HNHacker News
TopNewBestAskShowJobs

spyder

1,651 karma · joined August 31, 2010

submissionscomments
spyder··on AI-generated sad girl with piano performs the text of the MIT License
I have a dream:

https://app.suno.ai/song/d2f8e712-80ae-40ae-967b-90e13278da5...

spyder··on OpenVoice: Versatile instant voice cloning
Huh? Replacing human labor with machine is evil? You wouldn't even able to post this comment without that happening, because computers wouldn't exists or we wouldn't have time for that because many of us would work on farms to produce enough food without the use of human-replacing technologies.

In a similar way as machines allowed to produce abundance of food with less labor, the voice AI combined with AI translation can make information more accesible for the world. Voice actors wouldn't be able to voice act all the useful information in the world, (especially for the more niche topics and for the smaller languages) because it wouldn't worth to pay them and humans are also slower to than machines. We are not far from almost realtime voice translation from any language to any other one. Sure, we can do it with text-only translation, but voice makes it more accessible for lot of people. ( For example between 5–10% of the world has dislexya. )

spyder··on Generating music in the waveform domain (2020)
Stable-audio and MusicGen sounds better than Jukebox.

But the best so far is Suno.ai ( https://app.suno.ai ) especially with their V3 model they have very impressive results, the fidelity is not studio quality but they're getting very close.

It's very likely based on their TTS model they have released before (Bark), but trained on more data and with higher resolution.

https://github.com/suno-ai/bark

spyder··on Chronos: Learning the Language of Time Series
I think it could also have a connection with symbolic AI: The discrete tokens could be the symbols that many believe is useful or necessary for reasoning. It is also useful for compression, reducing memory requirements by the quantization and small integer representations.

https://en.wikipedia.org/wiki/Neuro-symbolic_AI

spyder··on Yi: Open Foundation Models by 01.AI
Yep, humans too have to think before answering most non-trivial questions, and especially the ones that include calculations. So it seems "obvious" that we should try to to give LLMs too some time to think before answering, for example with the popular methods of asking for step-by-step thinking, thinking out loud, and only giving the final answer at the end, and also asking it to proofread and correct it's answers at the end all can help with that.

Pause tokens (thinking tokens) are also an interesting method to achieve that and seems to have a positive effect on performance:

https://arxiv.org/abs/2310.02226

spyder··on Claude 3 model family
What's up with the weird list of the supported countries?

It isn't available in most European countries (except for Ukraine and UK) but on the other hand lot of African counties are listed...

https://www.anthropic.com/claude-ai-locations

spyder··on The Era of 1-bit LLMs: ternary parameters for cost-effective computing
From the BitNet paper:

"Straight-through estimator. To train our 1-bit model, we employ the straight-through estimator (STE)[BLC13] to approximate the gradient during backpropagation. This method bypasses the nondifferentiable functions, such as the Sign (Eq. 2) and Clip (Eq. 5) functions, during the backward pass. STE allows gradients to flow through the network without being affected by these non-differentiable functions, making it possible to train our quantized model."

also the author's (@shumingma) answer in the comments: https://huggingface.co/papers/2402.17764#65df17ed4d436404cdc...

spyder··on Exodus Bitcoin Wallet: $490k swindle
He would just give access for his brokerage or bank account to the lady, like it happens in many other scam cases.
spyder··on Sora: Creating video from text
It's possible it was pre-trained on 3D renderings first, because it's easy to get almost infinite synthetic data that way, and after that they continued the training on real videos.
spyder··on A decoder-only foundation model for time-series forecasting
Yea, and from this paper it looks like the tree model (Catboost supervised) is still beating their (zero-shot) performance and they don't show their supervised version yet, but they write at the end that they plan to do it in a future work. Will be interesting to see.
spyder··on Artificial Intelligence and Peace
computer technology since the 70s is really just "faster paper"

Huh... "paper" is just storage, the whole point of "computer technology" is to have a tool to do computation not just storage. So no, computers are not just faster paper.

spyder··on Disney unveils the HoloTile floor
Yep, and by tilting the spinning discs/cylinders you can have omnidirectional control by changing which part of the edge of the rotating cylinder is in contact with the surface above it. You don't need the ultrasound thing posted in the other comment, that patent may be a different path they've also tried.
spyder··on Scammy AI-Generated Books Are Flooding Amazon
The goal should be to quality check not the AI vs human distinction.

It shouldn't matter if something was written by AI or human if it is high quality , fact-checked. But I think AI (or the way we use it) still needs some improvement for that.

spyder··on How thermal management is changing in the age of the kilowatt chip
There are already such products, just search for "bitcoin heater" .
spyder··on Salad Fingers and the dawn of ‘weird YouTube’
just to add to the list of weird videos:

Cyriak: https://www.youtube.com/watch?v=FavUpD_IjVY

Buttered Side Down: https://www.youtube.com/@ButteredSideDown/videos

spyder··on Launched Playtoob, a simple YouTube companion, because my 5yo daughter loves it
Using AI could have prevented this. So the no AI, or algorithms can be also a negative thing as the example shows. I understand not using it for recommendation because it still has false positives, but using it for filtering could have prevented this. And false negatives for filtering is not so bad and could be just used to put filtered videos in a list that the parent can review, in the case that some good videos are getting accidentally filtered.
spyder··on Lockheed is now tracking phones and walkie-talkies from space
Yep, a video about it capturing a murder (blurry video from 7 years ago I'm sure it's even better today):

https://www.youtube.com/watch?v=mJLr0KMsRAA

spyder··on Cryptographers solve decades-old privacy problem
For example the Numerai hedge fund's data science tournament for crowdsourced stock market prediction is giving out their expensive hedge fund quality data to their users but it's transformed enough that the users don't actually know what the data is, yet the machine learning models are still working on it. To my knowledge it's not homomorphic encryption because that would be still too computational expensive, but it would be an ideal application for this.

https://numer.ai/

spyder··on I think I need to go lie down
Huh? UIs should be repetitive not impressive or original, so users don't have to learn different new interface for every app.
spyder··on David Attenborough is now narrating my life
In the tweet he says he is using https://elevenlabs.io for the voice, which is currently the best quality text-to-speech service and allows for voice cloning too.

And Elevenlabs is saying this about cloning other people's voices:

Can I clone anybody's voice?

Yes, as long as you have their consent. For Professional Voice Cloning, we have integrated robust security measures to make sure you can only clone your own voice. Unless you share it, your voice belongs and is available only to you.

But it looks like it's not enforced if the example in the tweet is possible.

But even if it would be enforced by the company, open-source models are getting close to them and they are harder to control. The public or commercial use of the output from these models can (and probably should) be limited by laws, but if the users want to use it privately (for example to narrate themselves) there is not much you can do to prevent it especially with the progress in this tech it will become more trivial to do it for anybody.

spyder··on Starship will attempt a launch this Friday
A small prototype was already built in Las Vegas, they can even change the "logos" on it :)
spyder··on New York restaurants fight back against reservations by bots
One of the reason is mentioned in the article:

"When guests drop $100 or more just to walk in the door, “people have [the] wrong expectation when they come” he says because those expectations might be unreasonably high."

spyder··on AI can catalogue a forest's inhabitants simply by listening
There are companies working on that using AI:

Acoustic sensors in the car: https://v2minc.com/#solution

Skoda made s sound analyzer mobile app 3 years ago: https://beebom.com/this-ai-app-detects-internal-issues-in-ca...

spyder··on Neurons in Large Language Models: Dead, N-Gram, Positional
There are already methods to do that in neural networks, it's called "pruning"

https://en.wikipedia.org/wiki/Pruning_(artificial_neural_net...

and, yes the article also mentions that it happens in mammal brains.

spyder··on How to Use AI to Do Stuff: An Opinionated Guide
Same and it pops up only when you are in the middle of reading (on scrolling down) which was so distracting I didn't even continue reading.

Probably they know or measured that immediately showing pop up on a page load gets ignored and closed so they sneaked in the middle of reading, together with the small print dismission text they look pretty UX evil.

spyder··on Measuring blood pressure for less than a dollar using a phone
This looks like it' using the method that was mentioned in the article too, it needs regular calibration with a cuff (which they include with the product).
spyder··on Crawling a quarter billion webpages in 40 hours (2012)
CommonCrawl / Open PageRank ?

https://www.domcop.com/top-10-million-websites

spyder··on First people sickened by Covid-19 were scientists at WIV: US government sources
Hmm... but intelligent people don't think they're that intelligent (Dunning–Kruger effect) :-)
spyder··on Apollo will close down on June 30th
Sure, and that's why they value API users around 20x more than their website users ? (based on the rough estimation in the post)
spyder··on Show HN: StonksGPT – A Natural Language search tool for Stocks and Finance data
Hugged to death? Console shows: 429 (Too Many Requests)
← PreviousPage 3 of 30Next →