متابعة زيارة ترامب لدولة الإمارات العربية المتحدة.

[RSA]

الموضوع صار سوالف مراهقين .. سيارة ابوي اقوى من سيارة ابوك

لا الامارات ولا السعودية محتاجين حد يثبت مراكزهم

 

[Khalifa ss]

اتسائل الى اي مركز سنقفز بعد الاستثمارات الهائلة في قطاع الذكاء الصناعي ...

تقديري الشخصي ،، ربما المركز الثاني او الثالث على ابعد تقدير.

الطريق طويل والإمارات بدأت مبكراً عندما كان الذكاء الاصطناعي مجرد مصطلح في الأوراق البحثية او فكرة في أفلام هوليوود كان لديها وزارة وانشأت جامعات ودعمت الموسسات الخاصة والحكومية في هذا المجال .. خطوات صغيرة نظرية وقتها لكن طريق الألف ميل يبداً بخطوة .. اليوم لازلنا في بداية الطريق ..

 

[Hegesias]

Starting from Llama-2 pretrained model weights, we continue pretraining the ALLaM-7B and ALLaM-13B models on 1.2T tokens, covering both English and Arabic languages.


We first demonstrate the feasibility of adapting an existing pretrained English model (Llama-2) to fluency in both Arabic and English through tokenizer and vocabulary expansion.


..



Quotes form the research paper which was written and published by the Saudi authorities themselves



Second an option form IBm

ALLaM is a series of powerful language models designed to advance Arabic Language Technology (ALT) developed by the National Center for Artificial Intelligence (NCAI) at the Saudi Data and AI Authority (SDAIA). These models are initialized with Llama-2 weights and are trained on both Arabic and English


Not here to argue facts .. sorry


..



Flashy names or bad names or whatever if you have credible sources that back your opinion or we can distinguish between personal opinion and correct facts

..

Let’s not rewrite history: yes, ALLaM builds on LLaMA-2 weights, so what? Every major model stands on the shoulders of predecessors. The real achievement is how Saudi Arabia took that base, expanded vocabularies, retrained on 1.2 trillion tokens including deep Arabic data, and created models truly tailored for the language and culture. That’s not a minor tweak; it’s a full-scale transformation

Saudi’s NCAI and SDAIA didn’t just slap a label on a foreign model, they engineered and deployed a powerful Arabic-focused AI stack from the ground up, backed by serious compute, data, and national strategy. Unlike the UAE’s flashy demos, this is real infrastructure, real deployment, real impact

If you want to argue, bring credible sources. But don’t dismiss Saudi’s contribution like it’s an afterthought just because it’s based on existing tech. That’s how all tech evolves. The question is who owns, controls, and applies it, and on that, Saudi is miles ahead​

عدوكم المصاريه, خذ حقائق هنا

ArabicMMLU: Assessing Massive Multitask Language Understanding in Arabic

The focus of language model evaluation has transitioned towards reasoning and knowledge-intensive tasks, driven by advancements in pretraining large models. While state-of-the-art models are partially trained on large Arabic texts, evaluating their performance in Arabic remains challenging due...

arxiv.org

 

[صقر الامارات]

يعجبني كذلك ، العمل بذكاء في هذا المجال المهم ..

وزارة للذكاء الصناعي ووزير ذكاء صناعي

اطلاق استراتيجية الامارات للذكاء الصناعي

منهج متكامل للطلبة في مجال الذكاء الصناعي

جامعة محمد بن زايد للذكاء الصناعي

تشريعات وقوانين وحوكمة وبنية تحتية متخصصة

استثمارات هائلة

والكثييير ....


كل ذلك اعطى الامارات مكانتها المستحقة في هذا المجال (والقادم بعون الله افضل).

 

[صقر الامارات]

الطريق طويل والإمارات بدأت مبكراً عندما كان الذكاء الاصطناعي مجرد مصطلح في الأوراق البحثية او فكرة في أفلام هوليوود كان لديها وزارة وانشأت جامعات ودعمت الموسسات الخاصة والحكومية في هذا المجال .. خطوات صغيرة نظرية وقتها لكن طريق الألف ميل يبداً بخطوة .. اليوم لازلنا في بداية الطريق ..

انت في بداية الطريق وتم تصنيفك ضمن الTOP 5 كقائد للذكاء الصناعي على مستوى العالم (GLOBAL AI LEADERS) ..

والقادم - بتكامل كل هذه الخطوات والبنى التحتية والتشريعات والاستثمارات الخ..) سيعزز هذا الموقف ويدفعه للأمام ان شاء الله.

 

[Hegesias]

الطريق طويل والإمارات بدأت مبكراً عندما كان الذكاء الاصطناعي مجرد مصطلح في الأوراق البحثية او فكرة في أفلام هوليوود كان لديها وزارة وانشأت جامعات ودعمت الموسسات الخاصة والحكومية في هذا المجال .. خطوات صغيرة نظرية وقتها لكن طريق الألف ميل يبداً بخطوة .. اليوم لازلنا في بداية الطريق ..

الاساس النظري للذكاء الصناعي بدأ في الثمانينيات الله يصلحك. عندها الامارات كان عمرها 10 سنين.

 

[Hegesias]

انت في بداية الطريق وتم تصنيفك ضمن الTOP 5 كقائد للذكاء الصناعي على مستوى العالم (GLOBAL AI LEADERS) ..

والقادم - بتكامل كل هذه الخطوات والبنى التحتية والتشريعات والاستثمارات الخ..) سيعزز هذا الموقف ويدفعه للأمام ان شاء الله.

عدوكم المصاريه

وقفزت المملكة 17 مركزاً، لتحتل الآن المرتبة 14 عالمياً، متجاوزة الإمارات العربية المتحدة كالدولة العربية الرائدة في مجال الذكاء الاصطناعي.

https://www.arabnews.com/node/2572046/media

 

[Khalifa ss]

Let’s not rewrite history: yes, ALLaM builds on LLaMA-2 weights, so what? Every major model stands on the shoulders of predecessors. The real achievement is how Saudi Arabia took that base, expanded vocabularies, retrained on 1.2 trillion tokens including deep Arabic data, and created models truly tailored for the language and culture. That’s not a minor tweak; it’s a full-scale transformation

Saudi’s NCAI and SDAIA didn’t just slap a label on a foreign model, they engineered and deployed a powerful Arabic-focused AI stack from the ground up, backed by serious compute, data, and national strategy. Unlike the UAE’s flashy demos, this is real infrastructure, real deployment, real impact

If you want to argue, bring credible sources. But don’t dismiss Saudi’s contribution like it’s an afterthought just because it’s based on existing tech. That’s how all tech evolves. The question is who owns, controls, and applies it—and on that, Saudi is miles ahead​

عدوكم المصاريه, خذ حقائق هنا

ArabicMMLU: Assessing Massive Multitask Language Understanding in Arabic

The focus of language model evaluation has transitioned towards reasoning and knowledge-intensive tasks, driven by advancements in pretraining large models. While state-of-the-art models are partially trained on large Arabic texts, evaluating their performance in Arabic remains challenging due...

arxiv.org

a flashy demo positioned in the top performing models “based on the research paper you attached”



Nice demos



Quotes from the paper itself


We find that Jais-chat 30B, the largest model in the Jais series, achieves the highest accuracy across all models evaluated, with 62.3% accuracy.



Jais-chat 30B outperforms GPT-3.5 (57.7%) by 4.6 percentage points, making it the strongest performer on ArabicMMLU.



Falcon isn’t Arabic focused LLM

Among the evaluated general-purpose multilingual models, Falcon and XGLM lag behind significantly, suggesting their tokenization and pretraining data are less optimized for Arabic



Nice research paper ..

 

[Khalifa ss]

الاساس النظري للذكاء الصناعي بدأ في الثمانينيات الله يصلحك. عندها الامارات كان عمرها 10 سنين.

صحيح ..

 

[Hegesias]

a flashy demo positioned in the top performing models “based on the research paper you attached”



Nice demos



Quotes from the paper itself


We find that Jais-chat 30B, the largest model in the Jais series, achieves the highest accuracy across all models evaluated, with 62.3% accuracy.



Jais-chat 30B outperforms GPT-3.5 (57.7%) by 4.6 percentage points, making it the strongest performer on ArabicMMLU.



Falcon isn’t Arabic focused LLM

Among the evaluated general-purpose multilingual models, Falcon and XGLM lag behind significantly, suggesting their tokenization and pretraining data are less optimized for Arabic



Nice research paper ..

Nice cherry picking

Yes, Jais-Chat 30B did well, on one Arabic benchmark, but let’s not pretend that alone makes it a game-changer. You’re pointing to a flashy moment in a controlled demo. That’s not general dominance, that’s selective lighting

Meanwhile, ALLaM was built specifically for Arabic, not as a multilingual model with some Arabic seasoning. That’s why it consistently outperforms general-purpose LLMs like Falcon, which, as the paper admits, lags behind in Arabic due to poor tokenization and weak pretraining data

You can quote performance numbers all day, but the deeper question is who’s investing in real, focused Arabic AI, and who’s just showing off good lighting for the press? We both know the answer

 

[Khalifa ss]

الموضوع صار سوالف مراهقين .. سيارة ابوي اقوى من سيارة ابوك

لا الامارات ولا السعودية محتاجين حد يثبت مراكزهم

من ناحيتي لا مقارنة ولا تهكم مجرد سرد وشرح للحقائق العلمية وأنا اعتقد ان البلدين يكملون بعضهم والسعودية طموحة جداً ولا شخصياً لا أتمنى لها اقل من ما أتمناه للإمارات وافرح بانجازها ..

 

[Khalifa ss]

Nice cherry picking

Yes, Jais-Chat 30B did well, on one Arabic benchmark, but let’s not pretend that alone makes it a game-changer. You’re pointing to a flashy moment in a controlled demo. That’s not general dominance, that’s selective lighting

Meanwhile, ALLaM was built specifically for Arabic, not as a multilingual model with some Arabic seasoning. That’s why it consistently outperforms general-purpose LLMs like Falcon, which, as the paper admits, lags behind in Arabic due to poor tokenization and weak pretraining data

You can quote performance numbers all day, but the deeper question is who’s investing in real, focused Arabic AI, and who’s just showing off good lighting for the press? We both know the answer

Nope let’s put things in the right order

Jais is an Arabic focused LLM that’s why it’s the top performing against all other models including gpt 3.5 itself meaning more than 70% of capacity is trained and developed for Arabic language usage.


Compared to AceGPT-chat (13B), both Jais-chat models (13B and 30B) exhibit substantially higher accuracy in areas including STEM, Social Science, Humanities, and Others.

…

Falcon isn’t multilingual by design it wasn’t developed for multilingual use that’s why it lags behind in Arabic tests which the paper clearly state:
…

Jais found it more challenging to answer questions from countries like Morocco which usually meant the accent and the context but that challenge did not impact its overall performance.

Jais performs best overall except in questions sourced from Morocco.


No cherry picking here you have the source and you can read it and quote it as much as you want ..

 

[Hegesias]

Nope let’s put things in the right order

Jais is an Arabic focused LLM that’s why it’s the top performing against all other models including gpt 3.5 itself meaning more than 70% of capacity is trained and developed for Arabic language usage.


Compared to AceGPT-chat (13B), both Jais-chat models (13B and 30B) exhibit substantially higher accuracy in areas including STEM, Social Science, Humanities, and Others.

…

Falcon isn’t multilingual by design it wasn’t developed for multilingual use that’s why it lags behind in Arabic tests which the paper clearly state:
…

Jais found it more challenging to answer questions from countries like Morocco which usually meant the accent and the context but that challenge did not impact its overall performance.

Jais performs best overall except in questions sourced from Morocco.


No cherry picking here you have the source and you can read it and quote it as much as you want ..

Let’s actually put things in order

Jais did well, as it should, being fine tuned for Arabic and backed by heavy compute. But the very paper you’re quoting from also shows ALLaM outperforming Jais on several key benchmarks, especially in Arabic centric QA tasks, which are far more representative of real world language understanding than selective MMLU metrics


If we’re going to talk about specialization, ALLaM is the one actually winning on deep Arabic language tasks, not just high level categories like “Social Science” with multiple choice questions. And let’s not ignore ALLaM pulling this off with a smaller model size and less compute noise, that’s optimization, not overkill


Also, saying Jais struggled only with Moroccan dialects is basically saying it tripped on a real world challenge, and that’s the point. You can’t dominate Arabic AI if you fall apart the moment dialects enter the room


So yes, nice paper. Read past the headlines, it’s ALLaM that walks out with the crown

 

[Khalifa ss]

Let’s actually put things in order

Jais did well, as it should, being fine tuned for Arabic and backed by heavy compute. But the very paper you’re quoting from also shows ALLaM outperforming Jais on several key benchmarks, especially in Arabic centric QA tasks, which are far more representative of real world language understanding than selective MMLU metrics


If we’re going to talk about specialization, ALLaM is the one actually winning on deep Arabic language tasks, not just high level categories like “Social Science” with multiple choice questions. And let’s not ignore ALLaM pulling this off with a smaller model size and less compute noise, that’s optimization, not overkill


Also, saying Jais struggled only with Moroccan dialects is basically saying it tripped on a real world challenge, and that’s the point. You can’t dominate Arabic AI if you fall apart the moment dialects enter the room


So yes, nice paper. Read past the headlines, it’s ALLaM that walks out with the crown

Quote it from the paper please where these things are mentioned .. my English may not be as good as yours that’s why I couldn’t check what you have mentioned.

 

[Hegesias]

ALLaM: Large Language Models for Arabic and English

We present ALLaM: Arabic Large Language Model, a series of large language models to support the ecosystem of Arabic Language Technologies (ALT). ALLaM is carefully trained considering the values of language alignment and knowledge transfer at scale. Our autoregressive decoder-only architecture...

arxiv.org

 

[Khalifa ss]

يعجبني كذلك ، العمل بذكاء في هذا المجال المهم ..

وزارة للذكاء الصناعي ووزير ذكاء صناعي

اطلاق استراتيجية الامارات للذكاء الصناعي

منهج متكامل للطلبة في مجال الذكاء الصناعي

جامعة محمد بن زايد للذكاء الصناعي

تشريعات وقوانين وحوكمة وبنية تحتية متخصصة

استثمارات هائلة

والكثييير ....


كل ذلك اعطى الامارات مكانتها المستحقة في هذا المجال (والقادم بعون الله افضل).

مستقبل اكثر اشراقاً لان الإمارات وقادتها يبذلون الغالي والنفيس لرفعة الوطن بشكل خاص في الشأن العلمي من الفضاء إلى الذاك الاصطناعي إلى العملات الرقمية .. هناك رؤية كبيرة مبهرة في الإمارات ونتائج ملموسة ..

ريادة عالمية

 

[Khalid88]

 

[صقر الامارات]

الامارات بدأت تصدر التكنولوجيا وتبني مراكز ذكاء صناعي حول العالم (على سبيل المثال لا الحصر)

مركز بيانات للذكاء الاصطناعي بسعة 1 جيجاواط: شراكة استراتيجية بين الإمارات وفرنسا

الإمارات العربية المتحدة وفرنسا تبنيان مركز بيانات للذكاء الاصطناعي بقدرة 1 جيجاواط: ما نعرفه عن هذا المشروع التكنولوجي الضخم

www.khaleejtimes.com

MSN

https://al-ain.com/article/italy-uae-artificial-intelligence-center

ويضاف ما تم الاعلان عنه من مركز هائل في الامارات بمقدوره ان يخدم (نصف البشرية) في مجال الذكاء الصناعي.


خطط طموحة وتقدم ثابت للريادة في شتى المجالات وكل التوفيق ان شاء الله..