Can an LLM use memories it was never trained on? No retraining, no long prompts.
Dynamic Engram injects new memories on demand as exact n-gram memory.
350M model: 95.5% correct on unseen memories, faster than RAG on CPU.
๐zenodo.org/records/23803703
๐คhuggingface.co/tetiai/dynamiโฆ
Triggers are now live in Teti: "every morning at 8, give me the top tech news" โ say it once, it happens on schedule. Right in your conversation.
teti.ai/hub/news/triggers-arโฆ
Vector DBs bet context stays scarce forever. We bet it won't.
memraw: no retrieval. The whole memory lives in the context, always. Next stop: memory inside the model itself โ Titans, Memory Layers are already heading there.
t.ly/16oy8
Two new Teti models just landed ๐
โก Aura 3 โ fast, multimodal, lowest price.
๐ง Metis 3 โ our most capable model, 91.1% on GPQA Diamond, near-frontier at a fraction of the cost.
Try both โ teti.ai
teti.ai/hub/news/introducingโฆ
#AI #LLM #TetiAI
People keep asking how we cost 4x less than Exa.
Teti.ai users and pixserp users feed each other. Every search makes the cache smarter, faster and cheaper for everyone. Opting in is your choice โ your data, your call.
Pay per use. ๐น๏ธ
๐ pixserp.com
Meet Teti ๐ก๏ธAI designed to amplify human brilliance, not replace it. Thinks deeply with you. Zero tracking, ever. Your thinking partner that evolves with you.
Try it: teti.ai/download-app
#AI #Tech #Innovation #ethics
