Colin

@colin@mckellar.social
16 Followers
170 Following
207 Posts
Native title lawyer, interested in video encoding, 3d printing, and technology generally.
@bigzaphod the authors were very involved in making the show. I wonder how many of the changes to the characters and/or the plot came from the authors themselves.
@camwilson I heard your interview on If you’re listening. It was great. I look forward to reading your book. I mostly consume books in audio form. Will your book come out on audio? Or should I pick up a paper version?
@liamvhogan practise/practice just what on earth

This guy kept a diary which, among other things, records having killed almost 20 Aboriginal people around geraldton 1850s. The diary is now being published, but will remove references to the killings, because “we are too close to the moment” to look at it “dispassionately”. And “the nation is not yet ready to understand its often bloody and brutal past”

If a nation is not ready to understand its bloody and brutal past 170 years later, it likely intends to never so do.

https://www.theguardian.com/australia-news/2025/aug/05/the-descendants-traditional-owners-call-for-diaries-of-wa-settler-who-killed-aboriginal-people-to-be-released-in-full-ntwnfb

Traditional owners call for diaries of WA settler who killed Aboriginal people to be released in full

Major Logue’s diaries use code to describe the frontier killings of Yamatji people but a local publisher plans to omit the coded sections from publication

The Guardian

Wow, "AI" is doing the thing is absolutely known for doing, which is pulling nonsense straight out of its cybernetic ass, who could have possibly predicted this utterly predictable turn on events

https://gizmodo.com/fdas-new-drug-approval-ai-is-generating-fake-studies-report-2000633153

#uspol

FDA's New Drug Approval AI Is Generating Fake Studies: Report

The AI, dubbed Elsa, is supposed to be making employees better at their jobs.

Gizmodo
@ian it is apparent that the AIs had been trained on these gazettes (they are available on the internet after-all) so the language is not a problem. I can imagine that AI technology could be very helpful with OCR, but as you say, the OCR engine should be running the show. Some of the AI output was so bad, it was almost as if the prompt was “generate a page of text inspired by this image”.

Improving AVIF in Open Source via the Halide blog

https://halide.cx/blog/improving-avif-in-open-source/index.html

I wanted to write this to explain some of what went into Tune 4/Tune IQ, & help outline some of my reasoning for starting Halide Compression in the first place. I hope you enjoy!

#av1 #avif #encoding

Improving AVIF in Open Source | Halide Blog

Had a couple, "I'm suddenly old" feelings in the last couple days during the process of getting some quotes for our roof (which is 20 years old) and maybe adding solar to it (if we can somehow swing it with a refinanced house loan or something).
The most surprising finding was while google docs does a really good job, Googles AI is a complete disaster at OCR. It hallucinated the majority of the test producing credible, but false output. I am worried that people may be using this, expecting accurate text detection.

In my day job I spend a lot of time reading and searching through old scanned government gazettes. I have long wondered how good their ocr was, and if I could find software that could do better. I compared a bunch of different options, and wrote about it here:

https://colinmckellar.com/2025/05/28/ocr-comparison/

OCR comparison – Something Something