thorum

22 ms·

331.
GitHub is removing the Trending Repositories page
75 points by thorum 4y ago|33 comments
332.
by thorum 4y ago
GitHub Copilot is built on GPT3.
333.
by thorum 4y ago
I'm sure those are factors, but it's mainly (1) GPU prices have dropped significantly, and (2) a lot of progress is happening in optimizing large language models to be more efficient while maintaining performance.
334.
OpenAI API pricing update FAQ(help.openai.com)
87 points by thorum 4y ago|193 comments
335.
by thorum 4y ago
Does anyone know if Azure’s OpenAI platform was down too?
336.
by thorum 4y ago
DALLE can’t render fine detail at a distance of more than a meter or so. Possibly because OpenAI limits how many diffusion steps are used for each image.
337.
by thorum 4y ago
One cool thing about Stable Diffusion (the current leading competitor to DALLE) is that you can create images in a similar style by configuring the seed number. Different prompts with the same seed number can be used to create different ima
338.
by thorum 4y ago
If you want to see more examples of what this AI is capable of, check out the subreddit: https://reddit.com/r/stablediffusion
339.
by thorum 4y ago
Stable Diffusion needs more detailed prompt engineering to get good results than DALLE. But it can also handle longer and more detailed prompts. So asking for people “with two legs” and “symmetrical eyes” actually works (to some extent).
340.
by thorum 4y ago
Where does Stability AI gets its funding? 70+ employees, massive GPU compute resources, and the first commercial product isn’t released yet?
341.
by thorum 4y ago
Until you consider the level of demand for this product, which is surely higher than OpenAI can scale to with the number of GPUs they have. If they price it lower they’ll be overwhelmed.
342.
by thorum 4y ago
DALLE images are still only 1024 px wide. Which has its uses, but I don’t think the stock photo industry is in real danger until someone figures out a better AI superresolution system that can produce larger and more detailed images.
343.
by thorum 4y ago
So this is a tool for developing and experimenting with quantum algorithms without needing access to an actual quantum computer. It runs in Colab and simulates the expected results on a regular computer. I don’t know whether there are other
344.
by thorum 4y ago
Yes, but they’re still averaging the results in the “it worked” category. Hypothetically, if it worked very well for 5 users and only a little for 95 users, an average score over all 100 will make it appear to not work for anyone. I’m not g
345.
by thorum 4y ago
I think the best approach would be a graph showing the distribution of users across different WPM differences, and if possible a second experiment with two non-BR fonts (to see how much random variation is expected). The positive results mi
346.
by thorum 4y ago
> Let's look at how many participants read faster with each font and their average speed gains. “Average speed gains” doesn’t seem like the best metric here. A graph showing the distribution of users across different WPM differences
347.
by thorum 4y ago
The article’s conclusion doesn’t seem to match the data they collected. They found that most people (52%) do read faster with bionic reading, the effect is generally quite small, but large for some minority of users - at least one user re
348.
by thorum 4y ago
Toasters are probably not a great example, but concept art is a big deal in most product development / creative industries. Tech, cars, Hollywood, fashion, anime, etc. The process usually involves finding and hiring an artist and worki
349.
by thorum 4y ago
OpenAI published an article about some of the things people are using DALL-E for: https://openai.com/blog/dall-e-2-extending-creativity/ I thought this quote was particularly insightful: > “Conceptualizing one’
350.
With Rising Book Bans, Librarians Have Come Under Attack(nytimes.com)
20 points by thorum 4y ago|4 comments
351.
by thorum 4y ago
The SQLite CoC, for anyone who's interested: https://sqlite.org/codeofethics.html
352.
by thorum 4y ago
The problem with GPT-3 is it’s too agreeable. It will agree to anything and promise anything - which is not a good idea in this context, to put it mildly.
353.
by thorum 4y ago
> The last thing I'm ever gonna do is feed any of my source code into something they've built. These are open source AI models which can be downloaded and used without requiring your source code to interact with Salesforce the
354.
Google AGI GitHub Repository(github.com)
1 points by thorum 4y ago|0 comments
355.
Cheater Latents(nonint.com)
1 points by thorum 4y ago|0 comments
356.
by thorum 4y ago
FWIW, Lambda has added support for Docker images, which can run anything. They’re not limited to 500 MB either. You just have to use one of Amazon’s base images that come with the Lambda runtime installed, or add it to your existing Dockerf
357.
by thorum 4y ago
Aside from voice quality (Tortoise is undefeated) and speed (Tortoise is vastly slower), there is a difference in consistency and control. Tortoise is a probabilistic system that uses one model to randomly generate 80+ different possible ou
358.
AI winter is well on its way (2018)(blog.piekniewski.info)
2 points by thorum 4y ago|0 comments
359.
by thorum 4y ago
> removed the line about replying Unknown to trick questions Did you need to remove this because it identified your question as a trick question? If so, doesn’t that indicate that the model does in fact know the difference?
360.
by thorum 4y ago
Lambda is great, if you can afford it: https://lambdalabs.com/service/gpu-cloud/pricing

More ›