4 ms·
So we can validate Bard using blogspam SEO articles, eventually having those written by Bard or ChatGPT? It's a LLM ouroboros
by candiddevmike 3y ago
So we can validate Bard using blogspam SEO articles, eventually having those written by Bard or ChatGPT? It's a LLM ouroboros
- vitiral 3y ago* the snake eating it's own tail This feedback loop, when you extend LLM to the horizon, is my primary point against this approach. When 90% of the new training data is from content (a worse version of) it previously generated you get a negative feedback loop to zero quality.