January 2026 - Position on AI

Given the fast moving nature of these things, and how heavily it's still being pushed by all the Big Tech interests, I guess it's time for another update on my current position on AI stuff

 

General Position

Get your stupid effing "Large Lying Model" / "Grand Theft Autocorrect" / "Pervasive Spying Narks" away from me!!!

For the record (even if it's useless actually saying this):  

In general, I DO NOT consent to any of "my data" or any of my work / creative outputs being used for AI training / processing, except for very specific case-by-case purposes.

While I'm asking nicely: Please cease and desist any operations where this may have happened, and delete all relevant records or thus-trained models  🙂 

 

For Coding / Software Development 

For work stuff, my current position on this stuff is:
* 1) If I'm stuck on trying to resolve a sticky "compiler says no" problem (BUT only after having given it a decent stab), then I will now consider getting AI to suggest a fix / solution, since it's there and may be able to help.

I will however scrutinise whether it passes the sniff test for bad code... (and have on a few occasions just thrown that approach out and changed tack, after seeing how bad it can get!)

* 2) For big tricky changesets, I will also ask it to do a pass over the code and check for any obvious issues (since we have found it can catch or at least flag a whole bunch of stuff that tired eyes might miss)

* 3) I've had mixed results getting it to do any bulk code generation (mainly unit tests), and TBH don't trust it to do actual feature implementation (which is usually the fun bit that us humans should get to do anyway!)... so: So far there's really not much need / use for it on that front from my end!

 

Oh... and this only applies to work problems, where work has said we should experiment with these tools + fenced off some "approved" tools that appear suitably tame.

 

(*) EDIT (22 Jan): 

I have also had some luck with getting it to generate a whole bunch of boilerplate for translating between one enum definition, and a protobuf protocol one (that I got it to generate from the first), and then writing all the conversion logic (that would've been tedious to do by hand). The best solution of course is to not need that boilerplate (i.e. the language / compiler should really be able to deduce it itself), but since our environment isn't set up for that sort of "experimental" stuff, having this tooling available when cases like this call for it is certainly nice...

 

For personal stuff:
Nah, I like the challenge / can deal with whatever it is over a longer timespan when I can be bothered...  😜

 

Exceptions (for personal stuff):

* Wrangling intractible CSS / Javascript problems - As those are nasty, and I suck at that

* Wrangling compile errors stemming from trying to build publicly available third-party OSS repos