Awsaf

Incoming Intern at @Microsoft | Graduate Student @ucsantabarbara || Prev: @amazon, @google, @weights_biases || @kaggle Grandmaster

[Awsaf @awsaf49]

🚀 Late Posting! My first-author paper has been accepted to @iclr_conf 2025 🎉

📝 Title: SONICS: Synthetic Or Not - Identifying Counterfeit Songs

🔗 Paper: https://t.co/oRaQDZTgQc (Final version coming soon!)

🔎 What’s inside?

🎶 SONICS Dataset:
✅ First large-scale dataset for detecting AI-generated fake songs (97k samples)
✅ 49k AI-generated songs (Suno & Udio) + 48k real songs (YouTube)
✅ Covers 5 different AI models (3 from Suno, 2 from Udio)

⚡ New Model: SpecTTTra 🔱
✅ Tackles long-context patterns (e.g., repeated verses, rhythms)
✅ 8% higher F1-score than ViTs while 38% faster, 26% less memory
✅ Disentangles spectral & temporal info, boosting efficiency

🧑‍🔬 Human vs. AI Benchmark:
🔹 AI-generated Suno songs are easier to detect than Udio
🔹 Udio songs sometimes create the "Karaoke effect"
🔹 Real songs have unique musical characteristics
🔹 AI models outperform humans at detecting fake songs

👀 Coming soon:
✔️ Open-source SONICS dataset & code
✔️ A @Gradio web app on @huggingface : Test if you can spot fake songs!

Huge thanks to my co-authors Zaber Ibn Abdul Hakim, Najibul Haque Sarker, Bishmoy Paul, & Prof. Shaikh A Fattah for their support! 🙌

#ICLR2025 #AIMusic #DeepfakeDetection #FakeSongs #AudioProcessing #Transformers #MachineLearning

[Awsaf @awsaf49]

@MinjiYoon90 hi, curious, was the team able to find out new or conflicting findings than qwen or olmo or min-max or llama team ?

[Awsaf @awsaf49]

@elonmusk I had such high hopes for @xai, but then all the co-founders left, and now I'm not sure what to expect from Grok. Those early announcement meetings were cool. You just knew something great was about to come. Now Grok has even opted out of HLE. Sigh.

[Awsaf @awsaf49]

but isn't budget awareness also make llm less intelligent? if i'm not wrong claude code w/ claude opus 4.6 was really good then they tried to reduce cost by making it less verbose with system prompt and default effort to medium, which ended up making the model dumb as people started to notice in benchmark. Then they reverted that.

https://t.co/2bS0IcRZ5l

[Awsaf @awsaf49]

@mariotomich fix slouching issue first!

[Awsaf @awsaf49]

@rodger @minakimes And here people complain when LLM memorizes!
"Memorization is all you need"

[Awsaf @awsaf49]

@SingularityRes this is hilarious, claude models are privately hosted on aws bedrock for internal usage. not sure why they have to pay anything to anthropic. they owns good portion of anthropic and I would guess this private usage comes with the deal

[Awsaf @awsaf49]

@zmwang lol that feels like 70% of SF xD

[Awsaf @awsaf49]

@nff_landslag jersey sucks but will support you guys tho!

[Awsaf @awsaf49]

@Lunens__

[Awsaf @awsaf49]

@sama It would be really nice to have access to such models as a researcher! but feels like you need to have special privileged to be a beta tester of such nice models :(

[Awsaf @awsaf49]

@OfficialLoganK Personal Intelligence has gotten really bad tho

[Awsaf @awsaf49]

@GeminiApp after the update personal intelligence has gotten very bad!

[Awsaf @awsaf49]

@karpathy Breaking news: Adrej Karpathy joined Anthropic to avoid token limit in claude code!

[Awsaf @awsaf49]

@ryanvogel You don't have token limit for claude code anymore

[Awsaf @awsaf49]

@ns123abc Keep your friends close and your enemies closer!

[Awsaf @awsaf49]

@Rainmaker1973 probably because they are so cute, how can such cute animals even harm humans. another example is panda :v

[Awsaf @awsaf49]

@tdietterich @arxiv @grok anyway to create alternative of @arxiv with @claudeai or sth?

[Awsaf @awsaf49]

Saw a Facebook ad selling NeurIPS / ICML / ACL co-authorship to high school students!

That’s how you know things are getting bad 🙄

@NeurIPSConf @icmlconf @aclmeeting

[Awsaf @awsaf49]

@yoavgo lol

[Awsaf @awsaf49]

@Polymarket 😐