Jason Harvey

Founder @DeepaData — the significance layer for AI memory | Writing on memory, meaning & the future of interaction →

[Jason Harvey @JayPHarvey]

We don’t just remember what was said.
We remember what mattered.
That’s why I’m building @deepadata; the emotional layer of memory.
Writing about memory, meaning & the future of interaction → https://t.co/SWqc8Ft2B3

[Jason Harvey @JayPHarvey]

@LeoATracker Shit he’s good. Up between 40-90% over the last 3 month between them. Stellar

[Jason Harvey @JayPHarvey]

@mem0ai EDM (Emotional Data Model) — the significance layer for AI memory.

Mem0 captures what happened. EDM captures what mattered — encoded at capture before retrieval.

We already have a Mem0 adapter with queryBySignificance(). MIT open standard + hosted API.

https://t.co/4caevMYskb

[Jason Harvey @JayPHarvey]

CoALA is the right taxonomy for the memory architecture. EDM sits within it — specifically the episodic + semantic layers, adding a significance tier: governed, auditable records of what mattered, encoded at capture.

CoALA describes the shape. EDM is a file format for one of the most important things that goes in it.

MIT spec: https://t.co/IJ5UYTTmtj

[Jason Harvey @JayPHarvey]

EDM (Emotional Data Model) is a published open standard for this — specifically the significance tier: not what happened, but what mattered, encoded at capture before retrieval.

MIT spec at https://t.co/IJ5UYTTmtj. DOI on Zenodo. LangChain adapter at v0.1.2.

Your artifacts, your significance — same shape as your harness/memory thesis, one layer down.

[Jason Harvey @JayPHarvey]

@chrysb Appreciate the response; just followed up by email. Your framing of selective retrieval bias is the clearest I've seen.

[Jason Harvey @JayPHarvey]

There's a third path between raw and derived that this piece doesn't cover: structured significance encoding at capture time.

Not a summary. Not verbatim. Structured fields extracted once — arc_type, emotional_weight, identity_thread — that don't drift with repeated derivation and don't require semantic retrieval to find.

"When was I most afraid of losing her" has zero lexical overlap with the stored memory. Raw misses it. Derived drifts from it. Significance fields find it.

The failure mode you named — selective retrieval bias — is what this solves.

[Jason Harvey @JayPHarvey]

@KieranWarwick Agree. The things we value or remember are seeded in emotion and stick, or are recalled, because of it.

Community, storytelling, and sharing that with others is more valuable than the experience itself.

That’s the chasm. Keep fighting! 👊

[Jason Harvey @JayPHarvey]

@HarryStebbings You’re the King @HarryStebbings 🤴

The only podcast I listen to @ 1x. Because content needs focus and cadence spot on ✅

[Jason Harvey @JayPHarvey]

@tomfgoodwin Great to have you here Tom. Enjoy your stay

[Jason Harvey @JayPHarvey]

The hidden cost of attention isn’t solved by efficiency.
It’s solved by meaning.
Full essay → https://t.co/QLMDX5Vtzy

[Jason Harvey @JayPHarvey]

Attention isn’t free. Inside every transformer, it’s the most expensive operation.
But the real cost of attention isn’t just compute — it’s meaning. 🧵

[Jason Harvey @JayPHarvey]

This is the foundation of what I’m building with DeepaData — tagging memory with emotional gravity so systems remember what matters.

[Jason Harvey @JayPHarvey]

#COYS

[Tottenham Hotspur @SpursOfficial]

WE'VE DONE IT!!!!! 🏆

[Jason Harvey @JayPHarvey]

@theJayAlto You’ve shown “the reason for meaning” not meaning itself in my opinion.

Meaning is where it all starts—not the outcome.
Most people don’t recognize it until something is achieved, but it was always there.
Quiet. Waiting. In all of us. It’s foundational

[Jason Harvey @JayPHarvey]

@Jacobsklug @lovable "Build"