Mark Miller

Building Lodge. Big ontologies

[Mark Miller @markpm39]

@ar0cket1 Not to mention data

[Mark Miller @markpm39]

@0x_Apex @cursor_ai oh wait https://t.co/p3TQzvSyHB

[ben hylak @benhylak]

um, guys?

[Mark Miller @markpm39]

@0x_Apex @cursor_ai Yeah I mean I just feel like benchmarks are a little arbitrary in general. They probably just pick and choose ones that make them look good

[Mark Miller @markpm39]

@kitten_beloved Well I’m cooked

[Mark Miller @markpm39]

@cursor_ai i would imagine spacexai wouldnt optimize for benchmarks so im optimistic

[Mark Miller @markpm39]

@cursor_ai I really hope these evals are representative of the model quality I will happily become a cursor/spacexai supporter over anthropic/openai

[Mark Miller @markpm39]

Yes, this is the goal. The primary bottleneck is building the structure of the business which requires some sort of structure (Lodge's ontology/operational layers), and then actually making it usable for an operator.

The first part is an engineering problem and the second is fortunately a very nice application of agents.

[Mark Miller @markpm39]

I would also like to add that trying to even define how you're going to model the real world and your operational workflows can indeed be conceptually complex. You have to consider issues such as:

- How am I going to model my business? What real-world objects do I care about? What relationships? How do I make sure the schema is philosophically valid?

- How am I going to define my operations? What objects do I need? What actions?

- How can I put these together for powerful automations?

These problems are well-addressed by AI agents. They are all difficult/tedious to construct manually, but easy to verify. (reminds me of an np problem).

An operator probably doesn't know how to define the schema necessary for his operation, or how to construct automations (one reason why platforms such as Palantir are too heavy for smaller companies).

However, the operator does have 1) a mental map of his business, and 2) an understanding of what his workflows look like, and what successful operations looks like.

Thus, if an agent has access to a platform that models the world as the operator knows it and a safe, well-defined operating layer with provenance based on the operator's workflows, it greatly reduces the burden on the operator, reducing the majority of his work to verifying rather than building.

This in turn vastly increases efficiency, all while removing the unnecessary degrees of freedom that cause security concerns with standard agentic automation.

[Mark Miller @markpm39]

Complex operations can become much simpler with great underlying data models.

Lodge solves this problem by building automations with an operational layer that sits above your ontology.

The ontology is what defines the real world objects you care about (business objects, people, etc.)

The operational layer is what lets you track the status of your operations (process ticket, tracking status, etc), and act on them.

This allows you to construct stateful operations over a live map of your business, which is far easier to maintain than dozens of individual automations across several domains.

[Mark Miller @markpm39]

Complex operations can become much simpler with great underlying data models.

Lodge solves this problem by building automations with an operational layer that sits above your ontology.

The ontology is what defines the real world objects you care about (business objects, people, etc.)

The operational layer is what lets you track the status of your operations (process ticket, tracking status, etc), and act on them.

This allows you to construct stateful operations over a live map of your business, which is far easier to maintain than dozens of individual automations across several domains.

[Mark Miller @markpm39]

how is there not a physical place in my city selling a c19 power cord

[Mark Miller @markpm39]

@HatforceSec @overleaf Nice! How does it compare to prism?

[Mark Miller @markpm39]

@gabu3d_pl how is ttft

[Mark Miller @markpm39]

@mouse_math Literally only heard Euler

[Mark Miller @markpm39]

@thdxr two economists are walking in a forest...

[Mark Miller @markpm39]

I built a an ontology over college basketball coaches (who they have coached under, who they coached)

one funny connection was Greg Gard (current coach of my college) -> Bo Ryan -> Steve Yader -> Ball State -> Billy Taylor (my dad's college teammate, coach of Elon) -> John MacLeod, Fran McCaffery (my dad's college coaches)

Apparently Ball State (which seems completely random and I would have never thought of) is the only thing linking these two coaching lineages I'm related to

[Mark Miller @markpm39]

@ThePrimeagen After all😭

[Mark Miller @markpm39]

@ninja_maths Very cool! How was this constructed, I.e. how do you model connections between subjects?

[Mark Miller @markpm39]

@Daniel_Alami "this i think remains an underexplored area due to lack of talent supply"🙌🙌