Low-hanging fruit I'd like to see picked

Things I’d like to see done in the world; I may pick these myself at some point; general tactic of publishing lists to the internet; harnessing the internet exobrain; explicitly describing demand to the market and hoping for supply; spitballing; having a place to come back to when I’m bored; providing an example for others to emulate (you should have a “low hanging fruit I’d like to see picked” post); asking for what I want; training my brain to spot free energy in the world;

Things which take ~30 minutes

• Prompt a full novel on hypersitionai.com; project by Aaron Silverbook et al to hyperstition the training data into making LLMs more aligned. According to Silverbook, Anthropic is fine-tuning on the 500 first novels
• Update my system prompt based on toying around with Claude Sonnet 4.5; people make/edit their system prompt way too little and yet there’s so much potential in expectation here. A common mistunderstanding when it comes to LLMs is that they are “tools” and in that sense relevant on the object level only. Only, if you spot a limitation in an LLM, or critique a certain style/approach it takes, you can just tell it that. In other words, any LLM meta can immediately become LLM object. This is why the usage –> system prompt update loop ought to be extremely short
• Write a links post on Twitter; many authors I respect write links posts / share links with their friends, something I happen to find eminently valuable from my end! But what do I read? Why don’t my friends know this? I’ve found links posts to be extremely valuable! Links wouldn’t make for a good Croissanthology.com post, so how about I just write it up on my Twitter?
• Set up an RSS feed; email inbox is a mess; some extremely valuable sources of information don’t even have a native subscription feed; avoids spam more easily; provides something high-value to idly scroll vs. e.g. the Twitter algorithm; writer-side incentives depend less on oft-signal-antiselecting attributes like “virality”, because you’re reading/seeing everything they write no matter what.

Things which take ~2 hours

• Create/edit Wikipedia page for “hyperstition”; I looked it up during the making of this post, and the Google AI overview for it is fine, but there’s no easy way to link to it¹; why is there no Wikipedia page for it yet?
• Write a post creating new Wikipedia pages Dynomight asks “I think about this post a lot—is anything that we write that isn’t a Wikipedia page self-conceit?” i.e. why write under one’s name/pseudonym when one could instead be read by vastly more people and have one’s research/thoughts become the reference doc in any given topic? Just so we can be credited? Are we that vain? (Of course, things are not so simple.) Untested hypothesis that to create new pages will not expose one to the Wikipedian deletion demons. How does one do this? Where is the online guide for this?

Things which take ~a day

• Attempt to fully automate oneself;

---