visarga

22 ms·

451.
by visarga 11mo ago
The problem is how we use it. A human sees not a photo but a video, and has long context before and after, not just that instance, we can also change position, a LLM can't do that at all.
452.
by visarga 11mo ago
Scaling the model size, the compute and the dataset hit a wall. Too large a model, or if it needs too much compute - it becomes too expensive to use. And the dataset .. we benefitted in one go from about multiple decades of content accumula
453.
by visarga 11mo ago
What matters is not who writes the words. The source of slop is competition for scarce attention between creatives, and retention drive for platforms. They optimize for slop, humans conform, AI is just a tool here. We are trying to solve an
454.
by visarga 11mo ago
It's not slop if it is inspired from good content. Basically you need to add your original spices into the soup to make it not slop, or have the LLM do deep research kind of work to contrast among hundreds of sources. Slop did not orig
455.
by visarga 11mo ago
Simple solution - run the same query on 3 different LLMs with different search integrations, if they concur chances of hallucination are low.
456.
by visarga 11mo ago
> the problem is when I copy paste the content and claim I wrote it Why is this the problem and not the reverse - using AI without adding anything original into the soup? I could paraphrase an AI response in my own words and it will be n
457.
by visarga 11mo ago
So we expect either 1. people using AI and copy pasting into the human-only network, or 2. other people claiming your text sounds like AI and ostracizing you for no good reason. It won't be a happy place - I know from anti-generative A
458.
by visarga 11mo ago
> a new human-only internet Only if those humans don't take their leads from AI. If they read AI and write, not much benefit.
459.
by visarga 11mo ago
Deep Research reports I think are above average internet quality, they collect hundreds of sources, synthesize and contrast them & provide backlinks. Almost like a generative wikipedia. I think all we can expect from internet informatio
460.
by visarga 11mo ago
Oh, if you finetune GPT-4 on an author it assumes the style so well that people prefer it to human experts doing the same job > "Readers Prefer Outputs of AI Trained on Copyrighted Books over Expert Human Writers"
461.
by visarga 11mo ago
That sidebar of past chats is where they go to be lost forever. Nobody came up with a UI that has decent search experience. It's like reddit internal search engine, but a bit worse.
462.
by visarga 11mo ago
But people post on social networks, blogs, newspapers and other widely read places, while LLMs post in chat rooms with 1 reader most of their outputs.
463.
by visarga 11mo ago
I don't see a big difference to humans, we are saying many unreasonable things too, validation is necessary. If you use internet, books or AI it is your job to test their validity. Anything can be bullshit, written by human or AI. In f
464.
by visarga 11mo ago
You don't use it that way. You use it to help you build and run experiments, and help you discuss your findings, and in the end helps you write your discoveries. You provide the content, and actual experiments provide the signal.
465.
by visarga 11mo ago
I think when it comes to LLMs, like software and books - usage is everything. You have to use it to get a benefit. LLMs by themselves produce no utility. And usage means a task coming from a person who puts something at stake, it is context
466.
by visarga 11mo ago
Wondering why a $4T company can't afford a smart installation assistant that can auto-detect problems and apply fixes as needed. I wasted too many days chasing driver and torch versions. It's probably the worst part of working in
467.
by visarga 11mo ago
Yes, this is correct, and it happens everywhere. App Store, Play Store, YouTube, Meta, X, Amazon and even Uber - they all play in two-sided markets exploiting both its users and providers at the same time.
468.
by visarga 11mo ago
> For example, I can't wrap my head around how a) a human could come up with a piece of writing that inarguably reads "novel" writing, while b) an AI could be guaranteed to not be able to do the same, under the same standa
469.
by visarga 11mo ago
You can select by source and audience, but not by topic. I found that a bit strange, topic would be the first thing we filter by. Why don't you dress up the summarization prompt with a task to generate tags?
470.
by visarga 11mo ago
Copyright kills works when rights cannot be negotiated, usually because the rights holder is not to be found, but in this case because the situation is just "complicated".
471.
by visarga 11mo ago
> I think nobody wants AI in Firefox I think the author does not speak for me
472.
by visarga 11mo ago
While Chinese models train on all Western cultural output, our own models are restricted. And in the corporate world the models of choice for finetuning are DeepSeek and Qwen, wonder why.
473.
by visarga 11mo ago
> Expanding copyright even more so that text / art that looks stylistically similar to another work is counted as infringing will, in the long run, give Disney's lawyers the power to punish folks for making content that even lo
474.
by visarga 11mo ago
Consider this - if I wanted to read A Game of Thrones, then I would read A Game of Thrones, not some bootleg LLM approximation. It is faster, more exact an cheaper to infringe by copying, a LLM is a terrible tool for infringement, it is slo
475.
by visarga 11mo ago
There is a huge point - those prompts have answers, followed by more prompts and answers. If you look at an AI answer in hindsight you can often spot if it was a good or bad response from the next messages. So you can derive a preference sc
476.
by visarga 11mo ago
You can use the LLM, but you don't also have the rest of the data they relied on. A LLM can generate everything if it starts from a minimal prompt, but this is a recipe for slop. If you come with materials, discuss them, their implicat
477.
by visarga 11mo ago
> You didn’t just start with five W’s, you went right for the most important one Also the "it's not A, it's B" template
478.
by visarga 11mo ago
Who hasn't posted a piece with "Reply" at the end?
479.
by visarga 11mo ago
> So, other companies innovated over the copyright holder's objections, and now we have to sort it out in the courts. I think they try to expand copyright from "protected expression" to "protected patterns and abstrac
480.
by visarga 11mo ago
> Great article, well written. >> The question is no longer whether a model can cover the job that was going to exist anyway. The question is whether a human can justify their presence next to a stack of models. >> The centra

More ›