alyxya

8 ms·

91.
by alyxya 1y ago
No that isn’t throwing out data. Activation functions perform a nonlinear transformation to increase the expressivity of a function. If you did two matrix multiplications without ReLU in between, your function contains less information than
92.
by alyxya 1y ago
Precision actually matters a decent amount in LLMs. Quantization is used strategically in places that’ll minimize performance degradation, and models are smart enough so some loss in performance still gives a good model. I’m skeptical how w
93.
by alyxya 1y ago
This seems to be the page that describes the low level details of what the hardware aims to do. https://extropic.ai/writing/tsu-101-an-entirely-new-type-of-... To me, the biggest limitation is that you’d need an entire
94.
by alyxya 1y ago
Deterring kids is still valuable. A lot of them won’t bother with figuring out how to get around the automated detection.
95.
by alyxya 1y ago
I worry that all this does is reduce future liability issues and make those users use another chatbot instead. I trust character.ai more than other chatbots for safety. > [Dr. Nina Vasan] said the company should work with child psycholog
96.
by alyxya 1y ago
I really like the homepage. It’s clear what the product does just after a few seconds and looks nice and smooth :) I’m curious if the annotation handles things like scrolling within a webpage. That would be a really cool feature if annotati
97.
by alyxya 1y ago
There’s two different kinds of users, on one side people are more hands off and want the model to autonomously handle longer tasks on its own with minimal guidance, and on the other side is users who want to interactively collaborate with t
98.
by alyxya 1y ago
Is the new model trained from scratch? What training data went into it?
99.
by alyxya 1y ago
I’ve generally seen providers allow you to opt in or out. What may vary is what the default is and what they may offer in exchange for using your data (perhaps they could offer higher rate limits).
100.
by alyxya 1y ago
I wonder if this custom model is trained on cursor users. There’s a lot of potential on how much better a custom model could be the closer it is integrated with the product. Having the model learn to adapt to different user preferences woul
101.
by alyxya 1y ago
With AI making it possible to use natural language to modify code, bare metal can make things easier to use with your own code and customization. Abstractions tend to be harder to reason about and have more limited functionality in exchange
102.
by alyxya 1y ago
I imagine youtube has the same automated system comply with eu, but the videos linked in the reddit post mention how quickly the appeals were denied, so accessing and contesting decisions alone doesn’t fix the automated system problem.
103.
by alyxya 1y ago
The main issue is how opaque the system is, though that may also be a feature to limit people’s abilities in figuring out how to bypass various restrictions. No one should have to risk the existence of their channel and livelihood to some u
104.
by alyxya 1y ago
I think the solution to continual learning is as simple as using context distillation. We know that models are good at in-context learning, so we just want an efficient way to distill context into the weights. I suspect context rot may come
105.
by alyxya 1y ago
Part of the concern I have is that OpenAI is contributing to these issues implicitly by helping companies automate away jobs. Maybe in the long term, society will adapt and continue to function, but many people will struggle to get by, and
106.
by alyxya 1y ago
I think of technology as offering a sliding scale for how much assistance it can provide. Your words could be literally the keys you press, or you could use some tool that fixes punctuation and spelling, or something that fixes the grammar
107.
by alyxya 1y ago
AI assisted blog posts could have an interleaved mix of AI and human written words where a person could edit the LLM’s output. If the whole blog post were simply a few prompts on ChatGPT with no human directly touching the output, then sure
108.
by alyxya 1y ago
I hadn't heard of Loudoun County before. I also did read the article and here's a sentence from it. > But while most locals the BBC spoke to opposed the data centres, the industry has many powerful proponents, including US Pres
109.
by alyxya 1y ago
It's sad seeing people impacted by these data centers considering how little voice they have. I don't think anyone wants to live near a data center unless they actually worked there.
110.
by alyxya 1y ago
That’s true, I just wanted to offer a counter perspective to the anti-AI sentiment in the blog post. I agree that the slop issue is probably more common and egregious, but it’s unhelpful to discount all AI assisted writing because of slop.
111.
by alyxya 1y ago
I try to make my best judgment regarding what to trust. It isn’t guaranteed that content written by humans is necessarily correct either. The nice thing about ChatGPT is that I can ask for sources, and sometimes I can rely on that source to
112.
by alyxya 1y ago
I personally don’t think I care if a blog post is AI generated or not. The only thing that matters to me is the content. I use ChatGPT to learn about a variety of different things, so if someone came up with an interesting set of prompts an
113.
by alyxya 1y ago
Intelligence is whatever we consider ourselves capable of. It turns out that computers are increasingly able to do whatever we can do. Maybe the only thing we can do is advanced pattern matching, but we didn't think of our intelligence
114.
by alyxya 1y ago
I didn't really convey my thoughts very well. I think of the actual valuable "more efficient ways of training" to be paradigm shifts between things like pretraining for learning raw knowledge, fine-tuning for making a model b
115.
by alyxya 1y ago
I think it should be able to learn multiplication with chain of thought. Without it, it's probably really difficult to generalize the multiplication of two n-digit integers when you have to accumulate up to n products of digits and han
116.
by alyxya 1y ago
I think people care too much about trying to innovate a new model architecture. Models are meant to create a compressed representation of its training data. Even if you came up with a more efficient compression, the capabilities of the mode
117.
by alyxya 1y ago
It’s a cool idea, but I don’t know how much people care about a photo provably being real. I take pictures with my phone because it’s simple and convenient. I get the vibe that it’s kinda like NFTs, where maybe some people would care if cer
118.
by alyxya 1y ago
I dislike the framing around IQ. It feels like IQ here just refers to how advanced a country's education system is. The real missed opportunity is not creating the right incentives for learning.
119.
by alyxya 1y ago
IMO the biggest limitation is that there's a lack of feedback in being a good interviewer. It's hard to get better at something you can't measure how well you did in any meaningful way.
120.
by alyxya 1y ago
It's possible algorithms simply drive up prices because price isn't the main factor some people use in deciding what to buy. Algorithms are probably able to learn that raising the price outweighs the decrease in the number of peop

More ›