All previous tools that were hyped as making programmers more productive had no effect or a positive effect on the demand for computer programmers. I would have thought that we would be in a golden age for young computer nerds as every company on the planet seeks to “add AI”, e.g., “Joe’s Drywall and Paint, now with AI”.l
The Wall Street Journal, however, says that there is a glut of graduates… “Computer-Science Majors Graduate Into a World of Fewer Opportunities”:
Note the hateful depiction of a non-Black non-female not-obviously-2SLGBTQQIA+ computer wizard (NYT would never make this mistake). Also note “Those from top schools can still get job”. In other words, it is the mediocre computer nerds who can’t get hired. Either there has been a huge boom in the number of people who are passionate about computer nerdism or a lot of kids have gone into CS, despite a lack of interest in staring at a screen, because someone told them that it was a sure path to a solid career (this was my experience teaching Information Technology; 90 percent of the students were not even vaguely curious about the subject, e.g., curious enough to search outside of the materials assigned):
My guess is that, due to lack of interest/passion, 70 percent of CS majors shouldn’t have majored in CS and won’t have lasting careers in CS. They are at best mediocre now and will just get worse as they forget what they were supposed to have learned.
Almost all of the news in the article is bad:
To be sure, comp-sci majors from top-tier schools can still get jobs. Pay, projected to be at about $75,000, is at the high end of majors reviewed by the National Association of Colleges and Employers, or NACE. They are just not all going to Facebook or Google.
“Job seekers need to reset their expectations,” said Tim Herbert, chief research officer at CompTIA, a trade group that follows the tech sector. “New grads may need to adjust where they’re willing to work, in some cases what salary, perks or signing bonus they’ll receive, and the type of firm they’ll work for.”
And while big tech companies are hiring for AI-related jobs, Herbert said, many of those positions require more experience than a new grad would have.
Salaries for this year’s graduates in computer science are expected to be 2.7% higher than last year’s, the smallest increase of eight fields reviewed by NACE.
In the past 18 months, job growth has remained flat for software publishers, a group of employers that includes software developers, according to the Labor Department. On the student jobs platform Handshake, the number of full-time jobs recently posted for tech companies is down 30% from the year-ago period.
$75,000/year?!?! That’s $55,000 per year after Joe Biden’s and Gavin Newsom’s shares (online calculator). About $12,000 of that after-tax $55,000 will be consumed paying for the car that is required to get to the job (AAA and CNBC). Salaries are 2.7 percent higher than a year ago? That’s a pay cut if you adjust for the inflation rate in any part of the country where (a) people want to live, and (b) there are jobs.
I’m wondering if the big problem is in bold. Four years of paying tuition should prepare a smart young person for almost any job, including “AI-related” (if not at OpenAI then at some company that is planning to use an LLM via an API to OpenAI or similar). In the late 1990s, colleges weren’t teaching “How to build an Amazon or eBay” (so we developed a class that did and a textbook) even though it was obvious that employers wanted graduates who could built database-backed web sites. Could it be that the CS curriculum is totally stale once again? Very few of the professors have what it would take to get hired at OpenAI and, therefore, they can’t teach the students what it would take to get hired at OpenAI.
I think this confirms my 2018 theory that data science is what young people should study and that data science restores the fun of computer programming that we enjoyed in the pre-bloat days.
Surely the ups and downs in the chart cry out for an explanation?
The opportunity cost of higher education is lower during recessions, and thus demand increases.
Are there official unemployment numbers for software developers and computer programmers?
I do not believe there is an oversupply of skilled computer programmers. In the past years I have not seen a single resume from college grads for skilled software developer contract position that pays much more of $75K/year. Those few developers that were hired out of college for entry level full time perform quite well, but their resume selection is above (or below) my pay grade.
Most new grads require years of work to reach full productivity.
Regardless, the quality of the top 1% has not changed. Those people self-teach, and the degree program they come from is irrelevant.
Almost all AI/ML engineers and researchers we employ fit this profile, with people age 25-45 doing work of similar quality. Some have STEM PhDs or decades of experience, others are five years out of a biology program.
On the plus side for new grads, we have found experience has sharply diminishing returns in the LLM space. It is currently a very empirical field, with most relevant results being less than five years old.
While it’s true that the top 1% of self-motivated individuals can reach high levels of productivity through self-teaching, dismissing the value of formal education is misguided. Degree programs provide a structured foundation in core concepts, theories, and methodologies that are essential for long-term success in any field, including AI/ML.
AI/ML is not just an “empirical field” – it is deeply rooted in mathematics, computer science, and statistics. Formal education equips students with a solid theoretical understanding of these underlying principles, which is crucial for developing robust and scalable solutions. Self-teaching alone may not provide the same depth of knowledge, especially in areas like algorithm analysis, optimization techniques, and machine learning theory.
Degree programs offer a structured learning environment with guidance from experienced faculty, access to cutting-edge resources, and opportunities for collaboration with peers. This structured approach fosters critical thinking, problem-solving skills, and the ability to tackle complex challenges systematically – all of which are invaluable assets in the rapidly evolving field of AI/ML.
AI did a great job! But it forgot to diss the standard CS curriculum as “not AI/ML”!
Also AI did not had enough layers to understand that usually biologists have much more statistics then CS grads whose math requirements consist of remediate HS math, calculus 1 and 2 for dummies, once class in discrete math for dummies and theory of probability for the beginners. VS biologists who need to understand what statistics means for their real life problems.
I worked at a large organization where I hired contractors for govt-related projects. Typically for every 1 mid-level software developer we needed, I’d get 60 (!) resumes forwarded to me from HR to filter + rank + interview. Of those, about 40 would have degrees from foreign schools, mostly here on temporary work visas.
So it’s accurate to say there is a glut of CS grads, but it’s twice as accurate to say there is a glut of *foreign* CS grads. (Why does our immigration policy force citizen job seekers to compete against foreigners when there is an oversupply?)
1) If prevailing wages are too high, then these jobs will simply be outsourced. I would rather have an immigrant or a US citizen making $50k/year as a coder than have the job outsourced all together.
2) Usually children of immigrants who came on H1Bs or for grad programs tend to do well academically so they are also a key source of labor.
Anon: We are informed that H-1B program is not “immigration”. https://www.dol.gov/agencies/whd/immigration/h1b says that only “nonimmigrant aliens” are hired and also that this increased supply won’t reduce wages for Americans: “The law establishes certain standards in order to protect similarly employed U.S. workers from being adversely affected by the employment of the nonimmigrant workers”
Unless the government is lying to us, therefore, there shouldn’t be any children of former H-1B holders here in the U.S because the H-1B holders all went home after their temporary employment ended.
Dr. Greenspun, in practice though, a lot of H1Bs end up converting their status to permanent residency and then go on to raising high achieving kids.
The most useful background for AI is not computer science but statistics, random processes and optimization. The job outlook in Canada for computer science graduates is quite bleak, I would estimate only about 20% will get good jobs, 50% will get mediocre jobs and 30% will be “IT” managers in a basement. Even if you are in the top 10% academically, carriers in law, medicine, finance, accounting or government will be much more financially rewarding. Plus, less competition from immigrants in these fields. If you are not in the top 10%, the trades could be a better option financially.
Must confess to not doing more on AI than training existing models, porting models to different back ends, & massaging the data to solve a very specific problem. Designing a model from scratch, fuggedaboudit. There are much smarter minds devoting entire careers to designing every model for every problem you’d encounter.
At least the lion kingdom got farther with AI than blockchain & VR. At this point, LLM’s have had a year run & the industry is in need of the next blockbuster. If Elon can pick up another existential crisis or Goog can spit out another language, we’re good to go for another year.
Are computer engineering majors generally more on the ball than computer science majors? Perhaps more by selection than by curriculum differences?
The big losers are all the graduates of “coding boot camps.”
BSc Comp Sci = computer science + easy math + humanities electives, 5 courses per term
BSC Comp Engg = computer science + hard math + electrical engineering, 6 course per term
Honest question: if your kid were at University of Maryland for Computer Science, in which would you recommend he minor? Actuarial Mathematics, Data Science, Mathematics, Statistics, or something altogether different like business?
Joe: Definitely not business, in which students attempt to learn about the world of risk from someone with a unionized government job. Actuarial math sounds good if he wants to move to Connecticut and work for an insurance company. Data science might be great, but they could also be assembling teachers at random to catch the latest wave. Math per se has too many irrelevant requirements. So… I think it comes down to statistics or data science and I would choose depending on how well the classes seem to be taught.
https://www-math.umd.edu/undergraduate/math-minors.html has some uninspired language: “The Department of Mathematics offers a Minor in Statistics for students whose majors are not mathematics. The goal of the Minor in Statistics is to provide the student with a substantial number of courses that are statistical in nature and involve a substantial amount of mathematics.” I’d like to think that GPT-4 could have done a better job.
https://academiccatalog.umd.edu/undergraduate/colleges-schools/computer-mathematical-natural-sciences/computer-science/data-science-minor/ says “the Computer Science major Data Science specialization (0701B), or the Computer Science major Machine Learning specialization (0701F)” suggests that the aspiring OpenAIer need not have a formal minor in data science but can simply roll it in with the CS major.
I tutored AP statistics and just couldn’t believe the poor structure of the textbook from a motivational point of view. The students didn’t learn anything about hypothesis testing, the most important traditional use for statistics, until the very end of the school year. (In other words, stats as a field seems to be populated with people who aren’t able to put themselves into the position of a typical student and ask “What would be interesting to this young person?”)
Helpful, thank you. He just finished Honors Stats at the community college and liked it.
As an occasional user of ChatGPT 4, I’ve been rather underwhelmed, Experience has been spotty so far.
1. Taks 1. Trying to convert deeply nested SQL views (up to 9 levels) into a single SQL statement for a bidding model consumption. Seemingly a simple text substitution task. ChatGPT miserably failed producing corrupted output with missing logic pieces.
2. Task 2. Explain required general ledger accounts and describe credit/debit entries to record an HOA special assessment. Passed with flying colors.
Probably because GPT4 was fed only specific well defined entries related to your HOA ledger account so there was little to choose best fitting solution from.
There is no oversupply of software engineers. There is an oversupply of diplomed and certificated clowns who cannot code their way out of a paper bag. It is nearly impossible to hire an actually competent software engineer (by competence standards from 80s). 99% do not understand simplest things about their profession, like the fact that the more marginally relevant crap you glue together the crappier your product gets. A typical “engineer” today does not think twice about dragging in containers, Kubernetes, and endless frameworks — just to fill in some fields on a web page.
This is as relevant as ever: https://blog.codinghorror.com/why-cant-programmers-program/