this was worth the effort.

Beyond Linux the rest of my migration to EU-based software is continuing - I’ve been experimenting with custom domains for e-mail (got it working, but still need to buy my new domain), as well as getting the last bit of data I had (this blog in fact) off of AWS and onto OVH (French cloud provider) and bunny.net (CDN from Slovenia). That still leaves the final major hurdle of Google Drive / Photos, but I’m confident I’ll be able to get that sorted in the next 3-6 months, time permitting.

The personal dashboard project and EVE Online

I had been working on a general-purpose personal dashboard whose first functionality would have been related to EVE Online - specifically an improvement on top of SMT, a 3rd party intel tool for EVE, where in addition of monitoring chat logs for relevant intel, it would also notify me if the combat log showed that combat had stopped (which meant I needed to warp to the next combat site).

Fortunately for me, moving to Linux made me aware that SMT is only supported for Windows. Rather than getting it to work via a compatibility layer, I instead looked for native alternatives and found RIFT, which is absolutely amazing. Not only does it handle intel monitoring and combat logs (which was the main functionality I was building my dashboard for) but it also:

  • comes with an integrated Jabber client for fleet pings;
  • has a really pretty display for intel showing character / corporation / alliance portraits, ship icons etc.;
  • it even pulls data from zKillboard, showing you nearby kills even if those are not reported in the intel channel (fantastic idea, wish I had thought of that);
  • it comes with predefined settings packages for major alliances like The Imperium, meaning I don’t need to spend any time configuring things like lists of intel channels;
  • it even supports pushing notifications to your phone!
  • plus a whole bunch of other features that I haven’t had the time or interest to check out just yet.

Honestly I am amazed by the complexity and feature richness of RIFT. Holy shit. It should therefore be no surprise that I’ve abandoned by work on the personal dashboard until I have a specific need for it.

Everything must include AI

(To be clear - whenever I say AI in this post, I specifically mean the wave of generative AI / LLMs that’s currently all the rage, not all forms of AI)

I can’t wait for the bubble to burst, I’m so sick of hearing of overblown, insecure (looking at you Moltbot!) and / or unnecessary applications of AI. Seriously, Copilot in Notepad and Paint? Or an AI description of each product in a local grocery-ordering app? C’mon man, I know what a banana is, I don’t need AI to burn some rainforest in order to tell me what a banana is.

Hopefully once the bubble bursts we’ll be able to focus on the few real applications of AI. In my mind AI is a great tool for anything that fulfils these technical conditions:

  • correctness / precision is not important;
  • replicability / stability (either short or long term) is not important;
  • if using AI to fulfil a task (such as using AI for customer support - which is currently a terrible idea), a non-AI (or “human”) alternative path should be easily and immediately available and you should not be brick-walled by AI;

Of course, even if these conditions are fulfilled AI is still horrible due to eating up so much electricity and water and stealing copyrighted work. Until we fix these problems AI will always remain a terrible technology, but for now let’s assume that we somehow magically fix these issues.

If these conditions are met then I think AI can be useful in some specific use-cases.

Can AI be your doctor? No, correctness is important.

Should you use AI as the foundation for your infrastructure? No, infrastructure needs to be stable in the long term and randomness is an inherent part of AI.

Can you use AI to cut jobs (for example for customer support)? Probably not (or not by a lot), since a human alternative path should be easily available. And if you’re planning not to have it - fuck you.

The primary instance that I’ve found that fulfils those conditions is the Fuzzy Search Engine. Specifically, I mean asking the AI various questions with the clear expectation that it will get things wrong. I’m not looking for an answer, I’m looking for keywords. Those keywords will then serve as topics to search for and learn about the traditional way, which means the correctness and replicability of the answer don’t matter (and in this usecase no human alternative is needed).

To be clear - AI would not be a search engine, it would be a “fuzzy” search engine - it would sometimes get things wrong, sometimes get things right, most times it would get things partly wrong and partly right. It can’t provide information, it can only generate ideas of what you could search for instead.

Even in this use case, with the clear expectation that it will fail, it still is annoying when it fails, but that’s mostly due to the confident tone it uses.

For instance, I was looking at how to set up a control scheme in Unity where D-Pad would perform an action, but LT + D-Pad would perform a different action. Unfortunately, just setting things up this way would trigger both actions when LT + D-Pad would be used. I asked ChatGPT and it said this is a very common problem and that the solution is basically to have a single action and check inside of it if LT is being used at the same time. Hacky and ugly, but I suppose it works.

However, I was doubtful that such an important system had such a glaring omission, so I continued digging and after about 10 minutes of searching online I found that Unity does support input consumption (which basically means that the more complex input rule, LT + D-Pad in this example, “consumes” the input when handled, preventing it from being consumed by the D-Pad-only action). Why this is not on by default is beyond me.

When asking AI why it didn’t offer that solution it said:

You’re right to be frustrated - and you’re also right that Input Consumption is the actual feature designed to solve exactly this problem. I should have mentioned it earlier. Thanks for calling that out.

So - fuzzy search engine - you won’t get correct, complete or precise answers, but you will ocassionally get keywords for things you might want to research further. Definitely not the groundbreaking, industry-shattering use of AI that is being thrown around these days, but still an actual real-world application of AI.

Anyway, can’t wait to see how the world will blow up tomorrow. See you next time!