> I imagine a lot of people feel like an engineer asking about the actual problem feels like getting goaded into some sort of pedantic debate.
Wow, that's describing it really well. I work for a guy who basically started a company by trying to vibe code his ideas into existence (late 2023 LLMs) and realized he would need actual developers to get anywhere.
It was excruciating trying to make him answer questions to get proper domain modelling going since LLM use had made him think of software as being wished into existence in a "declarative" way.
Not to mention the heavy contrast my persistent questioning had to a background of sycophantic yes-man claudespeak.
I was really wondering what these replies were talking about until this one - when I remembered a guy I used to work for that was just like this.
I think I have been pretty lucky the last 5-6 years of my career at least, where someone asking those questions is treated as trying to drive the team to a better result, rather than a pedant.
Eliezer Yudkowsky is not a scientist. He made a popular Harry Potter fanfiction series and a "rationality" blog-community that attracts "human biological diversity" enthusiasts.
Elizer Yudkowsky is a scientist, despite writing something you dislike 20 years ago and also having a blog. And no, he's not associated with race realism, that's just baseless libel.
>Schwitzgebel, Strasser, and Crosby fine-tuned GPT-3 on Dennett's corpus
This sounds like a completely different scenario. How many users who post LLM written blog posts are tuning the weights of their LLMs on a large corpus of their own original writing? I wouldn't doubt that this produces far more convincing and pleasant output than the disgusting slop from out of the box Claude.
>or if this is just liberal minded "see the big companies are out to get us and this is just the current example".
Of course, as opposed to the objective minded "there is no conspiracy ever, when it is repeatedly demonstrated that corporations will poison, experiment on and destroy human lives if it favors their bottom line and they think they can get away with it, this is not to be interpreted as a pattern or structural issue, it's just coincidences and 'bad' individual actors all the way down".
Could I hook up a SOTA model the 2D computer puzzle game Gruntz (1999) so that it can read it from screenshots and act on it through keyboard and mouse inputs in a way where it would learn how to play and progress through the game? I don't think so. I doubt we'd see any sign of progress in building an internal model of how the game works and the win states in its "thinking" tokens.
Anyone that can read English could do that though.
> Could I hook up a SOTA model the 2D computer puzzle game Gruntz (1999) so that it can read it from screenshots and act on it through keyboard and mouse inputs in a way where it would learn how to play and progress through the game? I don't think so. I doubt we'd see any sign of progress in building an internal model of how the game works and the win states in its "thinking" tokens.
You.. literally can? I have no idea what 90% of the people here are saying, it's like they've never even used one of these models before.
Can you? Let's say you prompt it with "this is a puzzle computer game, your objective is to progress through its levels" plus the controls from the instruction manual and tie it to a vision + KB and mouse harness.
Will it effectively create an internal model describing world objects and how they interact with each other, persist that so it doesn't get lost when it's context window gets filled up, then after it has sufficiently complete knowledge of the fundamentals after the tutorial levels successfully apply that model by making plans to solve the puzzles and execute them by clicking the right coordinates tied to the visual feedback?
I highly doubt it. To me it often just looks like people are defining narrow search spaces (e.g by having all of the task complexity pre-digested by the harness design), pointing a brute force engine at them, spending 20 thousand dollars in compute and then saying "hey look, it can do anything!".
Oh, I see. That’s a very different requirement. It’s not a technical limitation but a product decision to not allow training. An advantage of properly open source models is that you can train and tune them.
It’s an interesting challenge though. I might start to tackle it by having the model write its own tool program(s) to play the game. It’s possible that the model could choose that strategy itself from a high level prompt alone.
It can't even sprint because it's incapable of pressing two buttons at the same time. Maybe we get the two button tech before we start celebrating AGI.
1. It’s too slow for real-time games. To play mario, you’d need to step frame by frame like a TAS. I don’t know if Gruntz has real-time elements or not.
2. It will be expensive. You won’t get very far with a Plus subscription.
The models likely already have some knowledge on game objectives unless the game is really obscure, so it should do a decent job. It can figure out details of the mechanics along the way.
In my experience, using LLM's for generating code has definitely dulled my capacity to think about function hierarchies, composition and having a "picture" in my head of what will go where in the repository.
Raw "leetcode" algorithm intuition and database modelling/SQL are still largely intact for me because I never trust LLMs for those and want to actually understand what I'm building. But it's scary to think how easy it is to have your skills eroded and how quickly you become dependent on these tools.
From your comment, it seems like the only skill that has eroded in you apparently is the fact that you don't understand the structure of a repo that was written by someone else? Was that a skill you had pre LLM that you could just touch a repo you didn't work on and magically knew where everything went?
Furthermore, if you can see it so transparently and have put a stop gap, why do you think the rest of the world or programmers will not? Why will they fall into the temptation that you can resist? Especially if we are talking about that specific subset of people that are already really good at coding and understand the pitfalls pretty well.
Maybe you can make the argument that someone who purely vibecodes never gets the skills of raw algorithm intuition/specific knowledge but in those cases, no skills get eroded because there were none to begin with.
I wish the soy one were hidden on a corner on my local supermarket. There isn't even one anymore in my country, and that's the biggest soybean producer in the world.
Other plant milks are 3x the price of the cow one because of economies of scale and subsidies. It's all a big joke.
In some countries they're not allowed to call it "soy milk" which is a big joke. I understand why you can't call it "milk <font size=1>of the soy variety</font>" but nouns with adjectives can, in fact, refer to things the adjective-less noun doesn't refer to.
So the soy yoghurt in Germany is called Alpro - the brand name. The dairy industry even fucked themselves over, because even their real yoghurt doesn't meet the strict definition and has to be called "mild yoghurt" instead.
While it's great that green energy is now economically feasible, the idea that there will quickly (say, in the next 50 years) be a "green transition" that eliminates the usage of carbon fuels is still highly dubious.
What we've seen is new renewable capacity carving out a tiny sliver in the energy mix, sometimes even being able to meet all new demand in the installation period, but we still have all the existing demand running on the old stack.
There's also the fact that global society is not a real time strategy game being managed by a well intentioned min-maxing actor driving compounding efficiency gains in resource usage. It's a free for all where everyone is paranoid that they'll be destroyed if they don't use every dirty trick they can (on a global geopolitical level), entrenched oligarchies permanently lobbying for their inefficient, destructive structures to continue (on a domestic economy level) and so on.
Sometimes small tools are all you need, even if they're slop.
About a month ago I pirated an Argentinian movie and the only subtitles available in my language were out of sync and at a different speed/framerate so adjusting for delays wasn't enough. I was unsuccessful at fixing it with VLC and every other "online tool" I could find.
Knowing a srt file is just text with timestamps I vibe coded a python script to take in sample times throughout the movies so it could recalculate the rate and shifts and replace them in the file. It worked on the first try using only the deepseek web chat interface and my terminal.
Without AI I theoretically could have sat down with a pen and paper to figure out the math adjustment, then looked up python input handling syntax which I already forgot, typed something out and then hammered it into shape through trial and error over a few hours. But the friction and time investment of doing that would have been so great I would have just given up on watching the movie instead.
Not really. It's probably just international users not tolerating american exceptionalist hubris when they see it, especially now that the pretending is done.
Plus, what a shocker it is that a "hacker news" website would have many people with an affinity towards open source technology they can fiddle with and have autonomy over, and a distaste for tech monopolies trying to control and limit access to these tools.
Wow, that's describing it really well. I work for a guy who basically started a company by trying to vibe code his ideas into existence (late 2023 LLMs) and realized he would need actual developers to get anywhere.
It was excruciating trying to make him answer questions to get proper domain modelling going since LLM use had made him think of software as being wished into existence in a "declarative" way.
Not to mention the heavy contrast my persistent questioning had to a background of sycophantic yes-man claudespeak.