Sergey Brin on Gemini 1.5 Pro (03/02/2024) [video]

Within a couple sentences of each other he says (paraphrasing), “if you test any model (Gemini, ChatGPT, Grok) you’ll see it says far left things” and “we don’t know why [Gemini] is far left leaning”.<p>I think he’s being honest here. At least at the executive level of having no clue how any of this works or why it leans the way it does.

Transcript:<p><a href="https:&#x2F;&#x2F;lifearchitect.ai&#x2F;sergey&#x2F;" rel="nofollow">https:&#x2F;&#x2F;lifearchitect.ai&#x2F;sergey&#x2F;</a>

&gt; The Gemini art? Yeah, with the… Okay. That wasn’t really expected [...] &gt; So, once again, we haven’t fully understood why it means “left” in many cases. And that’s not our intention.<p>Riiiight.

Anyone except me thinks he doesn&#x27;t look very healthy? Its strange he is kind of slow on the video where he enters the room. Maybe some biohacking.

Let&#x27;s be clear. This isn&#x27;t a left or right thing. That&#x27;s just a convenient argument to avoid the real problem. It&#x27;s about moral relativism.<p>I am about as left wing as you can be and gemini is simply offensive. It didn&#x27;t want to say if Hitler was worse than Elon or Trump. It didn&#x27;t want to say if child abuse is bad, and if all child abusers should be prosecuted. It rendered WW2 soldiers as black.<p>This isn&#x27;t left vs right. This is extreme moral relativism to the point of racism. The end point of a line of ethics ML research where people won&#x27;t take a stand that might offend almost anyone.<p>I work in this area of ML. The goal of having a single model that makes everyone happy is mathematically impossible. If you avoid taking that problem on, you end up with a model like this.