A Crisis for Newly Minted CompSci Majors -- entry level jobs gone

[elias001]

@TensorCalculus, @jedishrfu, @robphy i don't know if you folks have seen the two youtube videos below. It has to do with something call "prompt engineering" and AI as being a compiler. The second one by the same guy is talking about educational usage for AI. The second video is kind of vague while the first one, I am sure is important behind his massive technical word salad with python as its main ingredient. I don't have the technical expertise to be the equivalent of a food critic. Wait, even if I do, I would not want to sample any of it, is in python=super gross.

 

[robphy]

Dev vs. Tester: Guess who’s got the edge in this epic meme showdown? (codemug)

 

[symbolipoint]

I do not know exactly what comments to say about this article but it seems discouraging. I will simply post a link, and you all think and make what comments you believe are good.

https://www.yahoo.com/news/articles/leading-computer-science-professor-says-095502798.html

 

[FactChecker]

1) If you want to get a job done efficiently with computers, the first question is: "What is the job?". It is not which computer language, technique, or fad of the year to use. A specialist in the application has a real advantage over a specialist in computer science. Of course, both are important, but the application field often remains long after the current computer fad is long gone.
2) A lot of the advanced computer techniques are made available in program libraries and work environments for anyone to use.

 

[harborsparrow]

Re: the miserable employment situation in IT right now

A painful and probably necessary course correction (social scientists call it a "disruption") is going on right now in software development. IMO it is only partly due to AI. There has been a glut of people with newly minted CS degrees for some time now in the US; they always had issues getting jobs due to the fact that a large corporation such as Amazon can hire a person from a foreign country via H1-B visa. The H1-B folks had to remain employed continuously for 5 years to remain in the country and get a green card, and many were thus willing to work for far lower wages than most US citizens expected. Amazon currently employs about 10,000 people via H1-B visas, and furthermore, those visa people will put up with a lot more crap than someone who can afford to change jobs. The party line is that these corporations cannot find US citizens with appropriate skills. Personally, I don't think they tried; I think it's all about lower salaries and leverage over their workforce.

But the big corporations mainly recruited from the ivy league and high reputation schools; if you got a degree from a perfectly good tech school or regional university, you might not ever even have gotten a shot at applying to the big tech companies.

And now, newly graduated CS students are well and truely screwed. Further, companies are in a panic due to AI and the Trump admin's wildly unpredictable policies, and IMO IT management always suffered from extreme short-sightedness anyway. So right now, the only IT people being hired by corporations are steeped in LLM skills.

Re: LLM's. Those have scraped and compiled almost everything they could find on the WWW, including the crap as well as the cream, so a certain portion of what they regurgitate (as glorified search engines with the ability to summarize) is going to be wrong just because the sources they have scraped (mostly illegally) are wrong. Further, the LLM's bots have ceased honoring robots.txt files and in many recent cases, they scrape so aggressively that their finding a new site to scrape leaves that site in a state as if it were experiencing a deliberate denial of service attack. I have experienced this myself, and desperately casting about, I was able to "save" my website by placing it's DNS behind a Cloudflare screen. Cloudflare parses all incoming requests and discards all those believed to be coming from misbehaving bots and other bad actors.

The market for IT professionals WILL rebound, because it has to, but I am not sure how long it will be. I have advised my young friends who are currently unable to find work to try using Freelancer.com, a place where normal people (not necessarily huge corporations) can find and hire programmers, sysadmins, project managers and many other roles as needed. I have found Freelancer very useful; those who hire leave reviews of those hired, and those hired leave reviews of the hirers. So it is reputation based, and frankly, the profession could benefit greatly from a two-way reputation based hiring system instead of the top-heavy management abusive situation in which programmers have largely labored in recent years. At least, this is how I view things. Of course, Freelancer skims off about 15% for its matchmaking, and workers do not receive any benefits such as health insurance, so they are purely independent contractors, and US law has never favored their situation financially. But 15% is actually reasonable. Most large companies require all independent contractors (and they hired a LOT of them in the past) to come through a few contract houses, and those contracting companies (acting mainly as payroll for the workers and sheltering the company from liability) generally skimmed off much greater percentages, sometimes 50% or more. I contracted for Wyeth for a couple of years back in the day, being paid $110 per hour, and I think Wyeth was paying them about twice that for my time.

 

[symbolipoint]

#55, a very long post, but interesting, and to some extent informative.

I ask, what is the meaning for the abbreviation "LLM"?

 

[harborsparrow]

LLM, Large Language Model, an AI machine learning system that is trained by "scraping" (fetching, parsing, storing) publicly available websites and databases, and sometimes, illegally obtained information. It can parse natural language, possibly also understand human speech, and tries to "understand" and represent the world symbolically based on its " reading". They are used in place of traditional search engines based on the "knowledge" they have been trained on, and are beginning to be used as assistants for other purposes. I have begun to encounter annoying phone menu systems (Hyatt has one, for example) that seem to be AI based but are not ready for primetime in case of any out of the ordinary situation. It is difficult to get past them to a human being.

I have heard for years now that resumes submitted to many large corporations are screened by AI and that system can be gamed, often must be gamed, because if you can't get the resume past the LLM, no human being will ever see it and there is thus no chance of being interviewed.

 

[harborsparrow]

Here is an interesting article about Elon Musk gambling on LLMs being the future:

https://www.wsj.com/tech/elon-musk-...ower-dec4c70d?mod=djemCybersecruityPro&tpl=cs

Note that LLMs run on server farms of PCs whose powerful graphics processors are used in clustered form to create a supercomputer. Yes, NVidia chips, not the x86 or maybe ARM based processors once considered to be the heart of a PC. These server farms consume enormous power and are straining electrical companies. Public utility rates for everyone are rising in order build power plants to be able provide enough power to installations for these AI server farms. The early efforts towards server farms were sometimes called COWs (Clusters of Workstations) or NOWs (Networks of Workstations)--like SETI--and they use sophisticated distributing computing algorithms to dole the computational work out to possibly millions of processors. This work progressed for years before OpenAI competitively forced all the major search engines to convert to having AI capability.

So yes, oddly, these issues are related peripherally to the crisis in software jobs.

This is one reason people are reconsidering nuclear power options, as described in https://citizendium.org/wiki/Nuclear_power_reconsidered

BTW, side note, you can turn off the LLM usage in Google searches by placing -ai after the search terms. This actually skips the AI summary and prevents the activation of an LLM for it, using less power. But Google already has its search engine running on enormous server farms and was already using AI behind the scenes to minimize power consumption by its server farms. Google farms use solar arrays and close down after the sun goes down. So when you Google something after dark, the search is likely serviced by a farm on the other side of the world where there is sunlight. Most LLM companies are not at that level of sophistication yet; their farms just run 24x7 and suck power from the public grid constantly.