Microsoft wrote this on Twitter today:
Please don’t just say “hello” in chat.
Seems they have significant costs of running the model and of course many users just write ‘hello’, which gives them neither valuable feedback nor can it be monetized on. That’s why the trend goes towards smaller and more efficient models, and Meta realized that by creating LLaMA, which with just 13 billion parameters outperforms GPT-3 with its 175 billion parameters.
streetyogi /
npub1nu…8qmun
2023-03-08 02:49:57
Author Public Key
npub1nujsvauvdlj8rat0rhvrksa56wuwr7k6mteh3pqtfnwnnmtxug9sg8qmunPublished at
2023-03-08 02:49:57Event JSON
{
"id": "77a7d546aa0bf7c752d0256b48fb02d04fc8c7362315a04369071fe7705bdfa9",
"pubkey": "9f2506778c6fe471f56f1dd83b43b4d3b8e1fadadaf378840b4cdd39ed66e20b",
"created_at": 1678243797,
"kind": 1,
"tags": [],
"content": "Microsoft wrote this on Twitter today:\n\nPlease don’t just say “hello” in chat.\n\nSeems they have significant costs of running the model and of course many users just write ‘hello’, which gives them neither valuable feedback nor can it be monetized on. That’s why the trend goes towards smaller and more efficient models, and Meta realized that by creating LLaMA, which with just 13 billion parameters outperforms GPT-3 with its 175 billion parameters.",
"sig": "0d1411b121b3ee04c11497b2f1d5eba631ea019a849e2172cce943c771d8d4684141c3430f281276f8bfa273b2679d4acf105e217546a7ef86e354286d6340f5"
}