Iiridayn
@iiridayn.info
Iiridayn.info. Authentication-focused Usable Security Researcher, experienced web and embedded systems programmer, Linux sysadmin, and SkyLords.com operator. Member of the Church of Jesus Christ of Latter-day Saints. I like to filk, read, and game.
It's literally a bit more than just remembering words. The mixing step ("attention" - I found www.3blue1brown.com/lessons/atte... useful) is genuinely fascinating. I took a ML course in 2005, and neural nets were useless past a few layers. Attention changed that by sandwiching layers.
September 14, 2026 at 10:06 PM
It's literally a bit more than just remembering words. The mixing step ("attention" - I found www.3blue1brown.com/lessons/atte... useful) is genuinely fascinating. I took a ML course in 2005, and neural nets were useless past a few layers. Attention changed that by sandwiching layers.
If you have done so, please return and #apologize for falsely accusing the individual of #bigotry. Such accusations are not consequence-free; they cause internal emotional pain and can cause severe external consequences. They damage your own #cause by harming #allies.
August 6, 2026 at 9:02 PM
If you have done so, please return and #apologize for falsely accusing the individual of #bigotry. Such accusations are not consequence-free; they cause internal emotional pain and can cause severe external consequences. They damage your own #cause by harming #allies.
Your knowledge of an issue is not universal. I've now learned more - #Substack has a reasonable position. Those calling to #boycott do too. But I worry that social pressure is prone to becoming similar extremism (cf "The Day The Nazi Died" outro). A #legislative change would fix Substack and others.
July 14, 2026 at 7:41 PM
Your knowledge of an issue is not universal. I've now learned more - #Substack has a reasonable position. Those calling to #boycott do too. But I worry that social pressure is prone to becoming similar extremism (cf "The Day The Nazi Died" outro). A #legislative change would fix Substack and others.
I've seen more focus on #characterization and side plots than before. My understanding is that authors believe that's what readers want - and certainly I see a _lot_ of negative feedback on works that don't provide. Personally I like stories focused on interesting ideas more. But, not mass market.
July 14, 2026 at 6:24 PM
I've seen more focus on #characterization and side plots than before. My understanding is that authors believe that's what readers want - and certainly I see a _lot_ of negative feedback on works that don't provide. Personally I like stories focused on interesting ideas more. But, not mass market.
Likely comes from choice to have a small standard library, excluding things like `left-pad`. Reimplementing things is bad, so pull what you need. But, you end up with a mix of overmodularization (left-pad) and large dependencies (express) which require everything for simple tasks (just in case).
July 13, 2026 at 8:38 PM
Likely comes from choice to have a small standard library, excluding things like `left-pad`. Reimplementing things is bad, so pull what you need. But, you end up with a mix of overmodularization (left-pad) and large dependencies (express) which require everything for simple tasks (just in case).
To answer "why would you ever want to decrease #security, and be merely sufficient?": security can always grow, but generally imposes a #usability burden. The value of security operates on thresholds (can/can't; often linear until exceeding reasonable levels). My goal is to trade less to be #secure.
June 25, 2026 at 8:27 PM
To answer "why would you ever want to decrease #security, and be merely sufficient?": security can always grow, but generally imposes a #usability burden. The value of security operates on thresholds (can/can't; often linear until exceeding reasonable levels). My goal is to trade less to be #secure.
#Claude's rebuttal, tersely summarized: #AI enables delegating thinking at an unprecedented scale, enabling harms previously rare. Learners need generative struggle to develop intuition. Non experts can attempt tasks they cannot verify, with potentially dangerous outcomes. The #scale makes this new.
June 3, 2026 at 7:47 AM
#Claude's rebuttal, tersely summarized: #AI enables delegating thinking at an unprecedented scale, enabling harms previously rare. Learners need generative struggle to develop intuition. Non experts can attempt tasks they cannot verify, with potentially dangerous outcomes. The #scale makes this new.