I think this is an interesting direction, but I think that step 2 of this would be to formulate some conjectures about the geometry of other LLMs, or testable hypotheses about how information flows wrt character counting. Even checking some intermediate training weights of Haiku would be interesting, so they’d still be working off of the same architecture.The biology metaphor they make is interesting, because I think a biologist would be the first to tell you that you need more than one datapoint.

That makes sense; however it does not seem like they check the LLM outputs against the known solution. Maybe I missed that in the article.

They study it because it already has a known solution.The point is to see how LLMs implement algorithms internally, starting with this simple easily understood algorithm.

it&#x27;s not strictly a counting task, the LLM sees same-sized-tokens, but a token corresponds to a variable number of characters (which is not directly fed into the model)like the difference between Unicode code-points and UTF-8 bytes, you can&#x27;t just count UTF-8 bytes to know how many code-points you have

There&#x27;s also a lot of analogising of this to visual&#x2F;spatial reasoning, even to the point of talking about &quot;visual illusions&quot;, when its clearly a counting task as the title says.It makes it tedious to figure out what they actually did (which sounds interesting) when it&#x27;s couched in such terms and presented in such an LLMified style.

&gt; The task we study is linebreaking in fixed-width text.I wonder why they focused specifically on a task that is already solved algorithmically. The paper does not seem to address this, and the references do not include any mentions of non-LLM approaches to the line-breaking problem.

A superior LLM for line length optimization:<a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=Y65FRxE7uMc" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=Y65FRxE7uMc</a>

The analogy stems from the notion that neural nets are &quot;grown&quot; rather than &quot;engineered&quot;. Chris Olah has an old, but good post with some specific examples: <a href="https:&#x2F;&#x2F;colah.github.io&#x2F;notes&#x2F;bio-analogies&#x2F;" rel="nofollow">https:&#x2F;&#x2F;colah.github.io&#x2F;notes&#x2F;bio-analogies&#x2F;</a>

&quot;not designed by humans&quot;? Since when? Unless you count cortical organoids &#x2F;wetware (grown in some instrumented petri dish) every artificial neural network, doesn&#x27;t matter how complicated, it is designed by humans. With equations and rules designed by humans. Backpropagation, optimization algorithms, genetic selections etc... all designed by humans.There is no biology here, and there are so many other words that describe perfectly what they are doing here, without twisting the meaning of another word.

Sure, but it makes no sense at all if you define biology as “the smell of a freshly opened can of tennis balls.” The original comment is probably better understood using a standard definition of the words it used, rather than either of our definitions.

It makes sense if you define &quot;biology&quot; as &quot;incredibly complicated system not designed by humans that we kind of poke at to try to understand it.&quot;

Utter disrespect for using the term &quot;biology&quot; relating to LLM. No one would call the analysis of a mechanical engine &quot;car biology&quot;.
It&#x27;s an artificial system, call it system analysis.

When models manipulate manifolds: The geometry of a counting task