fasterik · HackerTrans

4 ngày trướcdiscuss

I don't think it's overstating it at all. If anything, I would argue that code is more expressive than literature or music because it's more general. Consider the space of all possible video games, demos/intros, generative paintings, generative music, etc. which wouldn't be possible without code.

I think you're confusing the common use of a medium with expressiveness. Many uses of ordinary language also aren't that interesting, but we need to consider the full space of possibilities.

fasterik

4 ngày trướcdiscuss

You're still intentionally misreading the OP's statement. If you read it again in context, they're clearly saying that they think training bigger LLMs is not sufficient. I think I agree with that statement, but my confidence is pretty low.

No, I don't think LLMs are necessary for AGI at all. I think there are multiple paths to AGI, some of which involve LLMs and some which don't.

fasterik

4 ngày trướcdiscuss

>AGI won't rise out of AlphaZero and AlphaFold in the same way AGI won't rise out of Houdini chess engine.

That's a straw man. Nobody thinks AGI will rise out of domain-specific systems. The question is whether domain-specific systems are necessary for AGI.

Of course, the problem is that AGI isn't a well-defined concept. But if we define it as achieving superhuman performance across several hundred domains where there are objective measures of success, it doesn't seem far-fetched to predict that it will involve some general reasoning system paired with a bunch of specialized modules.

fasterik

4 ngày trướcdiscuss

Your examples (math, literature) involve natural language. It stands to reason that a general language model will be more competitive in those domains. If you want examples of successful domain-specific models, look at AlphaZero and AlphaFold. LLMs aren't anywhere close to achieving that level of competence at abstract strategy games or protein folding.

"This will never work" is a pretty confident assertion for a field that's so young and rapidly evolving.

fasterik

4 ngày trướcdiscuss

I still don't see why it's surprising, though. Humans and machines are subject to the same laws of physics. While we wouldn't expect LLMs to think exactly like humans, we should expect some degree of convergent evolution. There are always tradeoffs between storage capacity, accuracy, latency, relevance, fluency, etc.

Expecting LLMs to be magically perfect information retrieval machines is never going to be realistic.

fasterik

5 ngày trướcdiscuss

This doesn't seem that weird to me. Talk to any human and you'll find that their ability to recall specific names and facts is very context-dependent. Phrasing a question in one way can make it hard to answer, while providing certain words or cues can instantly "jog" the memory.

fasterik

6 ngày trướcdiscuss

There's already a good solution to buying and owning digital media: you pay money to download files that are playable offline.

When you pay for content locked to a platform, you're not buying an asset, you're paying for a service. The platforms grow around not only providing a convenient service to the end user, but also to the content creators, who publish on them with the expectation that their content is protected by DRM. Creators are free to choose where they publish, and end users are free to choose which services they use.

I don't think it makes sense for the government to define what it means to own a digital asset or to force every service platform to become a retailer and ownership-tracker. Where there's demand for DRM-free downloads or physical media, the market will respond.

fasterik

7 ngày trướcdiscuss

>adults often confuse learning with consuming material about learning

True, and even more insidious than that can be consuming the actual learning material (e.g. textbooks), but not doing the required work to integrate it. I find that I need to do projects to properly learn something. Once I actually start doing things, I quickly identify the parts I knew in theory from reading about them but had never put to the test by solving real problems.

fasterik

9 ngày trướcdiscuss

Computer graphics is intrinsically interesting and rewarding. It sits at the intersection of several important fields, from computer science to mathematics to theoretical physics to low-level programming.

Maybe steering away from it is good advice for someone who's looking for a career transition but doesn't care about what they're actually doing. But that's not a good way to go through life; my advice to such a person would be to follow what they find interesting and valuable, and constantly challenge themselves to learn new things. Then deciding whether or not to learn computer graphics is relatively straightforward and it will be a net positive for the right kind of person. Even if they don't make it a career, the skills transfer well to many other areas.

fasterik

11 ngày trướcdiscuss

>Peter Garber argues that the trade in common bulbs "was no more than a meaningless winter drinking game, played by a plague-ridden population that made use of the vibrant tulip market."

So basically, it was the GameStop of the 1630s. Humans never change.

fasterik

11 ngày trướcdiscuss

When I was a kid I got obsessed with Linux, but my family only had one PC in the living room. After an attempt to set up Windows/Linux dual boot where I messed up the partition table, my parents banned me from tinkering with it. Luckily I discovered Knoppix and other live distros, which allowed me to boot into a safe environment to play around in.

fasterik

13 ngày trướcdiscuss

>IMO, the pressures leading to degradation are all somehow linked to universalization

Liberalism is fundamentally a universalist idea. I mean "liberal" in the original sense: the cultivation of free and indepdendent human beings capable of governing themselves. A democratic society can't survive for long without universal liberal education.

We need to distinguish between education and the education system. You make good points that the system as it is today doesn't scale as much as we might like. But that doesn't mean we should abandon the goal of universal education. Unfortunately, I think the solution requires something much broader and deeper than more schools and more teachers. It requires a culture that values learning and independent thinking, parents who bring up children who are curious and willing to learn, and institutions that uphold these values in society.

We made a lot of progress over the past few centuries, but now there's an increasing number of people who want to question and undermine the core values of liberalism and replace them with something either more elitist, more authoritarian, or both.

fasterik

16 ngày trướcdiscuss

I think the question isn't whether (DOOM + 1) would have succeeded, but what they would have been able to accomplish with the Quake team if they'd all had the stamina and job satisfaction to make several games over the following decade. DOOM was already a massive success and they had the money to do whatever they wanted. It's undeniable that Quake was id's most groundbreaking game, but maybe (DOOM + 5) with the same team would have been even more so.

fasterik

23 ngày trướcdiscuss

It's the cadence and the density of cliches. I've seen Opus produce very similar output. That's not to say it's 100% that this was written by an LLM. They had to learn this style from human writing in the training data. It's just a very condensed form of bad writing.

fasterik

26 ngày trướcdiscuss

My prediction is that it will go the same way as the dot com bubble. The hypesters and fraudsters will eventually collide with objective reality, but the technology will persist and society at large will benefit from the infrastructure and the increased access to knowledge.

fasterik

26 ngày trướcdiscuss

I love computers too, but it doesn't resonate with me when people call AI "snake oil." The comparison suggests that the thing doesn't do what it's marketed to do. AI does more or less what it's marketed to do, sometimes badly.

I still write code by hand. But LLMs have been a legitimately useful tool when I've wanted to dig into a new field like computer graphics, theoretical physics, or numerical analysis. Or even just asking the LLM to write a piece of code and learning from its output. I think it makes me a better programmer because I can bootstrap the knowledge needed for a new project much faster and spend more time programming.

fasterik

28 ngày trướcdiscuss

I don't think "refusing to learn C++" is the right way to frame it. I want to use the language features that are actually useful to me, without being forced into a specific programming style. I can't speak for every "orthodox C++" programmer, but for me that means using exclusively plain-old-data structs, non-member functions, and "dumb" pointers. I have no issue with learning to use a C++ feature when it's directly useful to a problem I'm trying to solve.

As one example, I recently found templated lambdas useful in making animations.

https://www.youtube.com/watch?v=cw-h0dePYZM

fasterik

28 ngày trướcdiscuss

Regarding your last point, I think it's almost always wrong to move something discontinuously, but I do think designers should think a lot more about getting out of the way of the user. A 50-100 ms animation is more than enough for most motions and keeps the UI feeling snappy. Also, animation should be decoupled from input wherever possible. I hate it when I have to sit there waiting for an animation to complete before the app will start acknowledging my keystrokes.

fasterik

28 ngày trướcdiscuss

We're not recording reality, but we're trying to create convincing and aesthetically pleasing effects for brains that evolved in reality.

fasterik

28 ngày trướcdiscuss

Our culture has too much focus on landing a job and not enough focus on becoming the kind of person who can adapt and thrive in any situation.

Computer science isn't for everyone, and probably the people going into it for the money should look elsewhere. You should study computer science if you find it intrinsically interesting. If you fall into that category, it will teach you how to think about problems rigorously, how to find solutions and break them down into steps that can be stated unambiguously, and how to reason about the performance and real-world tradeoffs of complex systems. Those are skills that will never be outdated, even if programming becomes fully automated.

comments