Proud to have contributed to this work, congrats @joelniklaus
#SOTATV just interviewed @joelniklaus on his @COLM_conf Outstanding Award Paper, 𝐡𝐨𝐰 𝐜𝐚𝐧 𝐰𝐞 𝐬𝐲𝐧𝐭𝐡𝐞𝐬𝐢𝐳𝐞 𝐡𝐢𝐠𝐡 𝐪𝐮𝐚𝐥𝐢𝐭𝐲 𝐩𝐫𝐞𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐝𝐚𝐭𝐚? over 1T tokens rephrased, 90 experiments, 1.9 B Documents.
His hot take after all that: synthetic
