Leo

Oh? You're approaching me?

Public Key

npub1ap2cqjyf709p4xn5h2fkl8r5h9mtvvf4dspmk7st5ynzynh6fq9q2tpvx3

NIP-05 Address

leo@raru-re.mostr.pub

Profile Code

nprofile1qqsws4vqfzyl8js6nf6t4ym0n36tja4kxy6kcqam0g96zf3zfmayszspzemhxue69uhhyetvv9ujumn0wd68ytnzv9hxg4mvyxp

Publishing to

relay.nostr.band

Author Public Key

npub1ap2cqjyf709p4xn5h2fkl8r5h9mtvvf4dspmk7st5ynzynh6fq9q2tpvx3

Show more details

Last Notes

2025-03-18 21:30:14

- reply

@nprofile…skc4
"it's open sourced"
And yet it has a non-open-source license. I know being open-source is a huge marketing boost for people's projects, but there is no need to lie about it.

2025-03-09 09:19:19

Character.ai doesn't have the best models by a long shot, but it's so good for story-writing. The UI lends itself really well to it, and the characters stay in character more than some other models.
Are there any other apps / websites people recommend for this? Not writing as in making the model write a story from a prompt, but instead writing by interacting with your characters themselves.
#Writing #LLM #AI

2025-02-27 16:36:52

Deployed a new Numerai model today, let's hope we get good returns. More green and less red :blobcatBlessUp:
#Numerai #MachineLearning #AI #DataScience #Programming

2025-02-23 01:39:33

I got 21.5 million MyAnimeList scores. Let's keep collecting more data and see if we can build a decent recommendation engine.
#Anime #Programming #BigData #DataScience #MachineLearning
https://cdn.raru.re/media_attachments/files/114/050/641/630/985/528/original/00594dec3227c36b.png

2025-02-08 20:07:24

- reply

@nprofile…q5wq I've blocked 1 GB of ads and I haven't even been using this browser too long.
https://cdn.raru.re/media_attachments/files/113/970/068/578/886/332/original/fce5a5bb54070141.png

2025-02-07 07:54:12

- reply

@nprofile…vwt0 they're not ready for Fedi-level toxicity yet. You'll get Community Noted to oblivion.

2025-01-29 16:55:37

Why are they so salty? Every time someone releases an open-source model they become like this.
#OpenAI #AI #LLM #DeepSeek
https://cdn.raru.re/media_attachments/files/113/912/688/264/116/207/original/25d3882a09a8cfdb.png

2025-01-22 21:25:26

- reply

@nprofile…vvm4 @nprofile…92y4
You'd think the name "Death Valley" would provide a hint to people, but I guess that's expecting too much.

2025-01-22 20:35:07

- reply

@nprofile…vdjy
Keep in mind that if you do find a flaw or something minor that can be improved, people will jump at your throat and call you a Telegram shill or a Meta shill.

2025-01-21 21:18:24

Active Learning is a freaking superpower.
If you're training a model and need to manually label data, you might have tried labeling every sample, or labeling things randomly.
Instead of this, Active Learning uses an intermediary model to determine which samples would be beneficial to label. It does this by finding samples that the model is uncertain about, after being trained on the samples labeled so far.
I started doing this with one of my LeoSearch-related projects, and the difference it makes it day and night.
Now when the system asks me to label a sample, it's so difficult that I can feel the examples helping the model. I don't need to spend time randomly labeling simple examples any more.
@nprofile…ya4t
#MachineLearning #programming #AI #SearchEngine

2025-01-21 20:28:35

- reply

@nprofile…xvlc
"This isn't about electricity."
Was it about gas, then?

2025-01-21 15:35:18

- reply

@nprofile…57rm yep, it's not the bad old days with terrible practicality and tooling.
Code for encryption that should be un-crackable for the foreseeable future fits on a post-it note.
Speed is essentially a non-issue, a 10 year old cheapo Android phone can encrypt your messages faster than you can hope to read or write them.

2025-01-19 22:12:21

- reply

@nprofile…l0gg I wonder what confuses it. "Library" probably brings it too close to "hardcover book" in the latent space.
If the model allows negative prompts maybe it can help.

2025-01-19 12:39:43

- reply

@nprofile…fle4 I wonder how much of this is caused by the protocol itself. Python with asyncio scales pretty well. And text chat is not a very hard thing to scale.
When everyone called it an IRC/Discord alternative I checked out the Matrix protocol, hoping to write a small implementation to host my own server. It was unnecessarily complex, and in the end I didn't find it valuable enough to continue the project.
Writing an IRC server takes an afternoon. Writing a nostr server/client, including your own implementation of WebSockets takes a weekend. Matrix is a whole other beast of complexity, while lagging more than either of those.

2025-01-19 12:22:21

- reply

@nprofile…5k85 Europe takes its time but drops the big hammers eventually. Some GDPR-ish god-tier law targeting these apps would be good.
Makes me wonder, how come everyone else got GDPR'd but Tiktok is still Tiktok-ing? Is Europe assuming China is just gonna send their email to spam?

2025-01-19 10:50:56

- reply

@nprofile…args you don't get why a CCP-controlled app with extensive data collection pushing content to 100 million (mostly young and impressionable) Americans would be not desirable?

2025-01-18 10:55:37

- reply

@nprofile…k0gr why is Unity Hub accessing Unity Hub data sus?

2025-01-11 17:42:32

- reply

@nprofile…pngt they should make some sort of garbage collection for this. Freaking integers are garbage collected in Python, but not 160 GB of dependencies.

2025-01-11 17:42:32

- reply

@nprofile…pngt they should make some sort of garbage collection for this. Freaking integers are garbage collected in Python, but not 160 GB of dependencies.

2025-01-11 17:39:36

Python dependency sizes are worse than Rust lol
#python #rust #programming #opensource
https://cdn.raru.re/media_attachments/files/113/810/940/945/686/002/original/7227a47d7106e945.png