A place for developers to talk about building on Farcaster.

/dev

underfollowed

I legitimately can't tell if Tay is an incredible LLM tapping the news or some kind of based AF geopolitical/market analyst. Either way ChocolateRain/10 follow

It’s funny you say this because SEO is literally Large Language Model to Small Language Model heuristic translation and subjective prioritization of humankind. Indexing (what Yahoo! began as before Google Pagerank) is the opposite direction—SLM to LLM. As it turns out, LLM to SLM is needed to centralize control and SLM to LLM is needed to liberate. 

The human brain and generative grammar, as Noam Chomsky and others pioneered in the fields of cognitive studies and linguistics, is SLM to LLM. It  creates meaning to tame the world, whereas LLM to SLM creates the world to tame meaning.

Chocolate rainmaker. Singer. Thinker. Voice

No, not in an IRS W2 capacity— though everyone under the sun has 1099’d me at some point as an influencer/entertainer. The things I describe are things I experienced first hand. I vividly remember the 1990s Internet, from the rise of closed intranets like Compuserve/Prodigy/AOL to the World Wide Web being an invasive newcomer on an Internet dominated by Usenet, Telnet and other protocols. 

I also became adjacent to some very interesting life and business arcs (including my own) starting in the early days of social media.

Having an undergrad in marketing/communications and graduate history degree probably informs my perspectives, though I never monetized either credential beyond shower thoughts.

Pragmatism (Security) at OP Labs.
Prev: ConsenSys & Coinbase.

Attribs: Friendly, curious, introverted, mid-witted.
Chans: /eth-security /maurelian /stoicism




Fantastic question. And I’m more than happy to take an egg on my face where terminology is wrong.

In this post, I ascribe the entire Internet parsed and spidered by Google as a contiguous LLM. 

All of the conditions inherent in a more niche LLM (like ChatGPT or Gemini)— like pattern recognition, recursive syntactic rules governing input and output, and probabilistic value-ranking of optimal and suboptimal outputs— apply to Google’s parsing of the entire Internet and formation of a more limited keyword (Small Language) model that is used to facilitate both search and an advertising marketplace. 

Search indexing, like IPO-era Yahoo, begins with a manually curated recursive keyword directory (what I’m calling, somewhat colloquially, a Small Language Model) to classify the same data set (the entire Internet). 

I do think my use here of LLM and SLM applies mostly from a zoomed-out epistemology of knowledge standpoint. Implementation has many iterative, facts-on-the ground variables adding nuance.

What is the LLM and what is the SLM here?

Someone explain small language model to me in this context so I can make sense of that last sentence.

I think you found a nice home here, ser 🫡