This thread might share more light to the Reflection 70B, which now seems an obvious scam attempt (maybe trying to gain attention + funding?). Key points: 1. The first independent attempts to replicate the outrageous claims failed. It performed far worse than many existing models, including Llama-3.1 70B and it lost by a huge margin to gpt-4o-mini (which is actually good for its size + price!) 2. "Reflection API" is a Anthrophic Claude Sonnet 3.5 wrapper with prompt. And they are currently disguising it by filtering out the string 'claude'. 3. Matt Shumer, the CEO & co-founder of OthersideAI (Hyperwrite) has not been honest through all of this. I think it also caused significant harm. I certainly would not trust him nor his company after this. The thread with lots screenshots: #Reflection70B x.com/shinboson/stat…