I Learned C++ Before ChatGPT

Here’s something that my particular generation who finished University before the end of 2021 (those born up to 1999,) really have over the younger zoomers: we graduated “Pre-GPT”. I was part of the last few cohorts to finish University before the rise of LLMs,¹ and therefore one of the final cohorts to complete University on hard mode with no assistance. OpenAI’s text-davinci-002, the first rumble of things to come, wouldn’t appear until March the year after my graduation.

Before language models, cheating was rampant among unsavory kinds and particular groups, who would occasionally all be caught at once in hilarious fashion. Sites like Chegg hosted solution keys and past exams, which helped float students around lazy professors, though did enable the wholesale copying of assignments. Upright and honest students understood the impact to their own learning that taking any of these shortcuts would have, and most of these dishonest students have remained skill-less, untrustable, and dishonest in the workplace.

Here are some of the letter grades I earned at the University of Ottawa. These grades were achieved in a global environment where solutions could not be predicted or generated with a detailed prompt and the click of a mouse. The institution can guarantee these results are mine, a privilege that the students of today have lost:

(By the end of my last semester, my TGPAs were up at 9.5 (at uOttawa A = 9, A+ = 10) despite scraping as low as 4.50 (C = 4, C+ = 5) earlier in my degree. I was an excellent programmer, but certainly not a perfect student!)

Today, an identical list of grades is far less valuable - the mere existence of language models throws into question the fact that they were earned, as these grades could simply have been generated, particularly with online classes.

The introduction of LLMs and resulting loss of academic integrity can be profoundly felt everywhere. Since 2021, students in all faculties have been able to lean on ChatGPT as a crutch through their entire degree, claiming the credential while the language model completes the recitation of the material. It is easier to scrape and cheat through a complex degree than ever before, and as a result, the value of University degrees (even from real STEM fields, and particularly from text-intensive programs,) has been plummeting.

Tough in-person paper exams help to mitigate, but do not stifle the effect that language models have on trust in letter grades. As a people manager at IBM, I can say with certainty that the ability to implement real solutions is decaying, and letter grades and credentials don’t correlate to real world effectiveness. Incompetence is easier to hide, and competence is more difficult to ascertain.

Enhancing the output of technology workers with LLM pair programmers is most advantageous for skilled experts with pre-existing deep knowledge. New hires without existing domain expertise have suffered the most, having a tool at their disposal to delay learning forever.²

If these tools demand expertise, yet the tools can actively circumvent the friction that cultivates expertise, then what is the path for one to become an expert so they can effectively use these tools?

– Lars Faye, “AI Coding will Prevent Expertise”

Ultimately, pandora’s box has been opened, and we can’t go back. I don’t believe we can regain our pre-language-model state of institutional learning (for computer sciences,) and the reputation of academia will continue to suffer as they grapple with a worsening culture and these new technologies. What changes going forward is a heavier weight on real-world performance over credentials and degrees (a shift that was occurring anyways,) and a growing need for networks of trust.

For now, restricting language models to integration in products and document searching seems prudent. I will be instructing my teams and developers to learn without language models, and to “waste” time struggling to comprehend the machine - asking human experts when stuck for guidance. This seems to be the only way to build true competency on solid foundations – therefore I hate every false way.

Learn things the hard way!

Addendum¶

Related articles:

• https://larsfaye.com/articles/ai-coding-will-prevent-expertise
• https://publicznyprofil.github.io/ai_cant_do_your_work/
• https://www.ralston.ac/podcasts/ai-and-the-battle-for-the-soul-with-jonathan-pageau-lecture-whom-does-ai-serve-sacrifice-in-the-digital-age
• https://www.thesymbolicworld.com/content/the-possible-sinister-future-of-artificial-intelligence
• https://boydkane.com/essays/llms-could-control-their-host-machines-by-exploiting-inference-engines

A language model, an unthinking linguistic reiterator, will never feel the God-given human joy of creation in the simple things we do, like baking pies or writing a story.

The ongoing devouring of all knowledge only highlights the need for a personal library, so go build one!