clejack

Hi HN,

This project distills a model's word embeddings into human-interpretable "concept-vectors", i.e. vectors in which each component tracks concerns like semantics, syntax, and even statistics potentially, while associating each component with a human readable and human definable label.

See the readme[1] for details.

I want to acknowledge explicitly that this is a data design project. I have quite a bit of experience with data transformation and manipulation, but limited experience with NNs. I have not tested this on models, and I currently don't have the resources to build a comprehensive database to test it on models. I'm posting primarily for human feedback/criticism, and simply to share the idea since this is as far as I can currently take it.

[1] https://github.com/truehumanexe/concept_vector/tree/main

Yes, I recently got access to an annotations platform for llms, and I've found many projects associated with generating chain of thought outputs.

These COT outputs are the same sort of illusion as the general output. Someone is feeding them scripts of what it looks like to solve problems, so they generate outputs that look like problem solving.

I can't remember if I mentioned it previously on here, but an llm seems to be an extremely powerful synthesis machine. If you give it all of the individual components to solve a complex problem that humans might find intractable due to scope or bias, it may be able to crack the problem.

Rest assured, if you force students to learn basic english and math, the vast majority of students will experience this as being forced to study things they don't care about.

The difference with what I'm suggesting is that they won't be forced to learn about 7 or 8 different things they don't care about at the same time.

The allocation of teachers' time will be better with a more constrained curriculum, and the classes where students choose to learn about a subject will be a more engaged.

Framing learning things you're uninterested in as "learning to get over yourself" is odd. This isn't an ego problem, and dictating personality traits to such an extent is a questionable goal.

I think you have the right question, but I think the answer is a resounding "no." Old thinking towards education seems irrelevant in modern times. Children should be taught the basics of mathematics ,language, and technology as a necessity for interacting in the real world. This should include arithmetic up to applied algebra, grammar, reading, and, ideally, critical thinking. These core topics should have traditional testing and homework.

Everything else should be about exposure. So children are lectured on science, history, or whatever other subject; but they don't actually need a grade in these subjects during elementary school. This would reduce the work burden on students and teachers. The only purpose is to light a spark in those with true curiosity.

In high school, students should be able to choose topics of interest that they learned about in elementary school to do more intentional learning with tests and grades. Everyone else continues on a general path with the core subjects being tested and non-core subjects simply being lectured.

In college, those who chose a specific focus in highschool accelerated their learning for that subject. For others, if they didn't find anything interesting, they can go into a trade or whatever else they choose. If they are late bloomers, they go to college and cultivate their newly found interests with a larger back log of pre-reqs.

There's no point in "teaching" children things that they immediately forget only for them to go on to become a generic office worker or retail employee. We should cultivate those with the desire to be cultivated, and stop pretending that it's actually feasible to have an entire society of "intellectuals." There is a place in the world for those who don't care about learning, but there is little sense in throwing significant education resources at them.

I think you all understood my point but for the sake of clarity, I said "take anyone from any group," and I was really thinking along the lines of "take a new born from any group."

Even if the creator specifically had eugenics in mind, I think he stumbled upon a greater truth.

Consider this. You can take anyone from any group in your nation, place them in a different nation, with a different culture, and they will adopt the mannerisms and accents of that culture.

We focus on race constantly, but it's clear that culture drives the norms that we see in any group. And culture may be persistent (especially now with technology allowing every culture to potentially spread everywhere), but it's not intrinsic.

With this framing, I interpreted Idiocracy's intro as being about a culture of intelligence or learning being harder to maintain in a modern world, than a culture of apathy or fun.

I watched this movie really late. Let's say within the past 2 years or so. After watching it, all I could think was, "This isn't a comedy, it's a tragedy."

It felt way too close to home.

I have to point out that your original post is technically correct because you specified "medical expertise" as the focus of your argument and psychologists aren't MDs. The field has some questionable aspects (and outcomes) to be sure, but I don't think it's completely without merit, and as a consequence, I feel the spirit of your argument is still wrong. You said:

> At the end of the day, this is a cultural issue, not a medical one, and needs to be solved via cultural norms, not via political intervention based on contrived pretenses

It is possible to consider people's subjective experiences in tandem with the consequences of those experiences and make an empirical judgement. The consequences can be quantified, even though the subjective experience itself can't.

If we found that people began committing suicide after using social media, would you suggest this can't be studied, and that a government wouldn't have good reason to want to legislate against social media in these circumstances?

This is really all I'm trying to get at. Replace suicide with depression, reduced quality of life, addiction. Whatever you like. If it holds in the suicide case, it holds in all of them.

Just to make sure I wasn't misunderstanding you, I double checked the meaning of "normative." This is the first result from google:

"establishing, relating to, or deriving from a standard or norm, especially of behavior."

And other sources have something similar. I'm interpreting your comment as saying "(psychological) harm is subjective, and because it can not be measured empirically, it's not possible to have expertise on this topic."

Fortunately, there are real world consequences that can be measured. If I take an action that makes many people say "ow!" and we acknowledge that expression as an indicator of pain, even though I can't measure the exact level of pain each person is experiencing, I can measure how many people are saying "ow!" I can measure the relationship between the intensity of my action, and the number of people that respond negatively. There's plenty of room for empiricism here, and a whole field of mathematics (statistics) that supports handling "normative" experiences. They even have a distribution for it!

The foundation of law is not scientific exactness or scientific empiricism. It is the mechanism by which a state establishes norms. A law against murder does not stop murder, but it does tell you that society does not appreciate it.

If human biological intelligence is our reference for general intelligence, then being skeptical about AGI is reasonable given its current capabilities. This isn't biological narcissism, this is setting a datum (this wasn't written by chatgpt I promise).

Humans have a great capacity for problem solving and creativity which, at its heights, completely dwarfs other creatures on this planet. What else would we reference for general intelligence if not ourselves?

My skepticism towards AGI is primarily supported by my interactions with current systems that are contenders for having this property.

Here's a recent conversation with chatgpt.

https://chatgpt.com/share/69930acc-3680-8008-a6f3-ba36624cb2...

This system doesn't seem general to me it seems like a specialized tool that has really good logic mimicry abilities. I asked it if the silence response was hard coded, it said no then went on to explain how the silence was hard coded via a separate layer from the LLM portion which would just respond indefinitely.

It's output is extremely impressive, but general intelligence it is not.

On your final point about functional replacement not requiring biological mimicry. We don't know whether biological mimicry is required or not. We can only test things until we find out or gain some greater understanding of reality that allows us to prove how intelligence emerges.

When I was in my mid 20s, I interned at a machine shop building automotive parts. In general, the work was pretty easy. I was modifying things via cad, doing dry runs on the cnc machine, loading raw material, and then unloading finished products for processing.

Usually there was a cadence to things that allowed for a decent amount of downtime while the machine was running, but I once got to a job where the machine milled the parts so quickly, that I spent more time loading and unloading parts than anything else. Once I started the first part, I didn't actually rest until all of them were done. I ended up straining my back from the repetitive motion. I was shocked because I was in good shape and I wasn't really moving a significant amount.

If I talk about excessive concern for productivity (or profit) being a problem, certain people will roll their eyes. It's hard to separate a message from the various agendas we perceive around us. Regardless of personal feelings, there will always be a negative fallout for people when there's a sudden inversion in workflow like the one described in this article or the one I experienced during my internship.

So what you're saying is the laws of thermodynamics that I learned in school don't apply to biological systems?

I'm admittedly in no position to argue deeply about biology, or even physics for that matter, because I studied mechanical engineering. I dealt with physics in an applied manner.

With that said, this story comes to mind:

https://en.wikipedia.org/wiki/Angus_Barbieri

It seems to disagree with your idea that fat is not fuel. And the first law of Thermo seemed to be applying to this man. He had little to no energy input (via food) and a baseline energy expenditure from just existing, so his system burned stored energy via body fat.

The printing of money has primarily lied within the purview of the government from the start. Money is one of the few modern physical item, off the top of my head, that this statement applies to. Maybe there are seals or other official marks that this also applies to, but all of these items fall into a similar category.

So while the legislation, and implementation can be deemed problematic, the political desire to prevent counterfeit is not actually unreasonable.

Having particular objects be banned that aren't under the exclusive control of a government actually creates new precedent. Regardless of the technical feasibility that you keep bringing up, this legislation is undesirable because of what could come after.

"Absent corruption" is doing a lot of heavy lifting in your statement. The idea that the system can't fail raises the question what do you consider failure, and what do you consider corruption"

If prices increase and wages don't keep up with them, an increasing number of people become squeezed by their environment. This is a slow event, sure, but enough drops can fill a bucket. The fallout from this pressure on the general populace will be the failure that you're saying can't happen. This seems inevitable without an intervening event to reset things.

With that said, I don't think your concerns are unreasonable, and I'm not sure UBI by itself could solve anything. At a minimum price controls or government administering of food and housing would be necessary to keep prices from rising in response to the influx of cash everyone would receive, but the problem of people not working does seem like a big potential issue.

I believe there have been studies to the contrary, but those studies necessarily miss the universal part of ubi, so they don't have the negative feedback loops that could spring up in a real implementation.

Stardew valley was apparently solo developed, and if Google is accurate it has sold over 40 million copies. Even if he sold it for a dollar, the dev would be very successful by most standards.

For the folks who have more positive outlooks how often do you change your code after it's been generated?

I haven't used agents much for coding, but I noticed that when I do have something created with the slightest complexity, it's never perfect and I have to go back and change it. This is mostly fine, but when large chunks of code are created, I don't have much context for editing things manually.

It's like waking up in a new house that you've never seen before. Sure I recognize the type of rooms, the furniture, the outlets, appliances, plumbing, and so on when I see them; but my sense of orientation is strained.

This is my main issue at the moment.

I can't speak to the current system because I have stepped out of it temporarily, but if you haven't done so; please tell your daughter that she is not her job nor is she the labels she places upon herself. It's tragic that she's placed so much weight on her job if she still has you supporting her.

None of the things you listed are signs of merit, they are signs of pedigree. If people recommended them here, they did so in error.

There's certainly a difference between universities, but the most important differentiators are connections (has she exhausted these) and prestige. If those aren't working for her, the only thing left is personal projects. That is the true indicator of merit in the software space.

As for specific advice, your daughter is in a similar situation to me. I graduated thinking I had did all of the right things, and that my degree (mechanical engineering) was some sort of magic ticket. I was unemployed and then underemployed for a year or two. I eventually went to a job fair and got a job as a data analyst then, finally, moved into data engineering.

Reality has shown her that there isn't always a direct path to a goal. Are there other skills she has that she could use to get meaningful or interesting work?

Tell her to explore alternate jobs outside of her field or preferred industry, build up a portfolio of projects on the side to keep her skills sharp, and keep applying to her preferred role, but now at a much slower and deliberate pace.

Am I the only one who doesn't take this statement literally and immediately extrapolate it to all aspects of an individual's existence? "I have nothing to hide" is a broad statement that clearly encompasses "everything", but it's often said in the context of a specific thing that a person doesn't care about.

Those of you who would ask someone for financial information after they say this, would you also say "it's hot out side" if they described something as cool during the summer?

Ultimately, given the complexity of security, expecting there to be some cultural shift on privacy is silly unless it's made trivially easy. We can't get people to eat right, exercise, or control their screen time and social media use and all of those have more immediate and tangible consequences.

I appreciate the message, but I don't think the call to action is practical.

I'm from the future and I'm here to tell you how to defeat this current iteration of AI. Stop entering text into their prompt and they die. You're welcome.

So to avoid the suffering of gambling we need to simply suffer and toil so much in day to day existence that we don't have the capacity to engage with anything else?

Pity indeed. If people don't actually believe in these things and they're simply repressing themselves this can't be healthy.

I was happily lurking until I saw this astounding response.

"Every relationship with {men|women} I've been in has been bad, so romance is obviously worthless."

"My neighbors dog always barks at me, I didn't get why anyone likes dogs."

"I've had a bad experience with ${race} so I really wish we could get rid of them."

"I caught the flu, and it didn't kill me. I don't get why people are always worried about it."

"I've never worn a seatbelt, and I'm still alive. They're a waste of time."

"School was a bad experience for me personally, so best to get rid of it"

Are you serious right now?

While incompetence might be an issue, I think the greater problem is that Microsoft is rolling back control and generally sucks at UX.

Why does this app that's been working just fine as desktop software need to save anything to the cloud by default? It's conceptually odd.

I've used Google docs from the beginning, but I actively choose what docs I want on that platform.

All MS had to do was add "save to cloud" as an additional save option along with "save" and "save as" (maybe renamed as "save to desktop") then auto save could activate where your last save location was. This would be good design.

Reich can't prove what happened? Slaves being prohibited from education? Books being burned?

Reich can't prove the internal working of the president's mind, so statements of intent are speculative to a degree, but if you interacted with a person in such a way that you projected feelings of animosity and hatred towards him, I wouldn't need your explicit admission to determine that you hate him or that you are consistently behaving in ways that appear hateful.

The president's actions have certainly shown a disdain for the education system. Vance has explicitly expressed taking issue with universities and wanting to implement government control. Trump has mandated that universities allow review for their curriculum if they want to receive funding.

This article isn't alarmist, because the actions that anyone would need to be alarmed about have already happened. The only question at this point is whether or not you like the results.

Not only does a lot of news have no real value a lot of news does not generate value of any kind (real or otherwise) until someone reads it.

For example, an opinion piece is meaningless unless someone reads it, so writers find themselves in the same situation as every other artist, even if their writing isn't artistic in nature.

Attention is a finite resource. This might be unpleasant to hear, but just because you're working on something, doesn't mean it has intrinsic monetary value.

What twisted agenda? The kicked article is about his hopes for new Alzheimer's treatments. The article after that is about his goal to give away his money.

The main issues for problems like this fall into 3 categories

- Things that prevent you from starting the job. Org silos, security, and permissions

- Things that prevent you from doing the job. This is primarily data cleaning.

- Things that make the job more difficult. This involves poor tooling, and you'll struggle to break the stranglehold that SQL and python-pandas have in this area. I'll also add plotting libraries to this. Many of them suck in a seemingly unavoidable way.

On the second and third points llms will most likely own these soon enough, though maybe there's room to build something small and local that's more efficient if the scope of the agent is reduced?

The first point is organizational generally, and it's very difficult to solve outside of integrating your system into an environment which is the strategy pursued by companies like snowflake and databricks.

Honestly, it does make me wonder what they're thinking. Given the nature of what they're doing, the stable coin reference doesn't seem like it's necessary. Even for purposes of hype.

Telling investors you're going to make your own currency to bypass transactions fees from financial institutions seems like a fairly straightforward way to boost investor interest.

Wouldn't adding crypto mechanics to a virtual credit system just make for a costlier system due to mining?

If the transactions are maintained within an Amazon or Walmart ecosystem, there's no need for trustless verification, because you own the ecosystem. What's the point?

"cosmic sidekick for navigating life with the power of the stars...with a fresh, no-BS approach."

There's a great deal of humor here. Astrology is conceptually interesting, in the same way that me summoning a Greek, Norse, or Hindu deity in old school final fantasy is interesting. I get the feeling this place isn't filled with your target audience though.

Anyway, consider allowing users to type in their birth day in addition to (or maybe even entirely replacing) the day widgets. It's not immediately obvious where to click to navigate to a year only selection for instance.

And even though it may seem obvious, a prompt that says "enter your birthdate" or something similar oriented around the entry fields would be a good idea for the uninitiated but curious.