Jack000

27 ms·

1.
by Jack000 7mo ago
None of my relatives who were alive in 1969 have ever heard of the US moon landings. The information was totally suppressed at the time and in the intervening decades they simply had no reason to seek it out. The younger generations think i
2.
by Jack000 2y ago
None of the things you listed mean much when it comes to ethno-nationalism. There are a lot of Chinese that love American brands but still hate the US for nationalist reasons. Among my relatives I would say all are anti-US. About 5-6 of t
3.
by Jack000 2y ago
It's heavy and the extra encumbrance is not great for general purpose computing, though I never felt the need for extra screen real estate while I'm coding, watching youtube, sending emails etc. That said I still use my AVP regula
4.
by Jack000 2y ago
This is kind of the visual equivalent of asking an LLM to count letters. The failure is more related to the tokenization scheme than the underlying quality of the model. I'm not certain about the specific models tested, but some VLMs j
5.
by Jack000 3y ago
Chess is a great example because it highlights the subtle difference between LLMs and agents. What GPT3.5 does is not quite playing chess but creating realistic chess moves that a human might make. An LLM could play chess though, all it n
6.
by Jack000 3y ago
I think a lot of the confusion on whether LLMs can think stems from the fact that LLMs are purely models of language and solve intelligence as a kind of accidental side-effect. The real problem that an LLM is trying to solve is to create a
7.
by Jack000 3y ago
The thing is, it's not gibberish. A sufficiently small language model can be understood by humans: https://twitter.com/karpathy/status/1645115622517542913 The explanation is perfectly sensical, just too compl
8.
by Jack000 3y ago
I was writing an SVG editor web app but didn't know anything about the fabricjs api. Here's a portion of my prompts with chatgpt (april 2023 version, the date at the top is the share date). The initial code came from a different c
9.
by Jack000 3y ago
In pixel space a convnet uses pixel-wise convolutions and a pixel-kernel. If you represent a vector image as a polygon, the direct equivalent to a convolution would be the Minkowski sum of the vector image and a polygon-kernel. You could st
10.
by Jack000 3y ago
I think this approach isn't ideal because you're representing pixels as 150x150 unique bins. With only 71k fonts it's likely a lot of these bins are never used, especially at the corners. Since you're quantizing anyways,
11.
by Jack000 3y ago
> what will LLMs ever do for us? Hallucinations are an engineering problem and can be solved. Compute per dollar is still growing exponentially. Eventually this technology will be widely proliferated and cheap to operate.
12.
by Jack000 3y ago
I didn't say they would be better. Most people on earth lack any access to healthcare at all. https://shorturl.at/joA23
13.
by Jack000 3y ago
A doctor for every person, a teacher for every child, available any time and for free.
14.
by Jack000 3y ago
As Yann Lecun said, “in high dimension, there is no such thing as interpolation. In high dimension, everything is extrapolation.” It’s trivial to come up with a prompt that doesn’t exist in the dataset. To generalize, the model cannot memo
15.
by Jack000 3y ago
The foreign buyers tax, the empty home tax, and now the total ban on foreign buyers introduced this year has had zero effect on home prices. It's time to stop blaming foreigners. The solution is to build more and more quickly.
16.
by Jack000 3y ago
oxycotton.com and oxyloo.com are taken unfortunately washluv.com is still available though
17.
by Jack000 3y ago
There are still some good 6 letter names out there, if you're willing to use a brandable (made-up) word. Some I randomly made just now: pletha.com encryx.com oxiply.com
18.
by Jack000 3y ago
double descent is a different phenomenon from grokking
19.
by Jack000 3y ago
Not seeing anything about the dataset, are they still using LAION? There's no mention of LAION in the paper and the results look quite different from 1.5 so I'm guessing no. > the model may encounter challenges when synthesizin
20.
by Jack000 3y ago
I had a tough time with "Man eating shark" Why would a man eat a shark? Shouldn't it be the other way around?
21.
by Jack000 3y ago
This conflict only exists if you believe in a single, inviolable self. Which cell is the “real” one when it divides? Which branch of a tree is the true tree? Both versions of you have equal claim to the past, as long as the copy is truly id
22.
by Jack000 3y ago
no it's definitely GPT4. I'm not sure why the share page says "Model: Default" but it's "Model: GPT-4" in my webui. Seems like a bug in the share feature
23.
by Jack000 3y ago
I tried to replicate a few of my chats (the displayed date is incorrect, it seems to be the publish date instead of the original chat date): svg editor: early april: https://chat.openai.com/share/c235b48e-5a0e-4a89-af1c
24.
by Jack000 3y ago
Here's roughly how my conversation with GPT4 went Me: I want to make an svg editor, give me some suggestions. I mainly want something with mobile support. GPT4: gives me some options I look over the options and choose fabricjs Me: Star
25.
by Jack000 3y ago
Humans use language for communicating ideas, which is very different from language modeling. Here's some discussion on this topic: https://www.alignmentforum.org/posts/htrZrxduciZ5QaCjw/langu... I would add t
26.
by Jack000 3y ago
There’s too much focus on AGI. Language models do not emulate human minds - they are models of language. The emergent behavior from these models are only a side effect of their main training task, which is to build a model of all meaningful
27.
by Jack000 3y ago
This type of data is actually better than independent human text (specifically for training the LLM that originally produced the output) GPT4 is trained with PPO+RLHF. The web text that is produced by the LLM then fed back in will be more p
28.
by Jack000 3y ago
LLM failure modes are caused by the lack of external context - they perform poorly on visual tasks because they have no sense of vision for example. Hallucinations are another aspect of this - as embodied agents, humans and animals have a s
29.
by Jack000 3y ago
the point is that you have to want to in the first place. This is kind of like looking at “the flame” or some other modern art and saying “yeah that’s easy, I could have done that.” It’s missing the point entirely.
30.
by Jack000 3y ago
Actually I don't think most programmers are capable of designing something like spacex.com - every image on the site fits the minimalist theme (and there are hundreds). The photos are always dark in the appropriate areas so the text is

More ›