visarga

33 ms·

571.
by visarga 1y ago
> Similarly I get annoyed, if every pixel is some clickable action trigger. Twitter did that, every pixel was reacting to clicks. Selecting text was hard
572.
by visarga 1y ago
An hour of Netflix streaming consumes approximately 77 Wh according to IEA analysis showing streaming a Netflix video in 2019 typically consumed around 0.077 kWh of electricity per hour [1], while an hour of active GPT-4 chatting (assuming
573.
by visarga 1y ago
By the same logic, people working in transportation around the time Ford Model T was introduced did NOT diminish 100 years later. We went from about 3.2 million in 1910 (~8% of the workforce) to 6–16 million in 2023 (~4–10%, depending on de
574.
by visarga 1y ago
I did a little investigation. Turns out that GPT-4's training consumes as much energy as 300 cars in their lifetime, which comes about 50 GWh. Not really that much, could be just families on a short street burning that kind of energy.
575.
by visarga 1y ago
The article is about SROs not houses with 12 people.
576.
by visarga 1y ago
I think you just push the problem away, now you got homeless people.
577.
by visarga 1y ago
> I’m not sure what the article is complaining about. It's right there in the article: "And as SROs disappeared, homelessness—which had been rare from at least the end of the Great Depression to the late 1970s—exploded nationwi
578.
by visarga 1y ago
I was hoping to find some insights about why performance drops when using actual structured outputs. It's been a known problem. For example this paper "Let Me Speak Freely? A Study on the Impact of Format Restrictions on Performan
579.
by visarga 1y ago
> Cloudflare is leveraging its CDN dominance to become a kind of payment processor for the internet Seems like the moment is ripe for this move. In recent news Google partnered with stablecoins and traditional payment processors to creat
580.
by visarga 1y ago
They are all platforms: news companies, publishing houses, ISPs, recording labels, YouTube, Google, Meta, TikTok, Windows, App Store, Play Store, Amazon.. and as platforms they squeeze both the layer above and the one below, their customers
581.
by visarga 1y ago
Say AI company A pays and gets access, trains a model-A. Now come AI company B, C, D.. and they train on synthetic data from model A. Do they need to license the original content anymore? (morally and practically)
582.
by visarga 1y ago
It's because batch size is dynamic. So a different batch size will change the output even on temp 0.
583.
by visarga 1y ago
I think extraction from the model itself is a bad idea. But extraction from external sources, such as the deep research reports LLMs generate, or solving problems where we have validation of correctness is a good idea. The model is not vali
584.
by visarga 1y ago
Ok, so it just means "worded with AI", no relation to how much thinking went in
585.
by visarga 1y ago
> knowing that all output is statistical garbage "Bare" LLMs are rare today. Almost all of them are hooked to search engines, APIs, and code execution. So "closed book" fact retention is not an issue, we're not e
586.
by visarga 1y ago
What is "AI slop?" I could have a 2 hour long discussion with Claude and in the end have it formalize it as an article. Is that AI slop?
587.
by visarga 1y ago
> We're not just experiencing technological change. We're watching the basic economics of software development transform in real time. > We're not just <A> ... We're <B> Is this proof of LLM writing or
588.
by visarga 1y ago
You have no idea how much personal work went behind it. You just suspect it was worded with a LLM. I have been using embeddings for almost a decade and am well versed with their intricacies. I think this article has merit. The direction of
589.
by visarga 1y ago
> but have no architectural mechanism to separate facts from expressions Sure they do. Every time a bot searches, reads your site and formulates an answer it does not replicate your expression. First of all, it compares across 20.. 100 s
590.
by visarga 1y ago
I see this step progression: 1. people having real problems, like employment, housing, health, or education access 2. they go online (or watch TV) finding all sorts of extreme takes and biases; these theories provide simple explanations and
591.
by visarga 1y ago
Books also make us less capable at rote memorization. People used to do much more memorization. Search engines taught us to remember the keywords, not the facts. Calculators made us rarely do mental calculations. This is what happens - prog
592.
by visarga 1y ago
AI is used by students, teachers, researchers, software developers, marketers and other categories and the adoption rates are close to 90%. Even if it does not make us more productive we still like using it daily. But when used right, it do
593.
by visarga 1y ago
We know in the same sense we understand the rules in Conway's game of life - at low level - but don't understand what those rules will produce at high level (gliders, guns) except by executing and seeing. Analogous to knowing what
594.
by visarga 1y ago
Attention (technology) is all they need (to keep secret).
595.
by visarga 1y ago
Every decision they take based on evals influences the model.
596.
by visarga 1y ago
> but you still have no way of knowing whether A) it's the string of symbols you wanted Time will tell. You take that output, use it, and see the outcomes. That's exactly how brains work too. We don't spark knowledge from
597.
by visarga 1y ago
> That they're clearly optimizing for people's attention is more worrisome. Running LLMs is expensive and we can swap models easily. The fight for attention is on, it acts like an evolutionary pressure on LLMs. We already had t
598.
by visarga 1y ago
I think the Stochastic Parrots idea is pretty outdated and incorrect. LLMs are not parrots, we don't even need them to parrot, we already have perfect copying machines. LLMs are working on new things, that is their purpose, reproducing
599.
by visarga 1y ago
It's because they don't allow GPT in exams. Students are accustomed to using it.
600.
by visarga 1y ago
> That’s the causal link. But copyright was based on substantial similarity, not causal links. That is the subtle change. Copyright is expanding more and more. In my view, unless there is substantially similarity to the infringed work,

More ›