halflings

7 ms·

31.
by halflings 3y ago
Anyone from Scandinavia at least would be horrified by this story. It makes the US presidency look like a 3rd world dictatorship. If it's OK for a president to behave in this way, then you better believe this goes all the way down (and
32.
by halflings 3y ago
Growing pains. Remember how this all started (Google Magenta, circa 2019): https://www.youtube.com/watch?v=ZRnbbtqxBEc
33.
by halflings 3y ago
How about "AI will further democratize music"? Anyone can make music today, but ask any musician and they'll be mindblown by the idea that they can hum a melody and immediately generate a sax/percussion track matching it
34.
by halflings 3y ago
That's like killing a mosquito with a bazooka. Using GPT-4 vision on all users would be extremely expensive. Way simpler models can detect nudity. (and they do mention they had great success using those) And if the idea was to detect c
35.
by halflings 3y ago
It means "raspberry" in Arabic, and in Moroccan dialect at least it's slang for "good looking woman". Though it does sound like "tota", which in Moroccan dialect again is the childish synonym for penis :))
36.
by halflings 3y ago
Counterpoint: On iPhone, I use this feature all the time (mostly to set alarms tbh, but occasionally to start loading some music or else). There's a reason it's there and easily accessible. (and you can disable it from settings if
37.
by halflings 3y ago
Like you said, Bing has their own search index. Why isn't everyone using Bing?
38.
by halflings 3y ago
> Has the C++ standardization committee given a rationale why a module definition does not automatically also create a namespace with the same name? That would seem such a useful feature. I'm not sure what the official rationale is,
39.
by halflings 3y ago
But that is exactly what search engines do. Yet people find a way to game these "impossible to game" metrics. For example: "age of the article": search engines value recent content => suddenly you start seeing article
40.
by halflings 3y ago
What I mentioned doesn't depend on how LLMs work, the end result is the same (retrieving useful inputs to pass to your LLM). Just meant that a lot of people can just do this in-memory or in ad-hoc ways if they're not too latency c
41.
by halflings 3y ago
Minor note: you only need a vector database if you have so many possible inputs that linear retrieval is too slow. Arguably, for many use cases (e.g. searching through a document with ~200 passages), loading embeddings in memory and running
42.
by halflings 3y ago
Debatable. Unions can also be seen as there to benefit current union workers only. Milton Friedman is as libertarian as it gets, but he makes a good case about how unions only serve current employees, and have no incentive to see the busi
43.
by halflings 3y ago
> It led to a memorable and positive conversation. Was the conversation really positive? My first instinct was to think that this is incredibly awkward and harsh for the person that made the effort to buy you a gift; or to speak for myse
44.
by halflings 3y ago
I was able to run the 4bit quantized LLAMA2 7B on a 2070 Super, though latency was so-so. I was surprised by how fast it runs on an M2 MBP + llama.cpp; Way way faster than ChatGPT, and that's not even using the Apple neural engine.
45.
by halflings 3y ago
I don't think translation is a great use case for ChatGPT and LLAMA. These models are overwhelmingly trained on English, and LLAMA2 which should have more data from other languages is still focused on languages w/ Latin/Cyril
46.
by halflings 3y ago
> If NN inference were so fast, we would compile C programs with it instead of using deductive logical inference that is executed efficiently by the compiler. This is the definition of a strawman. Who is claiming that NN inference is alw
47.
by halflings 3y ago
This looks like a promotional comment to sell some kind of paid "AI Training" [1], doesn't address anything in the linked article. [1] https://max.io/teacher-training.html
48.
by halflings 3y ago
Extractive tasks are part of where LLMs shine, and where you get the least amount of hallucination as long as you fine-tune your model. By fine-tuning the model to extract a specific desired output from the text you give it, it learns that
49.
by halflings 3y ago
The interesting part is the sudden generalization. Simple models predicting simple things will generally slowly overfit, and regularization keeps that overfitting in check. This "grokking" phenomenon is when a model first starts b
50.
by halflings 3y ago
> That’s not exactly true [...] Instead, the brain actually works to actively distill knowledge that doesn’t need to be memorized verbatim into its essential components ...but that's exactly what OP said, no? I remember attending an
51.
by halflings 3y ago
You can check if translations are also better on Bard, that would partly answer that question I assume the low latency you get from Google Translate is not feasible with current LLMs like ChatGPT. Translate is used to translate sentences on
52.
by halflings 3y ago
Sure, then advocate for your government to build a public search engine. Why do you need to penalize private institutions if users prefer to use them over a government service?
53.
by halflings 3y ago
Banning anti-competitive behavior (which is what antitrust laws are for in the first place). For instance: a company actively blocking competition (e.g. coffee machine manufacturers that make it impossible for people to buy pods from anothe
54.
by halflings 3y ago
So the remaining traffic would be all expenses, no profit? Doesn't sound like capitalism anymore then, more of a public service.
55.
by halflings 3y ago
> Why isn't US govt doing this? Why isn't the UK, Sweden, Switzerland, Italy, ... doing this? Because it's a pretty strong form of controlled economy that distorts the market. How would this work for search engines? Would
56.
by halflings 3y ago
Yes, this looks like a (hopefully) better information retrieval system than running CTRL+F over a PDF. Nothing wrong with that :) information retrieval is hard. And CTRL+F won't render a table for queries like "what is the % incre
57.
by halflings 3y ago
> They also didn't come up with ChatGPT, arguably the first LLM good enough that can perform non trivial tasks of data recall and organization. Not only did Google make the main breakthroughs that all LLMs still rely on today (mainl
58.
by halflings 3y ago
> Hell, Netflix used to have an option where people could leave reviews on the movies they watched There's a very specific reason Netflix removed reviews / star ratings in favour of a simple thumbs up & down system: those w
59.
by halflings 3y ago
Yes it is. I had a phone accidentally fall in water 3-4 times, was pretty happy waterproof phones are a thing.
60.
by halflings 3y ago
It's not necessarily why they didn't reach out, but waiving certifications and other random credentials (e.g. completion of some Coursera course) is usually not the best strategy to get hired. Talking about concrete work you'

More ›