xcv123

32 ms·

481.
by xcv123 3y ago
So what? Review and test it before running in production. Humans also screw things up on first attempts. Test it, check the output, fix it, try again. In my experience with GPT-4 if you describe a bug it will correct it.
482.
by xcv123 3y ago
English is not his native language
483.
by xcv123 3y ago
Shorter but missing an important detail. The article explains that every 8th bit is lost to allow for control signalling that steals the 8th LSB on every 6th frame. Hence 7/8 * 64 = 56. https://en.wikipedia.org/wiki&#x2
484.
by xcv123 3y ago
I just found this list of papers. https://github.com/AGI-Edgerunners/LLM-Planning-Papers Here's one titled "Reasoning with Language Model is Planning with World Model" https://arxiv.org/a
485.
by xcv123 3y ago
I was responding to this. > The context window only has to do with the size of input it has access to - its not related to what it's outputting The sequential token generation process is closely related to the content of the context
486.
by xcv123 3y ago
> If you ask it a question where the training data (or input data = context) either didn't include the answer, or where it was not obvious how to get the right answer, that will not (unfortunately) stop it from confidently answering
487.
by xcv123 3y ago
> The context window only has to do with the size of input it has access to - its not related to what it's outputting, which is ultimately constrained by what it was trained on. Wait a minute. You are completely missing the entire &
488.
by xcv123 3y ago
> Was it? I've only heard of pre-training (predict next word) and subsequent RLHF + SFT "alignment" (incl. aligning to goal of being conversational). AFAIK the NLP skills that these LLMs achieve are all emergent rather tha
489.
by xcv123 3y ago
> even if it seems to understand what you are asking (after all, it was trained with the goal of sounding good It was trained not only to "sound good" aesthetically but also to solve a wide range of NLP tasks accurately. It not
490.
by xcv123 3y ago
An example of an important human task that is not "economically valuable" in this sense, is caregiving within a family, such as parenting, since parents are not employed in that role. OpenAI is not setting the goal post that far,
491.
by xcv123 3y ago
Yes that's a good point. But the algorithms are improving too.
492.
by xcv123 3y ago
> They aren’t really designed to do anything actually. LLMs are models of human languages - it’s literally in the name, Large Language Model . No. And the article you linked to does not say that (because Wolfram is not an idiot). Transfo
493.
by xcv123 3y ago
Yes you're right I only mentioned the size of the model. The rate of progress has been astonishing and we haven't reached the end, in terms of both of size and algorithmic sophistication of the models. There is no evidence that we
494.
by xcv123 3y ago
> Set temperature to 0, and the LLM will talk in circles and not really say anything interesting. Set it too high and it will output nonsense. Sounds like some people I know, at both extremes. > The whole design of LLMs don’t seem ver
495.
by xcv123 3y ago
> it’ll just stop being a stochastic parrot in just a few more million parameters. Is is not a stochastic parrot today. Deep learning models can solve problems, recognize patterns, and generate new creative output that is not explicitly
496.
by xcv123 3y ago
I know that, but deep learning is more than LLMs. Transformers aren't the final ultimate stage of deep learning. We haven't found the limit yet.
497.
by xcv123 3y ago
> Just give the approach more time, it shall continue to incrementally improve until it finally works someday, cuz reasons Yes. Because we haven't yet reached the limit of deep learning models. GPT-3.5 has 175 billion parameters. GP
498.
by xcv123 3y ago
You missed the point. Deep learning models are in the early stages of development. With recent advancements they can already outperform humans at many tasks that were considered to require AGI level machine intelligence just a few years ago
499.
by xcv123 3y ago
You're right, it's good value for what it is, not a piece of junk. But it doesn't compete in the same segment.
500.
by xcv123 3y ago
> I feel very comfortable saying, as a mathematician, that the ability to solve grade school maths problems would not be at all a predictor of ability to solve real mathematical problems at a research level. At some point in the past, yo
501.
by xcv123 3y ago
No. That's cheap plastic gamer junk. Those are toys. Different market segment. Not comparable. MacBook Pro competes with high end professional laptops, not consumer gaming laptops. Dell Precision with 4k screen is what you would compar
502.
by xcv123 3y ago
> I think people forget that $4k+ for a computer is Quadro workstation territory. M3 MacBook Pro 16" is $2499, roughly the same as an equivalent Dell XPS / Precision laptop (for the same price Dell typically has half the screen
503.
by xcv123 3y ago
> You'd think the latter would imply the former, but maybe not... I wouldn't put C++ devs on too high of a pedestal. I got away with writing shitty C++ code for years before I really knew what I was doing. It still worked thoug
504.
by xcv123 3y ago
> “Deep understanding of what is a computer, what is computer software, and how the two relate.” Seems like a joke, but many developers do not really understand what's going on behind the scenes. This gets straight to the point. The
505.
by xcv123 3y ago
> So far I haven't found anything near the quality of GPT-4 GPT-4 has an estimated 1.8 trillion parameters. Orders of magnitude beyond open source models and ~10x GPT-3.5 which has 175 billion parameters. https://the-deco
506.
by xcv123 3y ago
For the entertainment value
507.
by xcv123 3y ago
Narrow to a tiny training set? What are you talking about now? That has nothing to do with deep learning. GPT-3.5 was trained on at least 300 billion tokens. It has 96 layers in its neural network of 175 billion parameters. Each one of thos
508.
by xcv123 3y ago
LLMs can outperform humans on a variety of NLP tasks that require understanding. Formally, they are designed to solve "natural language understanding" tasks as a subset of "natural language processing" tasks. The word &q
509.
by xcv123 3y ago
> It is pretty hard to argue against that. If they are a monopoly or duopoly or cartel it is very easy to argue against that. https://www.investopedia.com/terms/a/antitrust.asp
510.
by xcv123 3y ago
And Xen was my least favourite part of Black Mesa by far. To each their own.

More ›