Ahmed Rashad@AhmedZRashad · 2026/07/21赛道影响4/ Full results: https://t.co/SZecuEF0ve: https://t.co/HJkVqQK0y6: https://t.co/caif5PGLlw
Want your model scored? Reply or DM.
AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。
Ahmed Rashad@AhmedZRashad · 2026/07/21观点输出3/ The takeaway:
German and English is basically solved. Arabic and Persian are where systems fall apart, and where the ranking actually means something for a multilingual product.
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ahmed Rashad@AhmedZRashad · 2026/07/21观点输出1/
We built the first real benchmark for code-switched speech.
The finding: most commercial ASR breaks the moment a speaker switches languages mid-sentence. That's not an edge case, it's how millions of people talk.
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ahmed Rashad@AhmedZRashad · 2026/07/08赛道影响RT @POWR_X: "The whole concept of saying AI will need zero humans, I 100% disagree with that," said @AhmedZRashad founder and CEO of @Mee…
AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。
Ahmed Rashad@AhmedZRashad · 2026/06/23观点输出RT @Freeman_Lewin: https://t.co/CImy04kOqQ
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ahmed Rashad@AhmedZRashad · 2026/06/20赛道影响Very true. Elegantly said
AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。
Ahmed Rashad@AhmedZRashad · 2026/05/29赛道影响RT @Vivek4real_: LARRY ELLISON: AI IS RAPIDLY COMMODITIZING BECAUSE MOST MODELS ARE TRAINED ON THE SAME PUBLIC INTERNET DATA.
THE REAL COM…
AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。
Ahmed Rashad@AhmedZRashad · 2026/05/22观点输出RT @DJ_CURFEW: Today we reduced headcount by 22%. The business is the strongest it's ever been. So I think it's important to be direct abou…
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ahmed Rashad@AhmedZRashad · 2026/05/21政策影响We just put two datasets on HuggingFace Perle. Here's what they actually represent.
Code-switched ASR across German, Persian, Saudi Arabic, and Egyptian Arabic, all mixing with English. Real bilingual speech, not the clean monolingual corpora most pipelines train on.
Note the Saudi vs Egyptian Arabic split. We don't just do languages, we do dialects. The way someone code-switches in Riyadh isn't how they do it in Cairo, and that distinction shows up in our data and in your model's performance.
The second set is CT paired with radiology reports. Multimodal, because a scan without the report doesn't teach a model to read it. And medical data is the hardest data to acquire, period. It doesn't sit in public buckets. It sits behind hospitals, privacy law, and institutional review. Sourcing it legally, de-identified, and at scale takes relationships and signed agreements. We have both.
What's public is a sample. Behind it: thousands of hours of code-switched speech across 74 languages and dialects, and a medical pipeline that extends to MRI and CBCT dental, all backed by real sourcing agreements.
Both datasets are linked in the comments.
Egocentric data is in the cooking. Stay tuned.
If your model is gated on data, let's talk.
Datasets are live:
ASR Code-Switch: https://t.co/JwmPWraw6o
Ahmed Rashad@AhmedZRashad · 2026/04/28赛道影响Hey @boardyai, know anyone who can help me?
AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。
评论
暂无评论