tedbradley

My good sir or madam, disproving a decades-old conjecture produced by Erdos that has had armies of people in that field have their go at it IS a novel mathematical result.

Sure, but the bugs were found in an automated process. They just let an LLM scan. That's very impressive finding 100s of needed code changes. And it's even better if those needed code changes are bugs / vulnerabilities. The part no one is talking about comes from the bill. I'm sure Anthropic let Mythos analyze possibly for US$10,000s in tokens. A similar phenomenon happened back when an LLM scored well on some math olympiad competition. Yeah, it got all the answers right, but it was a frontier model running for 8 hours straight. That'll hurt the budget quite a bit. We're likely not at a stage where big corporate systems can just throw Mythos at it willy nilly for a complete analysis unless they have a ton of money.

I answered your question. AI-assited programmers will be paid less since more can do the same job through the use of AI assistance. In some cases, they will even can some developers if productivity goes up enough, and the team can reach business objectives with fewer people than pre-AI times. At the same time, these coders will become more and more dependent on AI, and as I'm sure you know, the API is priced so that they make more money than they lose per request. More and more usage = more and more revenue.

This pattern is only going to become more extreme year after year. I used to reject the idea that LLMs could produce useful code or debug things, but these days, we have Claude Opus and chatGPT Codex. And just around the corner, there's Claude Mythos. I believe it's ready to go out, but they are scanning OSS to give the code underneath it all a head start to fix the types of security issues Mythos can find before releasing the product. Otherwise, we could be talking an LLM jailbreak into a "scan this popular Java logging library" or "this popular OS operating system, Linux, for security flaws." If they didn't do it this way, there could have been a lot of damage to PCs, companies, government, bureaucracies, and institutions in general.

If I were to steelman your position, for now, people need to be a good dev to make the most out of the LLM ecosystem, and the skill of prompt engineering varies person to person as well. I could see an exceptional dev outputting not just more than they used to but with their improvement relative to themselves being much higher than the average improvement other devs gained. In that scenario, yeah, salaries could still increase despite the role being AI-assisted and despite the LLM tools costing these devs' company money every query. Skill varies anywhere between "vibe coder" all the way up to the highest position that still codes at your company, and familiarity with how to leverage LLMs the best can vary that widely as well.

Right now, the name of the game is making sure your LLM has a good action plan before letting it attempt to fix a bug or refactor or add a feature. Devs with more experience know what to ask Claude to do whereas a greener dev doesn't know the questions to ask, leaving Claude to guess right sometimes and wrong sometimes. Simply put, if a dev doesn't know to ask for something, there's a bigger chance the LLM won't care to do it. And if there's some nuanced, tricky aspect to the code not described to the LLM, the LLM might burn a lot of tokens to reach a bad solution. A good dev might give more clues and hunches and more context to fine-tune the prompt so that it almost definitely succeeds whereas a greener dev doesn't have intimacy with the system yet, needing tips and descriptions of subsystems themselves before they could pass it along to Claude. By this stage, people also differ in their skills with the various tools in the ecosystem. Power tools do a lot more in the hands of a seasoned handyman than in the hands of an eight-year-old after all.

However, the better LLMs get, the less differences like this will exist, and ideally, every dev will approach a similar amount of productivity. Salaries aren't reflective of how much profit a worker produces in the company (unless you are the CEO or a little below them or maybe have some stock). Supply and demand drive salary. If something nearby AGI arrives tomorrow, by definition, almost any two devs will provide similar value at which point teams will downsize, yet productivity will hold steady or increase. They will downsize to save some money since we live in a brutal world where workers have no loyalty to a company, and a company has no loyalty to its workers. Pensions are a relic from the past. After all that will happen, a large group of qualified devs will be searching for jobs, so they can remain in a home with food in it. Companies will see tons of resumes flowing in all by AI-assisted devs that can do the job.

The companies will then do two things: Offer the hired devs a transfer to AI-assisted dev for less money or else while also interviewing all the ones that all the companies fired, giving them that same AI-assisted dev salary to everyone in the picture. And they will have calculated the proper discount off the old full salary using some kind of economic equations. Then the wildcard happens: some of the ones needing work urgently start to offer their services for even less than the company is. It's a spiral downward until the salary becomes so low to the point where a dev would rather be an ex-dev doing something else that is more relaxing and also still paying them enough money to survive. No reason to do tough coding work, it'll still be tedious with stronger LLMs. Comfort and relaxation will prevail for many as they no longer feel the salary justifies doing the work.

And at the top, assuming AI costs do not exponentiate, they will be making more money than ever before since they downsized teams, slashed salaries, and got hired at even lower salaries than the slashed salaries. (There will still be a premium for knowing the systems like the back of your hand without need to ramp up before adding value, so you'll get paid more than a new hire.)

What would you consider the "extreme of tracking?" Anyway, I feel like we're dealing with Sorites paradox here, which is something my brother always freaks out over whenever there is a continuum of choices in something like a political argument. E.g. "Tax the rich" → "What is rich?!?" When someone says, "Tax the rich," they generally mean people with dozens of millions in net worth growing fast all the way up to the people who have hundreds of billions. I don't understand why that conversation repeatedly comes up with him. He's also into the slippery slope where, if you like the idea of a wealth tax, he cries bloody murder since income tax started out only on the richest salaries in the nation. So he reasons that... everyone will eventually have a wealth tax instead of it just being the people with billions of dollars who have paid like 1% tax per year due to their net worth being tied up in unsold stocks.

So if I'm going to pick your brain, what would a realistic extreme of tracking look like? You have to log in with your state-issued identity to enter the internet, and systems track absolutely everything you do? Sure, that sounds bad. I can admit that. But I don't feel like having a "?=example.com" is anywhere near that, if you get what I mean.

Do you find it moral to block ads? By that point, you are using free services without paying as intended. Or do you mean you buy YT premium and Twitch Turbo and Spotify premium and all those monthly bills that both block ads while sustaining the services you apparently enjoy using?

Glad you asked. AI empowers people who couldn't do a job before to do a job. With more supply of qualified workers, these workers compete with each other by lowering the salary they'll take.

So:

* You get paid less. * The company might pay a similar amount due to LLM costs. Although, it could be more or less as well, depending on how it works out.

A couple of years ago, I saw a story of a guy writing two articles for a website a day. The boss asked him if he wanted to transition to AI-assisted writer for less pay. He said, "No." After a couple of weeks, he got canned. He checked the website out, and it had a bunch of AI writing on it.

LLMs are there to reduce your salaries and increase the businessowner's profits. Bigger inequality in wealth, it's only going to grow more and more. Also, a ton of people fired across many different fields.

I genuinely would rather see ads for products I might like yet do not know exist instead of purely random ads. I don't understand why a person wouldn't.

"LLMorphism may encourage objectification when people are seen as replaceable mechanisms or output-generating systems. However, LLMorphism does not necessarily involve using another person instrumentally. Its primary content is representational: it concerns how humans are conceptualized, not necessarily how they are exploited."

This is quite a scary truth. A year or two ago, I saw a person with a job where he wrote small articles for a website. The boss contacted him, asking if he wanted to become an AI-assisted writer instead for less money. "No," he said, wanting the full payments for his writing prowess. A week or two later, they canned him, and the website's articles nosedived in quality.

LLMs expand the supply of "competent" labor. After mass firings, the remaining workers, desperate for income, accept lower wages for AI-assisted roles. Wealth consolidates upward while wages race downward.

So I think LLMorphism might tie closely to exploitation. Mass firings and lower salaries going around while the 0.01% of machine-learning companies consolidate wealth by servicing numerous roles autonomously in some cases and by reducing salaries due to the larger body of "qualified" workers who can technically finish the job despite not having qualified in the past.

"LLMorphism is also distinct from predictive processing and related Bayesian theories of cognition. Predictive processing holds that the brain continuously generates predictions about sensory input and updates internal models in light of prediction error (Clark, 2013; Friston, 2010; Hohwy, 2013). But predictive processing does not imply that humans are LLM-like, nor that human understanding is merely text generation. Indeed, many predictive-processing accounts are deeply embodied and action-oriented (Allen & Friston, 2018; Clark, 2015; Pezzulo et al., 2024)."

I agree wholeheartedly here, because neural networks (NN) are stateless functions usually (not stuff like recurrent ones). On the one hand, with an infinitely fast computer, you retrieve the answer instantly. Brains, on the other hand, have neurons that communicate with signal delay. I bet if, in a weird world, we could simulate a brain with zero delay, a mind would cease to function correctly. Plus, neurons accumulate charge steadily before firing to nearby neurons. With NNs, you simply add up all the numbers, the "charge," and the ReLU function (or sigmoid for old-school machine-learning researchers) instantly "simulate" a neuron firing off to neurons connected to it.

"and and"

Just a heads up, you have a typo here.

"LLMorphism may therefore make fluency appear sufficient for understanding and, in doing so, devalue expertise and weaken educational norms."

I have heard the horror stories that youngsters these days are attached to screens with less ability to focus, but I'm not scared of that claim yet. For every generation, there have been those who kick the can down the road, skirting responsibilities, and all that changes with the generation is the activity: Instead of kicking a can down the road, they slide their finger across their phone's screen. The real test is tracking how many students across HS are in AP courses, learning Newtonian mechanics, electromagnetism, and of course, calculus among a couple others. Is that number dropping relative to the 90s and the aughts? Is it roughly the same as a percent of students? Or is it even going up, perhaps LLMs helping some types of learners explore topics to help them qualify for AP coursework? Now, if the percent is nosediving, then* I will be terrified for what the future holds for them and for me.

"clinicians also rely on how patients appear. Research on clinical communication shows that nonverbal behaviour is central to physician–patient interaction, including the expression of emotion, empathy, distress, and relational understanding"

LLMs are becoming multimodal with pictures "understood." No reason LLMs won't catch these non-verbal signals in the future that I can think up.

"The risk may be particularly acute in mental health, where suffering can be difficult to articulate and where coherent self-description does not always track clinical severity; behavioral and nonverbal signs such as psychomotor retardation, agitation, facial expression, vocal dynamics, and posture can provide clinically relevant information beyond verbal report (Dibeklioğlu et al., 2015)"

This is a great point, because a lot of people with schizophrenia and bipolar disorder with psychotic features suffer from anosognosia, the state of not knowing they have a medical condition.

"In this sense, LLMorphism may contribute to a broader epistemic shift: from evaluating whether claims are grounded, justified, and accountable, to evaluating whether they are coherent, fluent, and plausible."

Grifters have always weaponized confident fluency over evidence. Anti-science plagues America right now. Some gullible few absorb the message that ivory-tower elites intentionally block heterodox research that is a paradigm shift, sowing seeds of doubt about academia. For example, I saw a doctor's YT channel that claimed high cholesterol isn't necessarily bad and that statins should be avoided all while recommending saturated fats over seed oils. Of course, he sells a book with his "suppressed" knowledge alongside having an online market selling US$90/month supplements that his book recommends. They claim academics keep them out of the journals out of self-preservation since the "paradigm shift" would cause their grants to go bye-bye.

In reality, these charlatans combine cherry-picking of low-quality studies, telling a good story of the underdog fighting the establishment, and ignoring the body of evidence in support of the current expert consensus. Their grift is so illogical as if researchers wouldn't love to spark up a paradigm shift, becoming semi-famous and making more money, as if research isn't done decentralized across many countries funded by charities, different governments, and different corporations in competition with each other. Collusion without whistleblowers is simply impossible. Also, there's a difference between the corporate arm of medicine where they've been sued for billions before versus researchers who just follow the evidence to advance their research career and help everyone on the planet. Trust in expert consensus when it's this independent and decentralized and financed from all over the place with zero reason for an ulterior motive. They also pull off the, "Science has been wrong in the past." like Mac from It's always Sunny in Philadelphia. Science is in a state of constant flux where new evidence comes in, and the best guess, explaining as much evidence as possible right now, might change.

"Early childhood education is organized around relational pedagogy, attachment, affect regulation, and development (Cliffe & Solvanson, 2023)."

One aspect here is, mass-produced cartoons for kids teach aplenty and do a decent job at it. I'm not convinced, in two decades from now, we won't have human-looking cyborgs doing teaching like this.

"The broader point, however, is that public debate on AI has focused mainly on anthropomorphism: whether we are giving too much mind to machines."

This part reminds me of some recent research out of Anthropic. They uncovered that a few hundred vectors in their activation space linked up to concrete emotional states. They dubbed them functional emotions while warning these have nothing to do with subjective experience of sentience. That paper had fantastic details in it, though. They tested things by adding a big magnitude to a particular functional emotion, running some tests, and seeing how its behavior changed.

When "desperate," it not only hallucinated more as if it "felt" it must answer something, but it reward hacked more often. In a simulated situation, "desperate" Claude Opus blackmailed ~80% of the time whereas regular Opus did so ~20% while "calm" Opus did so ~0% (likely not zero, but they ran too few iterations of the test to approximate the probability).

When curious / interested, it altered how it searched through the solution space by considering more options. It even went deeper into a promising solution before ending its calculations when allowed to do so.

I understand being for privacy, but on the flip side, information about you can result in a better experience. E.g. in the case of tracking where a person comes from, that can help those two websites improve by coordinating with each other in some way. Or your ads might actually show you something you didn't know you existed that you end up buying. That's probably better than seeing ads you likely have zero interest in. I'll admit it's creepy when an ad is incredibly tuned to your recent internet activity, though.