So here's a thing I've done with LLMs that I simply couldn't do. I do this for my job: I took all the oncall incident reports and had the LLM summarize them looking to ascertain what system (of a pre-chosen list or root cause in the prompt) was actually at fault.
With minimal effort, I got about 97% accuracy with Hermes Nous 7b. And so now I can attribute incidents to root causes better, without having to go back and read hundreds of reports.
How would I do this otherwise?
⛈️ Information ⛈️ /
npub1qz…kgcu3
2024-01-20 02:22:46
Author Public Key
npub1qzk83mw277u9wuhm2hc5ufspr4y90nel4zenw9jyz8ahhmmdyvfqwkgcu3Published at
2024-01-20 02:22:46Event JSON
{
"id": "e3739a303153367ada4aae79afc0e160c496a2c01bb41459f519368ef65cc7f3",
"pubkey": "00ac78edcaf7b85772fb55f14e26011d4857cf3fa8b337164411fb7bef6d2312",
"created_at": 1705717366,
"kind": 1,
"tags": [
[
"proxy",
"https://mastodon.social/users/Elucidating/statuses/111785893305319446",
"activitypub"
]
],
"content": "So here's a thing I've done with LLMs that I simply couldn't do. I do this for my job: I took all the oncall incident reports and had the LLM summarize them looking to ascertain what system (of a pre-chosen list or root cause in the prompt) was actually at fault.\n\nWith minimal effort, I got about 97% accuracy with Hermes Nous 7b. And so now I can attribute incidents to root causes better, without having to go back and read hundreds of reports.\n\nHow would I do this otherwise?",
"sig": "40b524a8fdac35d074f3b23396a88768d9afffc6f4680813510d2a3a3531b97dcfc74c4c7f6eac3c7a83a4ae765622e602babefe4999cc75278f96d9e8b843ad"
}