Joined November 2017
- Tweets 367
- Following 336
- Followers 308
- Likes 358
72 Photos and videos
Coming soon: GenoMark.
Autoresearch at home.
ZK verification for every result.
Train locally, submit confidently, prove it cryptographically.
π§¬β‘
@karpathy @ensue_ai @MinaFoundation @MinaDevelopers @MinaProtocol @minacryptocom
1
3
120
Every AI memory system I tested fails at contradiction detection.
- Structured_memory 36%
- Mem0: 29%
- OpenAI RAG: 0%
We built a benchmark to understand why.
The problem isn't implementation.
ππ‘π ππ π¬ππππ€ π’π¬ π¦π’π¬π¬π’π§π π π₯ππ²ππ«. π§΅
3
1
272
Example failure:
store: "user is vegetarian"
store: "user ate steak"
RAG result:
"User ate steak"
Correct result:
truth_state = BOTH
confidence = 0.52
conflict_detected = True
100
Claude Code just added structured auto-memory.
Thatβs a big step forward for agent memory.
But it also highlights something important:
most AI systems still treat memory as stored notes.
The harder problem is belief maintenance. π§΅
1/10
4
1
145
If you're experimenting with Claude Codeβs new memory system:
you can layer belief maintenance on top.
I wrote a guide showing how to integrate MnemeBrain with Claude Code:
mnemebrain.ai/blog/mnemebraiβ¦
1
83
Example of belief maintenance:
belief: "user is vegetarian"
evidence: "user ate steak"
truth_state: BOTH
confidence: 0.52
conflict_detected: true
45
This is the layer weβre exploring with MnemeBrain.
Evidence
β Belief node
β Truth state
β Confidence decay
Belief maintenance instead of retrieval.
9/10
1
51
Claude Code is pushing the ecosystem forward with structured memory.
The next step is belief systems.
Memory β beliefs.
10/10
46
If you're building AI agents that remember things over months:
the missing layer isn't retrieval.
it's belief state.
mnemebrain.ai/
60
BMB is open.
The MemorySystem interface is public. Any system can implement an adapter.
We'd love to see adapters from: @mem0ai @langchain @zep_ai @llama_index
If you beat 100% β great. If you don't β now we know exactly why
`pip install mnemebrain
python run_bmb_benchmark.py`
1
111
The belief layer for AI agents.
Not a better memory library.
The TCP/IP moment for agent memory.
github.com/mnemebrain/mnemebβ¦
77