Mikela
castironflower@hachyderm.io
<p>ex- San Francisco Digital Services Director of Engineering, PivotBio, CfA, MoJo. Opinions are my own. Too gay, Gunpla and barebow hobbiest,wife+mother. I used to do science. <a href="https://hachyderm.io/tags/django" class="mention hashtag" rel="tag">#<span>django</span></a> <a href="https://hachyderm.io/tags/ansible" class="mention hashtag" rel="tag">#<span>ansible</span></a> too many JavaScripts and data engineering</p>
Posts
-
View post
going to be shameless and let people know the only reason #sfgov is online at all is that i fought tooth and nail to move off #php to #django you will notice sfgov.org is suspiciously gone now as its unpatchable old php its kindov sad they let me go because migrating was hard(but we did it) and not wanting to break gdpr but if your happy sf still has a website right now consider hiring me #blackoctober #php2shell
-
View post
Reader beware got a glow up after 5yrs docs still buggy but editor works great https://glassresistor.github.io/reader-beware/#builds
-
View post
and i made a thing still in early beta but using it as my daily driver https://github.com/GentianLine
-
View post
big thoughts: while i find it vanishingly possible that llms or models in general are conscious. models are the connections/tape drives of minds, more or less the cache of thought and sure a self training harness and model is more conscious. those examples burn out. models as intelligent though? maybe? i think is &quot;reasoning&quot; in a human mind and chain of thought in a computer *might* be matters of breadth of accuracy humans dont reason perfectly and often reason by talkin...
-
View post
A zero-training memory harness for open chat models (Qwen and friends). The same model that reads and writes text also classifies its own context as long-term / short-term memory while decoding; a small deterministic eviction engine then enforces the token budget. Design inspired by Context Language Models (Shao et al., arXiv:2609.37725): context is treated as an editable artifact mirrored to disk, and context management is delegated to the model via prompting with no fine-tuning, works off the...