lukev

32 ms·

571.
by lukev 3y ago
But how is the user supposed to know if their query is a weird complex edge case or not? The model certainly can't tell them.
572.
by lukev 3y ago
This is a great example of something GPT-4 gets confidently wrong, today. I just ran this query: Prompt: "The year is 894 AD. The capital of France is: Response: "In 894 AD, the capital of France was Paris." This is incorrect
573.
by lukev 3y ago
> none of this process is needed to answer the actual question which the LLM would know from its training data. I think this isn't true; even if the model has the answer stored implicitly in its weights, it has no way of "citin
574.
by lukev 3y ago
Yes, a bit, though an important feature here is it's still searching the underlying data sources (e.g. Google, Wikipedia, or others) and then using a LLM to summarize the results. The "natural language tokenizer" itself is of
575.
by lukev 3y ago
I wonder how much better an LLM could be given even better training data. For example, the total number of tokens contained in the physical and digital collections of a moderately-sized university library is (probably) equal to or on par wi
576.
by lukev 3y ago
I was being charitable and interpreting it as referring to any type of media usage. Agreed that the race to the bottom of low-quality stimulus content is a (probably separate) problem, since that trend was well underway long before generati
577.
by lukev 3y ago
I remain open to the question of how one should best engage with general intelligences or superintelligences (should they ever become reality.) I just meant to say I have no interest in "consuming content" from any generative AI t
578.
by lukev 3y ago
I still think it's unthinkable. "Consuming content" is fundamentally an act of communication. If there isn't another entity with intelligence and intent to communicate with, I'm not really interested. Although the
579.
by lukev 3y ago
Also, Musk yanking the plug on an active operation does not bode well for that sort of use case.
580.
by lukev 3y ago
What's the long-term reality catch-up mechanism in the social media space?
581.
by lukev 3y ago
> The internet is 99.99% scams and trash. Mostly because of Google.
582.
by lukev 3y ago
I'm curious about the terminology for the "functional representation" dataset. Is this a well-defined term? I've been thinking about similar approaches for getting more structured propositional knowledge into and out of
583.
by lukev 3y ago
Also tech, a good portion of the time.
584.
by lukev 3y ago
Having worked extensively in Java, Node, and Python, I’ll take the JVM ecosystem absolutely any day of the week. Me and my catheter will be over here delivering actual software while you figure out how React 32 broke your transcompiler.
585.
by lukev 3y ago
I'm not sure the definition of "intention" the article suggests is a useful one. He tries to make it sound like he's being conservative: > That is, we should ascribe intentions to a system if and only if it helps to p
586.
by lukev 3y ago
This doesn't make sense. The browser provides the user agent as a header in HTTP requests. They can't detect if or how the server is using that information. Or do you mean your friend's product is a browser plugin? In which
587.
by lukev 3y ago
If prices were perfectly liquid and symmetrical, sure. Unfortunately they're not and the system is pretty complex. There's a bunch of different ways to purchase energy and not all of them are sensitive to realtime price. Establish
588.
by lukev 3y ago
The original idea behind this is sound. Because production and consumption must net to ~0, from the grid's perspective cutting load is just as valuable as providing generative capacity. Imagine you own a factory that typically runs 24&
589.
by lukev 3y ago
I can empathize with this, however, I think it's worth pointing out that the value of therapy does not lie in the insights that are communicated. Even if they weren't in self-help books, you could in theory just go read the text
590.
by lukev 3y ago
This also doesn't account for non-safety related risks of self-driving cars, relative to human drivers. For example, it's quite common for an autonomous vehicle to slow down or stop in a way that disrupts traffic. This isn't
591.
by lukev 3y ago
This brings up a question I've had about OpenAI's fine-tuning APIs. The training format is a series of question/answer pairs: i.e, what we might think of as "supervised" learning. It can be challenging to build good
592.
by lukev 3y ago
The only way to be discerning with respect to most* crypto is not to participate. Even "succeeding" at crypto just means you sold early enough to a greater fool. * jury still out on the couple major cryptocurrencies, though I hope
593.
by lukev 3y ago
It's really not equivalent. The player does not issue the card.
594.
by lukev 3y ago
But what law is being broken, if NFTs aren't securities?
595.
by lukev 3y ago
But if it's not a security, is it fraud? If I sell you a hand drawn stick figure for $100, and you buy it because it's "art", then there's no rug to pull. You bought it, and it doesn't matter to the artist whet
596.
by lukev 3y ago
I agree that the dissenters bring up a lot of good questions that need answering. However, if you believe their fundamental argument that NFTs aren't really securities... what regulatory body should prevent this kind of behavior? Or
597.
by lukev 3y ago
The irony here is that you'll need to use (and regularly pay off) credit cards if you want to improve your credit.
598.
by lukev 3y ago
It only breaks the camera. The rest of the phone works fine. Plenty of market for that (albeit at a steep discount.) Also, it's a repairable issue. Happened to me and Apple fixed it quickly (covered by AppleCare+ in my case)
599.
by lukev 3y ago
https://www.trychroma.com It works as a library, you can require the PIP package and get started immediately.
600.
by lukev 3y ago
Hallucination, but also the ability to cite a source. A model using RAG can tell you why it answered a certain way, and cite chapter and verse from the underlying data. That's impossible just from model weights.

More ›