Back in January, I spent (literally) 10 hours understanding and implementing Longest Common Subsequence. Like many problems that can be solved using dynamic programming, 99% of the effort goes into understanding the algorithm. The code is quite simple:

def longest_common_subsequence(a: str, b: str) -> int:
    """
    Given two strings, `a` and `b`, compute the length of their Longest Common
    Subsequence (LCS).
    """
    m, n = len(a) + 1, len(b) + 1
    dp: list[list[int]] = [[0 for j in range(n)] for i in range(m)]

    for i in range(1, m):
        for j in range(1, n):
            if a[i - 1] == b[j - 1]:
                dp[i][j] = 1 + dp[i - 1][j - 1]
            else:
                dp[i][j] = max(dp[i - 1][j], dp[i][j - 1])

    return dp[i][j]

Fast-forward 8 months. While interviewing at a company¹, I got a technical take home test through CoderPad. The gist of the problem was to find a subsequence of string a containing string b. Now, CoderPad boasts on their website:

AI broke technical hiring.
We fixed it.

And they are not wrong. I had access to chat-based LLMs for the assessment, albeit weaker ones². The presumed goal of allowing a candidate to use an LLM during a coding interview is to remove the unfair advantage they provide. Instead of going on faith that a candidate will not use AI when asked politely, all candidates are given access to AI. This is where my thesis, namely, knowing stuff is still important in the age of AI, garners some supporting evidence.

If you were to dump the problem into the chat and say, “How would you solve this?” you would get back a rather straight-forward, unimaginative, contrived solution³. Without going into the details of the actual problem, an optimal solution is to use the Longest Common Subsequence algorithm, which is a novel, more creative way to solve the problem⁴.

Though a specific example, I believe this scenario applies more generally to most problems where the solution is inspired, rather than rote. In addition, it evidences a theory around AI that I have been mulling for the past couple months:

Having knowledge at your fingertips is still, and will continue to be, extremely important.

There seems to be a pernicious, Mephistophelian bargain developing in the tech industry where even very smart people are slowly allowing LLMs to do more of their thinking and problem solving⁵. This has the side effect of slowly eroding their skills and deep understanding of the problem space. It also causes their solutions to devolve toward the mean solution. Short-term productivity traded for long-term understanding.

Ditto to the fact that people are not going to wait around for you to input their question into an LLM before responding. In the normal course of dialog, even discrete, under-the-table prompting kind of just kills the vibe of the conversation.

I’ve personally wrestled with this new way of working, mourning the apparent death of the inspired for the fast and good-enough. From this, I’ve developed a simple framework, which I apply to my use of LLMs:

                             +-------------------------+
                             | Do I know how to do it? |
                             +-------------------------+
                                 /                 \
                             +-----+             +----+
                             | Yes |             | No |
                             +-----+             +----+
                                |                   |
             +--------------------+   +---------------------------------+
             | Have an LLM do it. |   | Learn how to do it with an LLM. |
             +--------------------+   +---------------------------------+

Striking a balance between generation and inspiration, I am able to be wildly productive, while still being able to, you know, code and problem solve. As one might expect, I still am very bullish on deeply understanding things, not only for its utility, but that it generally feels good to acquire knowledge.

All in all, I’m actually quite worried about the future of society. Why, when even very smart people can be duped into trading their skill for productivity would the general public not as well? I feel like this is a textbook application of Matthew 13:12, where those that are able to navigate the siren song will come out wildly ahead, while those who do not will wallow in slop. Maybe all the Sci-Fi cliches will come to pass and the human race will realize that to be human is to think and if all of our thinking is outsourced to machines we become less human. It follows then that we will choose to destroy technology in a fit of humanistic frenzy, since AI will be viewed as a source of existential competition. It does seem rather in keeping with our inherent drive for survival.

Anyway, closing the loop on the titular thread of this post, I progressed through the technical take home test in the interview process with a Move forward with high confidence, which I believed was crucial in receiving an offer. Had I not known LCS, I probably wouldn’t have moved forward, thus solidifying my belief that knowing stuff is still important in the age of AI.