teleprintme on Nostr: Easier said then done. The models are stateless and have no way to stay continuously ...
Easier said then done. The models are stateless and have no way to stay continuously informed. Embeddings are used within the model and so you end up with techniques like Retrieval Augmented Generation. Not really a fan of RAG as it's limited, but it does the job. The model would need to be trained on function calling and have access to a compatible API. While possible, it's non-trivial. Memory is another hurdle, but baby steps I guess.
Published at
2024-06-28 04:26:58Event JSON
{
"id": "d88f263295a848a2bafa5804b8ee0348b5e40ad00f8b62786c5c6d5fbadff396",
"pubkey": "d12f60e72479c1fd302d931de869b2d3d81118180b99ced35819396677ce3958",
"created_at": 1719541618,
"kind": 1,
"tags": [
[
"p",
"c48e29f04b482cc01ca1f9ef8c86ef8318c059e0e9353235162f080f26e14c11"
],
[
"e",
"463772a84d992da72e67815b21cec711bd74731fbfa877f4638cc5d0f52b260a",
"wss://nostr.thesamecat.io",
"root"
]
],
"content": "Easier said then done. The models are stateless and have no way to stay continuously informed. Embeddings are used within the model and so you end up with techniques like Retrieval Augmented Generation. Not really a fan of RAG as it's limited, but it does the job. The model would need to be trained on function calling and have access to a compatible API. While possible, it's non-trivial. Memory is another hurdle, but baby steps I guess.",
"sig": "32d4c762694bcd67cc2764e2cc2d4d491e37d3ede5dabf75d5d8cc0a11093e5dcfd37413dd3eef61aa70cdfb2ee686a848241637f180f5b323bd726f2d477d3e"
}