This I haven't heard. Did they put dedicated machinery in there that can *only* be used for LLM training?
Generally, the same math used to train LLMs can be used to solve other problems; I wouldn't be surprised if part of their computation of cost / benefit here was "Even if LLMs don't pan out, oh no, we filled a room with fast linear algebra solvers, guess we'll just have to... Model nuclear physics for the next generation of reactors or fold proteins or something."
Mark T. Tomczak
npub10z…aga0j
2025-08-22 14:30:45 UTC
in reply to nevent1q…nx23
Author Public Key
npub10zv2wul2hvcg6z9445zjdm5yldmyc6sqyr0u4xnnyygtfeym746s2aga0jSeen on
wss://relay.momostr.pinkPublished at
2025-08-22 14:30:45 UTCEvent JSON
{
"id": "22a7bfdb49c9a27809794fdee1c45f60918c8d72e2afd5004590fd32792e525b",
"pubkey": "7898a773eabb308d08b5ad0526ee84fb764c6a0020dfca9a732110b4e49bf575",
"created_at": 1755873045,
"kind": 1,
"tags": [
[
"p",
"f8347ec37d9f23959ba62dd8391fbcade87a7c071205c96f890432e5c047a6ed"
],
[
"p",
"92874a5a0147e0f352b0814a4d48ff03a046a9f2468c3aa851a641449df059d7"
],
[
"proxy",
"https://mastodon.fixermark.com/@mark/115072895923155398",
"web"
],
[
"p",
"3d18be4a0c506239bb80675a0ca83740092da744979fcb044b262d55b20cc42f"
],
[
"p",
"b9729a87d3f878303b190469f7e8dfdfb5389f0e43ae69c36bab6822ead6283c"
],
[
"e",
"60fbdf1581b953bf96200baddea4288f95e39f418ff8775cff1a01ebc9c22df1",
"",
"root",
"f8347ec37d9f23959ba62dd8391fbcade87a7c071205c96f890432e5c047a6ed"
],
[
"e",
"2ce67661baf774305840af510e9dd589f716d49bed5f359cff18a348b33af845",
"",
"reply",
"92874a5a0147e0f352b0814a4d48ff03a046a9f2468c3aa851a641449df059d7"
],
[
"proxy",
"https://mastodon.fixermark.com/users/mark/statuses/115072895923155398",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://mastodon.fixermark.com/users/mark/statuses/115072895923155398",
"pink.momostr"
],
[
"-"
]
],
"content": "This I haven't heard. Did they put dedicated machinery in there that can *only* be used for LLM training?\n\nGenerally, the same math used to train LLMs can be used to solve other problems; I wouldn't be surprised if part of their computation of cost / benefit here was \"Even if LLMs don't pan out, oh no, we filled a room with fast linear algebra solvers, guess we'll just have to... Model nuclear physics for the next generation of reactors or fold proteins or something.\"",
"sig": "7a8853b2726dc87f1127c2389e959d3b7fc39f2c1190adb2d9f34f171a8bc433abe3cfafcaab94d02365e016cc4e58ab41f9fba610ece717a9f4d0a3d6750ef6"
}