Super-Tiny Language Models Controversy

[Eckart Bindewald]

Just posted a short LinkedIn article debunking some claims that "super-tiny" language models have "comparable" performance with respect to much larger models. Turns out if you actually look in the paper the performance of the super-tiny model is at or below random guess probability for several datasets.... 
First I thought this must be the secondary reporting, but now I am looking more at the original authors who published a pre-print with weak results and contradicting strong claims. What is the process of asking for a revised version, does anyone here know? Contact the authors or Arxiv? 
Here is the link to the post: Problems with Super-Tiny Language Models - LinkedIn Article

[Rajorshi]

I think we should reach out to the authors for more clarification on this. On a related note, I'm disappointed that MarkTechPost missed this glaring anomaly

[Aman]

Thanks for sharing, @Eckart Bindewald!

And yes - agree with @Rajorshi that this needs to be brought up with the authors.