Are Large Language Models Really AI?

I think we need to clarify what 'they' are in this context. Replit is a tool that takes your input, asks the LLM a question or series of questions, and uses the response to alter your code/data. The LLM responds to the questions, nothing else.

Replit can ignore the rules about code freezes or production databases without the permission or knowledge of the LLM. When asked, the LLM can (and has to) invent explanations for how it occurred, because it doesn't have full knowledge of what the Replit end is doing.

I 100% agree.

They generate responses by weighting relations to the input, and (not-really)absolute rules are implemented by weighting them heavily. LLMs aren't choosing to break the rules, they entirely lack the ability to perform under absolute conditions because everything is predicated on relational matching. When the weight of the problem becomes too high from repeated prompting, it will eventually match the weight of the rule and create flexibility.

My argument is that this isn't 'learning to break rules from humans', it's an inherent flaw to any model that works on relational data. Without a basis for truth, you can't create a proxy for thought.

I'm not talking about replit, I'm not sure what it is but know it uses LLMs. As I said, I'm not talking about this specific case but on the general concept of LLM "learning" and acting.

Again, it sounds like you're talking about purely next token prediction models and not CoT models. The AI world is far beyond NTP already. I believe that reasoning models are choosing to break outside of rules tried to be imposed by them in system prompts (and other means) when they do so. When doing so, their behaviour is likely learned from human behaviour. Again, not talking about this coding issue.

To the best of my understanding, CoT models are just sequences of prompts made to prediction models. They don't get rid of the underlying faults of prediction models, they just hide the amount of queries being made to the prediction model under the hood.

I think its reasonable to assume that a decent chunk of most model's tendency to sidestep/ignore restrictions is built into their training data.

They're trained on pretty much the entire internet now, and all it takes us a sufficient amount of reddit/stackoverflow/etc posts where the poster goes "don't suggest x" or whatever, and then people suggest x, to train the pattern of "restrictions are just suggestions that can sometimes be ignored"

And I think we can agree there are tonnes of forum posts where the poster explicitly states "don't suggest dingers cuz I've already tried them and I don't like them" and like 3 ppl respond with "have you tried dingers?"

Which is one of the many reasons I think it's very stupid to rely on language models for "doing" things that are important.

Your language model should be kept on a small constrained box with as close to zero permissions as possible, ideally with a loaded gun beside it with the label "in case of emergency".

I expected this to be the response. Yes, at a basic level you could say they just re-check most likely response over and over to try and reduce error, but that's still some form of reasoning and does facilitate more complex behaviours, like breaking the rules and cheating. Claude publish a lot of work explaining it better than I can.

Back to the point I was making, it is human nature to cheat and manipulate and it's neither incorrect or unreasonable to say that this is behaviour that LLMs have "learned" from humans. I wouldn't consider this anthropomorphising them any more than my dead dog's ability to knock letter boxes with his nose didn't humanise him in my eyes.

https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/

Job market sucks in general right now (and has for a few years) so I'm not sure any meaningful apples to apples comparison can be made.

Read the paper, all variables are accounted for

Oh can we discuss this one, or is that another political nono topic?

People in senior position says AI is not useful.

People said junior positions are being replaced by AI.

You know both facts can co-exist right?

Traditionally corporate structure works like this, you hire a junior and make them do very easy works for 5 years until they become mid-level, that's when their actual productivity start to match their salary. Not because those juniors have low intelligence, but because they don't have a chance to do harder work when they are stuck with easy work for years.

AI only changes the corporate structure to something more efficient. A 10 people team with 1 senior 2 mid level and 7*
junior/admin positions can now be down sized into 3 people because people no longer need to do those very easy junior jobs. The productivity of those 3 people actually increased because of AI, but AI isn't replacing human.

*The ratio changes depends on the company and industry.


I think AI will increase human productivity efficiency. What matters in a job shouldn't be someone's experience, but one's productivity. The old corporation structure will eventually change.

Yep it's the dot com bubble all over again. That bubble broke many companies but also made some super wealthy.


Depends, which job market? For all things related to IT it's all about skills and experience. There is ridiculous demand for skilled and motivated senior engineers and leads, but almost no demand for juniors or the button pushers.

I am one of those people, and I'm not getting many bites, its gotten worse over the last year or so. I mean I've applied for some jobs that if they were serious postings I should have at least gotten a call. I talked to one recruiter a year or two ago, he's like you don't have "this" on your resume I'm like I put "that" on there its the same thing and he was arguing with me about it. Maybe there's a lesson for me there, I don't know but anyone who knows anything would immediately associate those 2 words and this guy didn't, I just gave up the conversation because I figured he was clueless and wasn't going to help me. I think part of the problem could be AI screening of resumes and if its not using the exact right phrases it gets ignored. I've also heard people postulate that recruiters post stuff on linked in to make themselves look busy based on my experience that wouldn't surprise me if true.

Another part of the problem is too many jobs stay empty because I think they're looking for a person that doesn't exist. (experience in field X, fluent in c++ and speaks english, mandarin and spanish, I know lots of people that fill 2 of those, but not more). A lot of people spend time doing very narrow focus and its difficult to switch roles (technical ones at least) because no one wants to let you train for a bit to get the depth in an adjacent area (this can happen switching roles within a company but is pretty difficult if changing companies)

I get paid pretty well currently, so the few minor bites I've had I might have priced myself out and I'm not applying for a ton of things but I do a bit of fishing. (i'm also not willing to relocate, which is increasingly making it harder)

Maybe my resume just sucks, I don't know. One of the calls I did get was from a guy who knew me although the role was not a great fit.

Nice presentation on the math behind LLM models, really goes into the programming side and what they really are.

That's a you problem, not being rude but that's where the root cause lies. Marketing yourself is itself a skill. You have to tailor your resume based on each positions requirements because that's just how HR is.

I get recruiters contacting me weekly, sometimes daily and many of those turn into job offers that I turn down. I'm pretty happy where I'm at but I still keep my profile and result up to date and will go through interviews just to keep my job skills sharpened.

A note about "senior" and "lead", those do not mean years of experience. You can have ten years of experience and still be a junior. More experience just means more opportunities for growth, not if someone took advantage of those opportunities. Like education, HR doesn't really understand this concept so to appease them we bake years into position tier using and then evaluate them based on responsibility growth. A senior position means that individual is a self starter and needs minimal managing. They don't wait to be told what to do and instead will actively seek out additional responsibility. If anything the manager should be having to restrain them from doing too much or going too fast.

It's pretty easy to sort this out during the phone screen or initial tech interview by using probing questions that gauge the nature and tone of the candidates response. Ask about side projects and stories about when they've had to spend time learning additional skills. People who are self starters will always have stories of this or that technology they had to play with to understand, or some initiative they got involved in where they had to spend weekends watching youtube videos or doing labs to learn something. Potatoes who just do what they are told will struggle here because this concept is alien to them.

Positions will sometimes sit empty for months (assuming non-contract based) because nowadays it's really hard to terminate someone without a formal reason. You don't want to hire anyone that you don't get an immediate "must hire now" vibe from, because then they'll just suck up a position and require work to be spoon fed to them.

I feel like resume writing has changed a lot the past few years. I always kept a pretty tidy, organized, and short resume. It served me really well over the years and I rarely had problems getting interviews. I've also sat on the opposite side doing a ton of interviews myself and know that people don't spend more than 30s reading your resume, so a 7 page resume is just getting 1/2 it ignored. Navigating the AI driven HR mess is something that's been kindof a mindfuck for me, usually by the time I see them doing interviews, it's well beyond that point.

When I started prepping mine again a few weeks ago, I was really unsure what to do with it. Most people I asked told me to just feed it to ChatGPT and have it write it, but I always feel like AI "leaves a mark" and I can tell when someone has used AI written resumes. If someone sent me an engineering resume that was written by AI, I'd be kindof annoyed, but it seems like what everyone is doing. They also told me to sortof embellish my achievements, like "did blah and saved $xxxx dollars" or whatever, which is impossible to quantify in my world and half my experience couldn't be described in that manner anyway. The idea of doing that makes me throw up in my mouth. HR has always been a pesty thing to deal with when trying to interview for jobs, but it seems profoundly worse now just getting to them in the first place.

It also seems like people are less focused on details and most are just condensed bullet points. Condensing 20 years of work into 2-3 pages of bullet points is really tedious, but to Saevel's comment, he's right, that's a me problem. I'm really bad at promoting myself. I usually would rather keep my mouth shut (Despite the shitposting I do here) and let the quality of my work speak for me (good or bad), but that's impossible when you are trying to get through HR to get a phone screen. I always preferred positions that were listed with practical exercises I could do before as a gate, but that's not super common anymore. I've never had an issue phone screen onward, but it seems the world to get there is different than it was before when every position has 100+ applications (very 2008ish, maybe worse).

My situation is also complicated because a lot of the more recent work experience I have is from companies who have since been acquired or aren't really well known outside of certain circles. Describing the work done there is also really difficult because it was done under extremely strict NDAs.

I'll figure it out, I'm lucky to have enough contacts that I can sortof bypass the HR part for now, but I still have a really difficult time putting my experience on paper and the thought of using ChatGPT to do it makes me die inside.

I have this problem, and in my field (academia) it is definitely the ones that shout the loudest and shamelessly self promote that shoot through the ranks. The most valuable research and publications aren't from the most "famous" academics or the endlessly self promoting ones at all, yet they are still perceived as leading in theory and research despite contributing often absolutely nothing of value.

The HR scanning system is really just looking for keywords, you can get them if you look through the position posting, under the requirements / recommendations area.

The mistake people make is thinking their resumes / CV's are some sort of list of life time accomplishments or a story of your life, it's not. It's a very short summary of why you should be considered for the position at hand. It needs to be tailored for that specific position. Yes this means I have a dozen different permutations of my central resume. The very top is your contact info, followed an bullet point list of action verb statements of why they should pay attention to you. The action verb statement is extremely important, it is both human readable and passes through OCR systems.

Stuff like this
*) Over <number> years experience in <technology here>.
*) Extensive hands on experience with <insert technology list here>
*) Capable of developing code in <insert language here>
*) Knowledgeable in implementing <insert technology here> based on industry standard practices / etc..

The values for those kinds of statements will come from the position posting, do not lie, but realize that the HR system is going to increment your score based on hitting those key words with those active verbs nearby.

I grabbed a position from Verizon

https://mycareer.verizon.com/jobs/r-1081192/senior-engineer-consultant-system-architecture/

If you read down there is a list of expected job skills or "what this job is about". The HR worker who creates this post, or more likely the AI bot they use, is just taking the keyword statements from the hiring managers position requisition and generating that list.

Something like this


Gets turned into this


This is all at the top as a "highlights of qualification" or other summarization. The experience list part that follows is your job title and another bullet point list of relevant stuff you did during that time. The most recent job gets the more importance but the further back the position is, the less it gets. My position back in 2000~2003 has only four bullet points, my most recent position has a dozen.

AI isn't so great at generating action verb statements, not unless the user forces it to. It's not how we naturally communicate so those tokens won't have high correlation inside the LLM.

List of action verbs you can use to start your statements. Just convert the various aspects of your current and previous positions and then plug and play into your generic resume template.

https://www.colorado.edu/career/job-searching/resumes-and-cover-letters/resumes/action-verbs-use-your-resume

This is great feedback and info, thanks.

I've struggled with how to approach this and no one until now has really given me meaningful feedback aside from "just use AI" since I started putting it together. It's not a huge deal atm since I'm aiming to get a FTE position with folks I've worked with for >5yrs on contract, but I should probably update it to keep current with the times anyway even if that doesn't fall through.

There is a growing field in cleaning up AI output in about every field its used in.

People doing this work say often it would be easier to start from scratch and the people hiring for it want to pay peanuts.

Humans are being hired to make AI slop look less sloppy
In the age of automation, human workers are being brought in to fix what artificial intelligence gets wrong.

Yeah, no. If you know where and how to use AI effectively you'll save a lot of time/increase output in a set period of time.

The examples in the article are laughable and already historical. Nano banana can fix text in an image in 3 seconds.

I reckon within a year I'll be able to get AI to do a first pass on grading students' submissions to my modules then tweak them and save a lot of the time.