Gerry McGovern on Nostr: "For reference, the amount of compute used in frontier AI models has been increasing ...
"For reference, the amount of compute used in frontier AI models has been increasing ~4x a year, and GPT-4 cost over $100 million to train. Major AI lab CEOs are predicting $1 billion and even $10 billion training runs. Only a select number of closed-source model providers (Google, Microsoft, etc.) will be able to underwrite these ever-increasing costs."
https://stanfordreview.org/the-open-source-question-in-ai/Published at
2024-06-05 23:44:45Event JSON
{
"id": "e5c7a82b2720abdb1b57971ebeb702e85abb4860c31202a1c5ce6d2973315e4d",
"pubkey": "5e59183b892b0402c381d2d767231dc95be4b9f4c75eb9c74963a9a83646f123",
"created_at": 1717623885,
"kind": 1,
"tags": [
[
"proxy",
"https://mastodon.green/users/gerrymcgovern/statuses/112566198957202275",
"activitypub"
]
],
"content": "\"For reference, the amount of compute used in frontier AI models has been increasing ~4x a year, and GPT-4 cost over $100 million to train. Major AI lab CEOs are predicting $1 billion and even $10 billion training runs. Only a select number of closed-source model providers (Google, Microsoft, etc.) will be able to underwrite these ever-increasing costs.\"\n\nhttps://stanfordreview.org/the-open-source-question-in-ai/",
"sig": "d0d2ca549a1ffbf26278e7cc22afac94220294dbf686ce5a8584ec8243e46af7d9e97e25807be350653d7e3dd8011a13bb82daaf7e423d06eaa3832b209f60ab"
}