Ammar

ex. hacker | ai trader | exploring options in research | I contribute and write papers | prev. @DondersInst | @laion_ai | present. sleeping ai

[Ammar @sleepincat4]

@albfresco @OpenAI built to look like Maluma but works in IT

[Ammar @sleepincat4]

@prof_kamilov can I mention that I impressed a UC Berkeley professor so much that we went on a coffee date?

[Ammar @sleepincat4]

@TaliaRinger tried to learn lean before realizing I am not worth it

[Ammar @sleepincat4]

@bryancsk cuz it’s still relevant, read AlexNet and you’ll know why

[Ammar @sleepincat4]

@pmddomingos except America is a rebel colony of europe that got fed-up from its cultural roots and killed all natives. colonised won’t do colonisation.

[Ammar @sleepincat4]

@max_simchowitz exactly! you need a cold heart to write heartfelt reviews

[Ammar @sleepincat4]

@yuzu_jpg I don’t use Meta AI

[Ammar @sleepincat4]

@QCXINT_ I’m not surprised that it’s Chinese

[Ammar @sleepincat4]

@osanseviero Better audio quality like MOSS models

[Ammar @sleepincat4]

@notsky you shouldn’t state position but mentioning lab is fine otherwise you won’t know whom is from where

[Ammar @sleepincat4]

@notsky @notsky updated it and changed how the audio is streamed and made distributed cdn. thanks for the feedback, i realised the global lock wasn't being precise enough and causing problems.

[Ammar @sleepincat4]

would appreciate retweets https://t.co/CibjvIoyrq

- will have official announcement soon
- later a subsequent paper
- please, stress test and report bugs
- feel free to ask questions

[Ammar @sleepincat4]

#1 bug

After redownloading audio, player stops working, working on fix

[Ammar @sleepincat4]

would appreciate retweets https://t.co/CibjvIoyrq

- will have official announcement soon
- later a subsequent paper
- please, stress test and report bugs
- feel free to ask questions

[Ammar @sleepincat4]

@notsky Update schedule broke I think, will update shortly

[Ammar @sleepincat4]

@natolambert That’ll break Gemma philosophy of being small

[Ammar @sleepincat4]

creating benchmarks that don’t stale rather evolve with advancement of models.

Lastly, it is still in early phase, we expect people to stress test the website and report bugs so we can have a stable release. 8/8

[Ammar @sleepincat4]

releasing a new eval concept called active benchmark for ai in music: https://t.co/Bk0OMnK1gj

(Thread)

[Ammar @sleepincat4]

4. plan to add more songs in coming future

5. We only add AI made songs, if someone uploads human song, that too be treated like a mystery cuz we also won’t know as strip all info.

6. It is a community run effort that aims to sustain a long term vision for 7/8

[Ammar @sleepincat4]

Key details

1. We don’t save whole songs, users can upload songs only random 10s clip will be saved from each songs, that too uploaded on public hf repo.

2. We expect users to have the rights for the songs they upload

3. We have added a base 2000 clips from suno 6/8

[Ammar @sleepincat4]

active benchmark was created via inspiration from @CarnegieMellon MusicArena leaderboard except that is a stale leaderboard vs this allows much more freedom for annotators and contributors. It is practically a community run effort 5/8

[Ammar @sleepincat4]

these annotations are anonymous and we ask users to use passkeys for login and no other alternative method and track users. we have basic set of questions and way of adding new ones every other moment.

4/8