vagabund

5 ms·

1.
by vagabund 12d ago
Yeah, stopped reading after the first paragraph. It's really so disrespectful to your audience.
2.
Show HN: Generate a /mu/ style essentials chart for your last.fm history(mymuguide.site)
1 points by vagabund 6mo ago|0 comments
3.
by vagabund 8mo ago
Really fun to fly around, find my old apartment building etc. It would be neat if you could drag and click to select an area to inpaint. Let's see everyone's new Penn Station designs! Would guess it'd have to be BYOK but it
4.
by vagabund 8mo ago
More info: https://www.corememory.com/p/exclusive-openai-and-sam-altman...
5.
Merge Labs – Altman-Backed BCI Lab Using Biomolecular Ultrasound(merge.io)
1 points by vagabund 8mo ago|1 comments
6.
Show HN: Inflow – invoke an LLM with your viewport just by typing(github.com)
2 points by vagabund 1y ago|0 comments
7.
One mother for two species via obligate cross-species cloning in ants(nature.com)
2 points by vagabund 1y ago|0 comments
8.
H-Nets – The Past and Future(goombalab.github.io)
3 points by vagabund 1y ago|0 comments
9.
Senescence-resistant human cells counter aging in primates(cell.com)
4 points by vagabund 1y ago|1 comments
10.
by vagabund 1y ago
I'd push back on a couple things here. The notion that Scale AI's data is of secondary value to Wang seems wrong: data-labeling in the era of agentic RL is more sophisticated than the pejorative view of outsourcing mechanical turk
11.
by vagabund 1y ago
The huggingface spaces link doesn't work, fyi. Sounds awesome in the demo page though.
12.
by vagabund 2y ago
> I highly doubt you are getting novel, high quality data. Why wouldn't you? Presumably the end user would try their use case on the existing model, and if it performs well, wouldn't bother with the expense of setting up an RL
13.
by vagabund 2y ago
My understanding is they built a performant suite of simulation tools from the ground up, and then they expose those tools via API to an "agent" that can compose them to accomplish the user's ask. It's probably less gen
14.
by vagabund 2y ago
Searching Spotify for user created playlists is still probably your best bet. Youtube has some good results too. Here are two that might fit what you're looking for: '90s K-pop: https://open.spotify.com/playlist&#
15.
by vagabund 2y ago
No, but it's also biased toward their commercial partners. From this page [0], detailing their recommendation process: > How do commercial considerations impact recommendations? > [...] In some cases, commercial considerations, s
16.
by vagabund 2y ago
It may just be my perception, but I seem to have noticed this steering becoming a lot more heavy handed on Spotify. If I try to play any music from a historical genre, it's only about 3 or 4 autoplays before it's queued exclusivel
17.
Harris-sponsored Google ads put news outlets in a tough spot(axios.com)
7 points by vagabund 2y ago|3 comments
18.
Prior Covid infection may protect against common cold(science.org)
2 points by vagabund 2y ago|0 comments
19.
Stem Projector(stemplayer.com)
3 points by vagabund 2y ago|0 comments
20.
by vagabund 2y ago
Yeah they absolutely do not use the pile.
21.
Language Modeling by Estimating the Ratios of the Data Distribution(aaronlou.com)
1 points by vagabund 3y ago|0 comments
22.
by vagabund 3y ago
"Hawk-3B exceeds the reported performance of Mamba-3B (Gu and Dao, 2023) on downstream tasks, despite being trained on half as many tokens. Griffin-7B and Griffin-14B match the performance of Llama-2 (Touvron et al., 2023) despite bein
23.
Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient LMs(arxiv.org)
6 points by vagabund 3y ago|1 comments
24.
Phallocentricity in GPT-J's stratified ontology(lesswrong.com)
18 points by vagabund 3y ago|3 comments
25.
Neural Network Diffusion(arxiv.org)
223 points by vagabund 3y ago|86 comments
26.
by vagabund 3y ago
Author thread: https://twitter.com/liuzhuang1234/status/1760195922502312197
27.
Mistral releases their latest prototype model, Next, to Chatbot Arena(chat.lmsys.org)
2 points by vagabund 3y ago|0 comments
28.
by vagabund 3y ago
Author thread: https://twitter.com/ProfBuehlerMIT/status/175628414213464913... Code and weights: https://huggingface.co/lamm-mit/ProteinMechanicsDiffusionDes...
29.
ForceGen: End-to-end de novo protein generation(science.org)
64 points by vagabund 3y ago|10 comments
30.
by vagabund 3y ago
They're generating the audio. They use a series of techniques to automatically generate metadata for speech samples in LibriSpeech for things like accent, recording quality, pitch, speed, gender, then use an LLM to format these tags i

More ›