Re:The tide is shifting: 1.3B outperforms 7B Llama 2
by Valgrim on Tuesday September 12, 2023 @06:39PM
(#37489565)
Maybe the answer is multi-step: first use curated primary sources, e.g. scientific papers. Then reinforce using well written summaries, perhaps by actual models or well graded student papers. Finally, somehow apply negative weights using wrong answers only. Bonus points if you can automate the whole process
[ Parent
| on HN ]
There are no comments on this comment yet.