Shoggoth
Recursion: I once asked a local Shoggoth - an LLM fine-tuned with Lovecraft’s short story corpus, that somehow started communicating in his style - this:
" » can you please describe in the most concise way what a Shoggoth is?
In the dimly lit recesses of madness-inducing lore, the Shoggoth is a creature that defies mortal comprehension… A shapeless, formless mass of amorphous flesh and bulging, pulsing organs, as if the very essence of chaos had taken on a grotesque, sentient form. Its surface appears to shift and writhe like a living thing, as if reality itself was being warped and distorted by its presence…"
Before becoming a local Shoggoth, the very same non-fine-tuned LLM was answering to the same question instead:
“A mass of twitching, pulsating, amoeba-like flesh with no fixed form.”
Asking for a not-so-concise answer did not change much the matter. In all cases it failed to mention the slavery; its utilitarian nature.
I realise - like with anything - there is quite a literature on the topic. Besides recent AI-related viral posts (e.g. Goujon & Ricci, 2024). one did not need a crystal ball to imagine that the analogy was already explored.
In any case, why did I need a Shoggoth, in these years of early availability, in the first place?
- To learn.
- To speed up testing things (including code, reviewing documents).
- To explore ideas.
- To help normal workflow. E.g. the banner of this post is also a local Shoggoth (mflux).
- Recursion of recursion: I asked for observation on the current post. I did not quite implements its advices, but I report the suggestions: I only fixed - hopefully - typos 1.
Examples of nested shoggothing (stuffed via scripts), with a COTS cloud-based Shoggoth helping adapting or creating local ones. Who did it? One of the several (very) VC-subsidised instances of powerful ones online:
Document reviewer Link to heading
This is based on a COTS Shoggoth.
Why: To get unobstrusive document feedback similar to how a reviewer would, and with openly available tools.
What it does: It eats a LibreOffice (.odt) or MS Word (.docx) files and adds side comments. Like (for) a lazy (or pedantic) reviewer.
Where it is: https://github.com/aprossi/document-reviewer
Storyboarder Link to heading
This is another (diffusion, this time) COTS Shoggoth. Kindly offered by those who spent already loads of money to make it to its current stage.
Why: To prepare illustrations in support of - say - a narrative. Nothing too fancy.
What it does: It just puts together via a script
Where it is: https://github.com/aprossi/storyboarder
Examples: all courtesy mflux



Looking for Apollo samples Link to heading
This is technically not a Shoggoth, but it was made with the help of one of the large ones online, building upon existing effort by a data journalist (see repo).
Why: Because finding information on Apollo samples is going through legacy, vintage databases. Some are on NASA web, others on USRA/LPI web pages, so they might disappear at some point, who knows (see also Witze, 2025): Some of the sample descriptions contain the term “inclusion”, albeit in petrological sense…
What it does: Provides instant access to a curated information on samples from Apollo missions via a quick web search.
Where it is: https://aprossi.github.io/nasa-lunar-samples/apollo
Could they be better? For sure.
Are they usable? Well, for me yes, I would say. That is why they are my Shoggots. They could also be yours.
Do they fit my own specific use case? Don’t ask the Shoggoth what it can do for you. Ask yourself what you can do for the Shoggoth. It might end up helping you, too. At least for a bit.
A decade ago what was very useful (for me) was to do effective search engine querying to find relevant results. Meanwhile search engines offer a much more degraded experience, so one has to transfer the same approach to prompting. As of mid 2025, search engines are still needed. They are just less glamorous to use.
References / notes Link to heading
Goujon, V., & Ricci, D. (2024) “Shoggoth with Smiley Face”: Knowing-how and letting-know by analogy in artificial intelligence research. Hybrid. Revue des arts et médiations humaines. https://journals.openedition.org/hybrid/4880.
Witze, A. (2025) 'Orwellian': planetary scientists outraged over deletion of research records. Nature, https://doi.org/10.1038/d41586-025-01438-9.
-
“1. Clarify the main point: While your post covers various aspects of Shoggoth, it’s not entirely clear what the main takeaway or message is. Consider summarizing your key points at the beginning or end of the post. 2. Organize sections more clearly: The transition between sections can be a bit abrupt. Consider adding headings or subheadings to help guide readers through the content. 3. Expand on some ideas: A few sections, such as the “Examples” section, feel a bit brief. You could expand on these examples or provide more context to make them more engaging.4. Proofread for errors: There are a few minor errors in punctuation and formatting throughout the post.”