Vitalik Buterin
@vitalik.eth
A quick test of the new llama3 models (and old ones): L3 70b: https://i.imgur.com/HPgNLnW.png L3 8b: https://i.imgur.com/HTxZmy9.png Mixtral 8x7b: https://i.imgur.com/qjEk93V.png ChatGPT: https://i.imgur.com/DIQ5loP.png ChatGPT is correct, L3 70b almost; others are wrong.
55 replies
240 recasts
1310 reactions
Hottogo
@hottogo
Love tests like these. Have you seen recent tests on ChatGPT v ChatDKG showing ChatGPT hallucinating without the use of a decentralized RAG in the form of a knowledge graph? Very interesting evolution of these models. https://twitter.com/DrevZiga/status/1755625036486988174?t=AU1KDEXEyeFd43hKHDaUyg&s=19
0 reply
0 recast
0 reaction