Summary of "LLMs Are Databases - So Query Them"

Overview

This video argues that (at least the FFN/knowledge component of) transformer-based LLMs can be treated as a real graph database rather than a metaphor. The speaker claims:


Core technical claim: LLM weights as a graph database

The model is described as having a three-stage architecture:

  1. Syntax/context understanding (early transformer layers)
  2. Knowledge retrieval from weights (middle layers, represented as graph edges/features)
  3. Output formation / token commitment (later layers)

The speaker further maps these components to transformer parts:


Tooling / method: Larql queries over model weights

The speaker uses Larql, described as a “language to query large language models.” They connect Larql to the weights of Google Gemma 3 (34B) (and mention Gemma 4 as a future topic).

They probe the internal representations into a knowledge-graph-like structure:


“Describe” and entity-level browsing (read capabilities)

Using describe France, the speaker claims the model’s internal graph can be browsed as follows:

They also emphasize polysemanticity (noise overlap):


SQL-like graph queries (edges/features as database tables)

The speaker demonstrates queries analogous to SQL over the internal graph.

Query borders for France

Example (as described):

The claim is that:

Query nationality for France

Example:

They claim nationality-related outputs cluster across features/layers (e.g., Germany/Sweden/Italy; plus other sets in different layers).

Nearest-neighbor queries

Example:

This is used to show “country clusters” around France, with other countries appearing in the nearest set.


FFN “features” explained as edge mechanics

The speaker claims:

They interpret the FFN as the knowledge store:


Probing feature reuse across layers (why polysemy happens)

The speaker shows that feature indices are reused at every layer, but their semantic meaning changes across layers.


Relation schema discovery (“show relations”)

They use show relations to list discovered relation types and counts.

Key point: the speaker claims this relational structure was learned during training, not manually provided.


Inference tracing: answer derivation via “graph walk”

The speaker demonstrates actual inference:

A key Larql capability highlighted:

They claim their Larql inference uses:


Write capability: insert facts (training-free editing)

A major demonstration is described in two phases.

1) Baseline: hallucinations for unknown facts

2) Insert into the graph

They perform an insert:

  1. insert into edges (entity, relation, target) values (Atlantis, capital, Poseidon)

After insertion:

They then check for leakage/breakage:

Patch overlay vs compilation

Inserted edits initially live in a patch overlay:

To make changes permanent, they demonstrate compilation:

They can export and reload:


Claimed implications (product/tech impact)

Because:

the speaker argues:


Future items mentioned


Main speakers / sources

Category ?

Technology


Share this summary


Is the summary off?

If you think the summary is inaccurate, you can reprocess it with the latest model.

Video