>>74290010Alex seems to have no experience in AIs at all, only dabbling in Unity mods and webdev, whilst Vedal has quite a few previous AI projects. Out of all Alex's 85 starred repos, only 1 was AI-related.
In addition, recall how Vedal assured that Alex that NNs will work better with relative positions rather than absolute positions
(during that call about the among us mod a long time ago).
This suggests that Vedal is the only one with AI experience, of the two. Alex probably doesn't do the Neuro's backend AI stuff at all, and instead, at most, helps out on the Unity frontend of Neuro.
If Vedal is being helped at all (AI-wise), it's probably someone else (who for some reason is very private).
https://web.archive.org/web/20230309141700/https://github.com/Vedal987?tab=repositorieshttps://github.com/Alexejhero/>>74290926I think Vedal knew more about LLMs than you are suggesting, though to be fair, I don't even know pytorch, so I could be misjudging. During that session, Vedal was confused about where Karpathy was going and later had revelations. This implies that Vedal already knew what a correct LLM training/inference program would look like, and could reason about whether Karpathy's example was like that. Someone who knew nothing of the training algorithm for LLMs would simply copy the code, line by line, without doubt or confusion.
