Hacker News Reader

hnenjoyer_93on YouTube-Commons: Audio transcripts of 2,063,066 YouTube videos, CC-By license

According to the preview, some of these (30 thousand results?) are "Wikipedia article audios"[1], so text turned into speech and now turned back to text

[1]: https://huggingface.co/datasets/PleIAs/YouTube-Commons/viewe...

hnenjoyer_93on StableLM-2-12B

emad_9608 on Stable Audio 2.0 release post[1]:

> Team is working on an open version of this for https://github.com/Stability-AI/stable-audio-tools Dataset just taking some time. Lots of improvements to come like speech, customisation, comfy & more.

[1]: https://old.reddit.com/r/StableDiffusion/comments/1buruzc/co...

Edit: reformatted to one paragraph because HN didn't like the newlines

hnenjoyer_93on Urgent security alert for Fedora 41 and Rawhide users

Also see https://news.ycombinator.com/item?id=39865810

hnenjoyer_93on Why can't anyone agree on how dangerous AI will be?

I briefly skimmed through the text, and it seems to cover only the anti-AI arguments and how some people agree and disagree. At the same time there seems to be no pro-AI arguments, like improving the efficiency of work, medical advancements and other useful things. Maybe that's what was meant with such a title but I would like the author to cover the whole picture next time

hnenjoyer_93on Stop Saying AI

A new winter is never coming, the last few years have shown the potential of machine learning and its use is only going to increase

hnenjoyer_93on Show HN: Wife couldn't find a dev job so I built a tool to automate the search

I should have really worded it better connecting that to the above statements. The problem is that in this particular case a cloud solution would be a better choice for this exact task (unlike for Word or Photoshop). Also the author said they might consider that in another comment

hnenjoyer_93on Show HN: Wife couldn't find a dev job so I built a tool to automate the search

Aside from what the sister comment said, the app being closed source means we can't reliably check what it actually does, how it uses the data and whether it will get us IP bans on the job websites. Paying a subscription for using our own computer's resources and network to scrape the sites is just wrong, I wouldn't complain if it was some web service running in a cloud that just sent notifications though

hnenjoyer_93on Show HN: Wife couldn't find a dev job so I built a tool to automate the search

Not open source and you require a subscription to run a scraper from the user's own device?

Edit: I see that it's free for now but still

hnenjoyer_93on Nikon to acquire RED

Maybe, Redcode is based on JPEG2000

https://youtu.be/IJ_uo-x7Dc0

hnenjoyer_93on Nikon to acquire RED

Z8 and Z9 already record 8K60 12 bit RAW internally, but of course more cameras joining the list is welcome

hnenjoyer_93on Nikon to acquire RED

Interesting that some time ago Red filed a lawsuit against Nikon for using a RAW video compression algorithm that Red claimed infringed its patents, but later dismissed it:

https://dpreview.com/news/5338402238/red-is-suing-nikon-for-...

https://dpreview.com/news/9301564383/nikon-denies-red-s-laws...

https://www.newsshooter.com/2023/04/28/red-patent-lawsuit-ag...

hnenjoyer_93on Samsung is making it harder to know what type of OLED TV you're getting

Vision Pro is Micro OLED, not microLED

hnenjoyer_93on Claude 3 model family

https://chat.lmsys.org/

Choose Arena (side-by-side), it has Claude 3 Opus, Sonnet and GPT-4

hnenjoyer_93on ArtPrompt: ASCII Art-Based Jailbreak Attacks Against Aligned LLMs

I wonder if this can be extended to work with general prompts telling the LLM how to behave, such as a DAN mode

hnenjoyer_93on Show HN: I built a tool to help you search for CivitAI's models by art style

Why do you have a fake "See what our users say" section? For example, "Emma Johnson" is https://unsplash.com/photos/grayscale-photography-of-woman-z..., and "Jackson Smith" is https://unsplash.com/photos/man-sitting-near-gray-steel-roll.... The names are generic and the job titles lead to Wikipedia pages

hnenjoyer_93on Playground v2.5 Technical Report, SoTA Open-Source Image Gen Model [pdf]

They mostly compare it to the base SDXL. There is only one comparison with a community finetune on page 12 (RealStock v2, I suppose the one from Fooocus) and they don't even link to the model. I would like to see more comparisons with community models based on SDXL since they usually generate better images

hnenjoyer_93on Show HN: Hacker News Telegram Bot

No privacy policy, no contact info?

hnenjoyer_93on Show HN: Refractify – Optical software against myopia

This seems great, but I have 2 questions:

1. Since this is for "early myopia", does it mean it is useless for adults? If not, which groups of adults will find it useful?

2. What do you think about this and is this even related?

> The overall findings are equivocal with under‐correction causing a faster rate of myopia progression. There is no strong evidence of benefits from un‐correction, monovision or over‐correction. Hence, current clinical advice advocates for the full‐correction of myopia

https://www.tandfonline.com/doi/full/10.1111/cxo.12978

Edit to clarify, full correction with glasses should mean always seeing the sharpest picture possible