zone411

25 ms·

301.
by zone411 4y ago
If you don't learn anything from reading, I think you're in the minority. I agree that active doing, writing, thinking are often better but those aren't the only ways people learn.
302.
by zone411 4y ago
If you want to run it locally, the Automatic1111 repo that was mentioned does have most up-to-date models, but it's also messy and breaks often. But is your GPU up to the task?
303.
by zone411 4y ago
You could first see if you like it using https://huggingface.co/spaces/hysts/ControlNet . I quickly made these https://twitter.com/LechMazur/status/1626668677473918981
304.
by zone411 4y ago
Note that this article is from four months ago. Their Patreon and Kickstarter have been suspended but they collected money on their website and they will first focus on making yet another anime model. The big news in the Stable Diffusion la
305.
by zone411 4y ago
This article seems to be targeted towards beginners. From the title, I was expecting it to cover something like https://github.com/joboccara/NamedType , which is a technique I would recommend.
306.
by zone411 4y ago
Does anyone have experience running Visual Studio (not VSCode) in Parallels? I'd like to update my older MacBook Pro with Boot Camp but I'm not leaving Windows Visual Studio for a poor native Mac version.
307.
by zone411 4y ago
This specific example already feels a bit outdated because Bing can use web searches and pull the titles of actual papers (I tried it and it works well). Improving factual consistency of LLMs is an active research area and we should see som
308.
by zone411 4y ago
I can recommend "The Man Who Solved the Market: How Jim Simons Launched the Quant Revolution" by Gregory Zuckerman, but of course don't expect to learn any market-beating strategies from it. Quanta Magazine, a source for arti
309.
by zone411 4y ago
You can use the generated text for further training if you have a human curator who determines its quality. I've been training my model that helps generating melodies using some of the melodies I have created with it.
310.
by zone411 4y ago
Exactly, I made the same point before. Google's ads will become more effective AND people will spend more time on Google's properties. This narrative that Google will lose out because people won't view ads never made any sens
311.
by zone411 4y ago
They pretty much admitted that they did fumble the ball: https://www.nytimes.com/2022/12/21/technology/ai-chatgpt-goo... . But I agree that Google's research is #1 and they will come back strong.
312.
by zone411 4y ago
Google definitely made a strategic mistake by being slow to react and only taking action late. But Bing only has a limited period of time before Google catches up, and it doesn't seem like it will be enough to take away significant mar
313.
by zone411 4y ago
If you view the source of any Wikipedia page, they purposefully include "nofollow" tags, so Google ignores these links!
314.
by zone411 4y ago
I mean my sites are cited over a thousand of times, but it only makes sense: a tiny percentage of visitors who view the Wikipedia page even reach the bottom of the page and then click on one of the links. And there is no benefit in terms of
315.
by zone411 4y ago
Yes, it's pretty much game over for free-to-access factual content sites. I've been focusing heavily on AI in recent years, so I saw it coming. It's been a death by a thousand cuts, with Google incorporating long snippets, et
316.
by zone411 4y ago
Look at my profile, I own content-heavy sites and have for many years. I can show you logs - Wikipedia does virtually nothing. And the content of my sites has been regurgitated by Wikipedia thousands of times.
317.
by zone411 4y ago
...which do nothing for the websites that had the original content.
318.
by zone411 4y ago
No. In addition to what the sibling posts cited, this https://651d7eef-05d1-4785-8f04-93b49cc8d71f.filesusr.com/ug... says: >...ways of using the statements by reading them _prior_ to any other components of the applicat
319.
by zone411 4y ago
They only found images that had been replicated 100+ times in Stable Diffusion's training set. Stable Diffusion did a poor job deduplicating.
320.
by zone411 4y ago
You might be interested in the crowdsourced prediction which currently stands at 2039: https://www.metaculus.com/questions/5121/date-of-artificial-... . My personal shorter-term prediction, based on my experience a
321.
by zone411 4y ago
The abstract mentions fine-tuning, not full pre-training?
322.
by zone411 4y ago
Long story short, training requires intensive device-to-device communication. Distributed training is possible in theory but so inefficient that it's not worth it. Here is a new paper that looks to be the most promising approach yet:
323.
by zone411 4y ago
I looked up if there are any studies about this. I found this one from 2014: https://journals.sagepub.com/doi/abs/10.1177/109861111452204... . Paywalled but referenced elsewhere as "more in-service traini
324.
by zone411 4y ago
> Music gen is hard because music is inherently a composed of many many different instruments, each with unique sound and function. Simply training end-to-end will almost always end up badly. Yes! That's what I figured when I starte
325.
by zone411 4y ago
Google is a public company so its shareholders want the stock price to be higher ASAP. Being an unquestionable leader in AI with a product as good as ChatGPT (and maybe an image model available to the public) would help in that goal.
326.
by zone411 4y ago
I think the opposite: Google will do great. Google+Deepmind's ML research groups are leading everyone. While it's true that they are late in getting something like ChatGPT to market, I think it was a wakeup call and it will be goo
327.
by zone411 4y ago
...and the ability to synthetize both into a response to a question it has never encountered before
328.
by zone411 4y ago
The author said that he comes up with new questions every year.
329.
by zone411 4y ago
Oh how quickly the goalposts shift... If the test was done correctly and there was no data leakage, this absolutely does not compare to someone equipped with the answer key.
330.
by zone411 4y ago
You wouldn't sell the model, you would sell its outputs. I can legally buy products made by factories in Bangladesh that would violate all sorts of laws if they were operating in the U.S. It will be for international trade agreements t

More ›