Andrej Karpathy: "I was given early access to Grok 3 earlier today"

If I had to explain current-day LLMs to someone from 2010, I&#x27;d use this paragraph as my opening quote:<p>&quot;Grok 3 knows there are 3 &quot;r&quot; in &quot;strawberry&quot;, but then it also told me there are only 3 &quot;L&quot; in LOLLAPALOOZA. Turning on Thinking solves this.&quot;

In conclusion: <i>&quot;For now, big congrats to the xAI team, they clearly have huge velocity and momentum and I am excited to add Grok 3 to my &quot;LLM council&quot; and hear what it thinks going forward.&quot;</i>

Grok has an advantage in its access to Twitter data.<p>I imagine soon you&#x27;ll be able to ask it what the world is talking about today and get some interesting responses.

Related ongoing thread:<p><i>Grok3 Launch [video]</i> - <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=43085957">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=43085957</a> - Feb 2025 (985 comments)

&gt; Model still appears to be just a bit too overly sensitive to &quot;complex ethical issues&quot;, e.g. generated a 1 page essay basically refusing to answer whether it might be ethically justifiable to misgender someone if it meant saving 1 million people from dying.<p>The real &quot;mind virus&quot; is actually these idiotic trolley problems. Maybe if an LLM wanted to be helpful it should tell you this is a stupid question.

What is the &quot;emoji hidden message&quot; meant to be testing? This went around about a couple of weeks ago and it&#x27;s an interesting bug&#x2F;vuln, I suppose, but why do we care if an LLM catches it?

I wonder how much stock people put into people like Andrej&#x27;s opinion on an Elon Musk project? I would imagine the overwhelming thing hanging over this is &quot;If I say something that annoys that man, he is going to call me a pedophile, direct millions of anonymous people to attack me and more than likely will attempt to fuck with my job via my bosses&quot;.<p>Let&#x27;s say the model is mediocre. Do you think Karpathy <i>could</i> come out on X and say &quot;this model sucks&quot;? Or do you think that even if it sucks people are going to come out and say positive things because they don&#x27;t want the blow back?

&gt; Model still appears to be just a bit too overly sensitive to &quot;complex ethical issues&quot;, e.g. generated a 1 page essay basically refusing to answer whether it might be ethically justifiable to misgender someone if it meant saving 1 million people from dying.<p>I think the models response is actually the morally and intellectually correct thing to do here.