thorum

30 ms·

361.
Flexible diffusion modeling of long videos(plai.cs.ubc.ca)
65 points by thorum 4y ago|16 comments
362.
by thorum 4y ago
Does anyone actually want porn of fake, AI generated “people”? Seems like most of the demand would be deepfakes of real people, which is both highly unethical and a good way to get your business sued out of existence.
363.
by thorum 4y ago
The developer behind Tortoise is experimenting with using diffusion for music generation: https://nonint.com/2022/05/04/friends-dont-let-friends-train...
364.
by thorum 4y ago
I suspect the solution to keeping a long story coherent is using the model at different levels of abstraction. A human writer doesn’t sit down and write a complete novel in one sitting. They go through a process of planning, character devel
365.
by thorum 4y ago
Imagen seems better at capturing details/nuance from the prompt, but subjectively the DALLE-2 images feel more “real” to me. Not sure why. Something about the lighting?
366.
by thorum 4y ago
That only lasts until the community copies the paper and catches up. For example the open source DALLE-2 implementation is coming along great: https://github.com/lucidrains/DALLE2-pytorch
367.
by thorum 4y ago
I read it as a comment on the UX / developer experience, which can superior with Step Functions vs competition regardless of whether Step Functions is an appropriate (or even physically possible) option for non-AWS projects.
368.
by thorum 4y ago
The lack of existing research to explain (or contradict) people’s reported experience is what makes projects like this so interesting, and worth devoting resources to researching further.
369.
by thorum 4y ago
From comments on HN, Reddit and Twitter, this approach clearly does help many people read faster and with less visual strain - just not everyone. Others say they find it harder to read. I looked at a couple of the iPhone apps that integrate
370.
by thorum 4y ago
I wonder if the effectiveness of bionic reading would be improved by using a gradient of weights for each character - instead of just bolding the first half of the word, as this extension does. For example, Open Sans has six levels of weigh
371.
by thorum 4y ago
Icelandic horses shows zero-shot task generalization on Icelandic natural language prompts, outperforming GPT-3 on many tasks, while being 16x faster on land and having beautiful manes. Very impressive work!
372.
by thorum 4y ago
See also the recently published Tortoise TTS, which IMO sounds even better: https://github.com/neonbjb/tortoise-tts
373.
by thorum 4y ago
Most of the best, most comprehensive PDF libraries are written for Java. There are libraries for other languages but they tend to be incomplete or flawed. There’s also some great paid libraries for C#, but if you want free, I’d recommend lo
374.
by thorum 4y ago
A quick summary of the Limitations section: - "OPT-175B does not work well with declarative instructions or point-blank interrogatives." - "OPT-175B also tends to be repetitive and can easily get stuck in a loop. While sampli
375.
by thorum 4y ago
I wonder what the intended use of this technology is?
376.
Jurassic-X: Crossing the Neuro-Symbolic Chasm with the MRKL System(ai21.com)
6 points by thorum 4y ago|1 comments
377.
by thorum 4y ago
Very cool! https://github.com/thesephist/modelexicon Looks like it’s powered by GPT-J. My understanding is that GPT-J has comparable performance to OpenAI’s Curie model on many tasks (their second-best variant of GPT-3
378.
by thorum 4y ago
Does anyone know if Aurora PostgreSQL is a full drop-in replacement for Postgres, or is it "Postgres compatible" but missing some features (like CockroachDB, for example)? I am very interested in this service but some of my projec
379.
by thorum 5y ago
At that point they'll probably just have their own AI that generates the perfect response to any search query, and return it as the top result.
380.
by thorum 5y ago
If I'm understanding the quoted interview correctly, Google is talking about AI generated spam - like when you ask GPT-3 to write you an article about XYZ topic and it spits out 500 words of well-written, plausible sounding gibberish -
381.
SASS Text Crop – Remove the extra space around text caused by line height(text-crop.eightshapes.com)
1 points by thorum 5y ago|0 comments
382.
by thorum 5y ago
CSS Cascade Layers are now supported by all major browsers and provide a good, standard alternative to !important in cases where you need to override CSS that you can’t easily modify, like stylesheets from a third party CMS or framework, or
383.
by thorum 5y ago
The WCAG guidelines do a good job of answering questions like these, in my experience. Regarding resizing text, the WCAG guidelines provide the following example of a website that meets the standards: "A user uses a zoom function in hi
384.
by thorum 5y ago
Can you give an example of what you mean - a scenario where WCAG guidelines are too ambiguous? https://www.w3.org/WAI/WCAG21/quickref/
385.
by thorum 5y ago
Agreed, but keep in mind that person is already technically required to follow standards for things like PII, sales tax, security etc. in addition to accessibility. Small businesses can outsource these requirements to a service like Shopify
386.
by thorum 5y ago
The standard typically referenced in legal cases is the Web Content Accessibility Guidelines, Level AA. WCAG guidelines are generally clear and straightforward. Showing a good faith effort to comply with those requirements, is the bar fo
387.
by thorum 5y ago
Yes. WCAG (the W3C guidelines referenced as the gold standard for web accessibility) require both captions for people who are deaf and audio descriptions (an audio track with a person describing anything important that is happening in the v
388.
by thorum 5y ago
How big of a concern is collision with dust at these speeds? At .3 light speed would this material survive hitting a micrometeor?
389.
GitHub Actions Is Down(githubstatus.com)
9 points by thorum 5y ago|0 comments
390.
by thorum 5y ago
My favorite is when the link you’re trying to open jumps away 1 nanosecond before you click on it and you wind up clicking something else entirely.

More ›