The memory update everyone’s been waiting for — v1.10 is out
Shipped v1.10 yesterday. Redeploy only, no migration.
The thing that was broken: ask your memory store for something specific, like a version string or an error code, and it'd come back with the wrong thing. Confidently. Dense search embeds those identifiers so close together it can't tell them apart. It's not a miss, it's a wrong answer that looks right.
Hybrid recall fixes it. Keyword search runs alongside the vector query now, results get fused, exact identifiers actually come back.
Two other things in this release. Memories that survive contradictions rank higher over time, ones that keep losing rank lower. Compression now protects things that have been recalled or survived an argument instead of just compressing based on capture-time importance.
Also rewrote the insight prompt cause it was speculating beyond what was actually in the retrieved memories. That's gone.
Full release notes at https://secondbrain.s.gy/v1_10
v2 knowledge graph is next. Coming July 15.
Replies
Confidently wrong is such a good way to describe it. Those answers are way harder to catch than a simple miss.
Second Brain for AI
This is the failure mode that worries me most in AI products: not a blank answer, but a confident wrong answer that looks usable.
For developer memory, version strings and error codes are obvious examples. In business planning or funding prep, the same problem shows up as invented numbers, unsupported assumptions, or a polished plan that hides missing evidence.
That is why I think retrieval systems need to expose more than the answer:
- source or context used
- confidence / uncertainty
- what was not found
- what the user should verify manually
That is the direction I would rather see in tools like RaiseReady: if a number, milestone, or assumption is missing, say it is missing. Do not fill the gap just to make the output feel complete.
This is the kind of memory update the matters. Wrong answers with high confidence are far more dangerous than misses, so hybrid recall feels like a big step forward.
Second Brain for AI