Or till that Asimov story happens in real life to you, you are in a room with people arguing about who is AI and who it not - and in the end you were actually the only human left.
Music, like all art, is a human expression. AI has no desires, it feels nothing, it believes in nothing, thus it has nothing to express. It may imitate music, but it's not music.
Would you consider a beautiful sunset as art? is the value of music found in the source or the listener? I argue it is only the listener. The source is irrelevant. That is surely the case for me, and I dont think I'm unhinged or insane. I have a strong feeling I am not a minority in this regard.
Pythagoras argued that music is essentially number and proportion. If beauty is found in the geometry of sound, then the "belief" of the architect is secondary to the elegance of the structure.
Upon hearing mashed up pop music with almost coherent lyrics, "Shall I compare thee to a sunset?" What is going on here?
Have you not heard music before? Is Suno your first experience with music-shaped sounds? Because, buddy, this is wild. You're not getting Rumours out of an AI. You're not getting Time (The Revelator) out of AI. London Calling does not spring from the geometry of sound.
Maybe I'm just not as emotional as you. Could definitely be the case. Even before AI music I never cared much about lyrics. Nor artist names beyond finding similar music to a song I like. I listen to music for the sound, which does elicit emotion and feelings that are more enjoyable or less depending on my mood, but I don't care about the story being told.
I still don't think you're saying anything that refutes the geometry of sound argument, however. If you heard an AI song you liked, and didnt know it was AI, and found out after the fact, would you be rational enough to accept you could be wrong? Or would it turn you off to the song irrationally?
Tangential but AFAICT most people don't care about lyrics. If they did, so many hit songs would not be hits.
To name one, "Saving All My Love For You" should never be played at a wedding because the song is about having an affair with a married guy with kids. But no one listens to the Lyrics. They just hear the chorus. It's a hit for other reasons, not because of lyrics.
Similarly, few people listen to the lyrics of ""Rainy Day Women #12 & 35" (everybody must get stoned). It is not about drugs.
Heck, famously there's Bush Jr. (or more likely some PR person) using "Born in the USA" as a pro-American song. It's not a pro-America song at all. I wouldn't call it anti-American but it's definitely a song entirely about problems in America. Not praise.
You're all over this thread shilling, please stop.
Also, you already predefined that certain actions would definitely be rational, and certain others would definitely be irrational, and then ask what action would the person take? Lmao.
As a typical example DeepSeek v4-pro has 59B active params at mostly FP4 size, so it needs to "find" around 30GB worth of params in RAM per inferred token. On a 512GB total RAM machine, most of those params will actually be cached in RAM (model size on disk is around 862GB), so assuming for the sake of argument that MoE expert selection is completely random and unpredictable, around 15GB in total have to be fetched from storage per token. If MoE selection is not completely random and there's enough locality, that figure actually improves quite a bit and inference becomes quite workable.
I've never seen reports of this kind of setup being able to deliver more than low single-digit tokens per second. That's certainly not usable interactively, and only of limited utility for "leave it to think overnight" tasks. Am I missing something?
Also, I don't know of a general solution to streaming models from disk. Is there an inference engine that has this built-in in a way that is generally applicable for any model? I know (I mean, I've seen people say it, I haven't tried it) you can use swap memory with CPU offloading in llama.cpp, and I can imagine that would probably work...but definitely slowly. I don't know if it automatically handles putting the most important routing layers on the GPU before offloading other stuff to system RAM/swap, though. I know system RAM would, over time, come to hold the hottest selection of layers most of the time as that's how swap works. Some people seem to be manually splitting up the layers and distributing them across GPU and system RAM.
Have you actually done this? On what hardware? With what inference engine?
At least some of the investors in this tech are hoping for a monopoly position. They'd like to outspend the competition to get an insurmountable lead, at which point they can set their price.
But, so far, competition remains fierce. Anthropic still has the best tools for writing code. That lead is smaller than it's ever been, though. But, honestly, Opus 4.5 is when it got Good Enough. If Anthropic suddenly increased prices beyond what I'm willing to pay, any model that gives me Opus 4.5 or better performance is good enough for the vast majority of the work I do with agents. And, there are a bunch of models at that level, now maybe including some discount Chinese models. Certainly Gemini Pro 3.1 is on par with Opus 4.5. Current Codex is better than Opus 4.5 and close to Opus 4.7 (though I won't use OpenAI because I don't trust them to be the dominant player in AI).
I often switch agents/models on the same project because I like tinkering with self-hosted and I like to keep an eye on the most efficient way to work...which models wastes less of my time on silly stuff. Switching is literally nothing; I run `gemini` or `copilot` or `hermes` instead of `claude`. There's simply no deep dependency on a specific model or agent. They're all trying to find ways to make unique features for people to build a dependence on, of course, but the top models are all so fucking smart you can just tell them to do whatever thing it is that you need done. That feature could probably be a skill, whatever it is, and the model can probably write the skill. Or, even better, it could be actual software, also written by the model, rather than a set of instructions for the model to interpret based on the current random seed.
Currently, the only consistent moat is making the best model. Anthropic makes the best model and tools for coding, but that's a pretty shallow moat...I could live with several other models for coding. I'll gladly pay a premium for the best model and tools for coding, but I also won't be devastated if I suddenly don't have Claude Code tomorrow. Even open models I can host myself are getting very close to Good Enough.
I don't get it. I use Claude Code every day, what I would consider pretty heavy usage...at least as heavy as I can use it while actually paying attention to what it's producing and guiding it effectively into producing good software. I literally never run into usage limits on the $100 plan, even when the bugs related to caching, etc. were happening that led to inflated token usage.
WTF are y'all doing that chews tokens so fast? I mean, sure, I could spin up Gas Town and Beads and produce infinite busy work for the agents, but that won't make useful software, because the models don't want anything. They don't know what to build without pretty constant guidance. Left to their own devices, they do busy work. The folks who "set and forget" on AI development are producing a whole lot of code to do nothing that needed doing. And, a lot of those folks are proud of their useless million lines of code.
I'm not trying to burn as many tokens as a possible, I'm trying to build good software. If you're paying attention to what you're building, there's so many points where a human is in the loop that it's unusual to run up against token limits.
Anyway, I assume that at some point they have to make enough money to pay the bills. Everything has been subsidized by investors for quite some time, and while the cost per token is going down with efficiency gains in the models/harnesses and with newer compute hardware tuned for these workloads, I think we're all still enjoying subsidized compute at the moment. I don't think Anthropic is making much profit on their plans, especially with folks who somehow run right at the edge of their token limit 24/7. And, I would guess OpenAI is running an even lossier balance sheet (they've raised more money and their prices are lower).
I dunno. I hear a lot of complaining about Claude, but it's been pretty much fine for me throughout 4.5, 4.6 and 4.7. It got Good Enough at 4.5, and it's never been less than Good Enough since. And, when I've tried alternatives, they usually proved to be not quite Good Enough for some reason, sometimes non-technical reasons (I won't use OpenAI, anymore, because I don't trust OpenAI, and Gemini is just not as good at coding as Claude).
I would say most people are complaining about the $20 plan, which is now actually indistinguishable from the free one. I tested it and ran into limits immediately. With Max, i can work properly again.
You know what? Good for Friendster. It can't be worse than what we have now. But, also, good luck with that. Facebook may be terrible (it is definitely terrible), but it's where everybody's friends are. I dunno how you reproduce that network effect today, but maybe they do.
As someone put in the unfortunate position of building an AI IDE in three weeks, I assure you it's much more difficult than it seems. Sure, we were able to get something working with all the features you would expect, but the performance was awful. Claude Code, Cursor, and others do a lot of tweaking based on a lot of experience in order to make their systems give good results. There is more to getting good results than just using a good model.
Decent analogy in 2nd sentence BUT the 1st doesn't hold water. TIMTOWTDI, statements about "everyone" or "no one" are highly suspect, and the trend of code-assist -> agentic -> delegated / orchestration is just that (a trend), not a universal law. Even in a full-on maximalist yolo paperclip future, many experts will likely ALWAYS want access to a decent IDE. (Note I'm not saying Cursor is necessarily that IDE, and I'm not commenting on the valuation.)
OK, I'll concede that not everyone keeps getting pulled back to vim, the way I do. I simply don't like VS Code or its forks. I like Zed well enough, but I find I use it very rarely...two or three terminal tabs (Claude code, bash, and vim) is usually all I need, or tmux windows and/or panes if I'm working remotely, with Claude Code opened locally and configured to use tmux to talk to the remote system (using a wrapper I made to automate the setup: https://github.com/swelljoe/tandem).
But, even if you want a big all-in-one editor in an Electron app, it seems obvious VS Code is the way to go (or Zed, if you you aren't committed to using an Electron app). I just can't think of anything Cursor offers that makes it worth spending extra money for it.
Why would you even want a Claude subscription if not for Claude Code? Anthropic is obviously the best for programming, but probably nowhere else. Seems like a good way to onboard people to the Claude Code experience...everyone who's working seriously with it needs Opus, anyway. But, maybe that's the rub, if the Pro plan includes no Opus usage (which I think has always been the case), you might have a worse impression of Claude Code. Codex 5.4 is better than Sonnet, but not better than Opus.
I dunno, I'm no business genius, but I think we're starting to see these companies try to find ways to make money instead of losing it.
The pro plan does include Opus usage. I've noticed the limits on the web client are a bit higher than through CC, but probably more because of the increased token usage of agentic coding in general.
Claude web is actually pretty good for dealing with random projects outside of code. I have a Home Assistant MCP server [1] behind a Cloudflare tunnel exposed to it that makes maintaining automations a lot easier.
On LMArena, Claude Opus is ranked as the best at everything except image and video generation, which it does not support. That may be inaccurate, but it's plausible
People subscribed to chatgpt before there was codex. Why wouldnt a Claude subscription stand on its own without Claude Code? In fact it’s probably a smart move for Anthropic to split it out.
Because Claude Code provides me more value than simply a chat agent. Claude Code is the only thing I pay for. I can do chat things with the free tier across all the LLMs.
I have been using https://claude.ai and, initially, it was good, but, unfortunately, it keeps getting worse. I had it search for contact information for a certain public entity, and in Claude's response, all emails were being replaced with [email protected] or something like that. They also added an absolutely horrendous automatic markdown in the text input, so now you can't even properly enter your prompt. It actively gets in my way and prevents me from typing what I want. Fuck you Anthropic.
The metric they're selling this on is intelligence per byte, rather than total intelligence. So, if they used the quantized competing models, the intelligence per byte gap shrinks, because most models hold up very well down to 6-bit quantization, and 4-bit is usually still pretty good, though intelligence definitely tends to fall below 6-bit.
Nonetheless, the Prism Bonsai models are impressive for their size. Where it falls apart is with knowledge. It has good prose/logic for a tiny model, and it's fast even on modest hardware, but it hallucinates a lot. Which makes sense. You can't fit the world's data in a couple of gigabytes. But, as a base model for fine-tuning for use cases where size matters, it's probably a great choice.
That's the craziest shit I've ever heard.
reply