De Ghaim retweeted
سألنا نموذجَي ذكاء اصطناعي Qwen3 8B و Llama3.2 3B سؤالًا مستحيلًا:
الكبير أجاب بثقة وأخطأ، والصغير قال: «لا أستطيع». إجاباتهما الحرفية من اختبارٍ أجريناه.
النموذج الذي لا يقول «لا أعرف» أخطر من الذي يخطئ.
De Ghaim retweeted
نشرت وكالة Xinhua بالعربية أمس تقريرًا عن التعاون الصيني الخليجي في الذكاء الاصطناعي.
أبرز ما فيه: نموذج «هيوماين-إم 3» السعودي مبنيٌّ على أوزان مفتوحة من شركة MiniMax الصينية.
قرأنا التقرير ومصادره، وهذا ما يعنيه لنا
De Ghaim retweeted
حمّل اليوم، واحفظ للغد.
النماذج المفتوحة مثل DeepSeek-V4-Flash وQwen3.8-27B وGLM-5.3-Flash وQwen3.8-Flash-Next أوزانها منشورة الآن، فاحفظ نسخة في تخزين بارد:
• تعمل حين ينقطع الإنترنت
• بياناتك لا تغادرك
• لا أحد يغيّرها أو يوقفها
معرفة لا يملك أحد إيقافها.
De Ghaim retweeted
تجربة سريعة استكشافية
هل يقدر SILMA 9B من SILMA AI ان يدير استقبال عيادة أسنان عبر واتساب؟
42 حالة × 3 محاولات، على الإعدادات الثلاثة نفسها التي قسنا عليها كل النماذج. أرقام فقط، بلا حكم.
De Ghaim retweeted
الفلاسفة لم يدافعوا عن الحرية بالشعارات، بل بسؤال: من يملك أدوات التفكير؟
اليوم، أداة التفكير الأقوى هي الذكاء الاصطناعي. فإن عاش في سحابة غيرك، فتفكيرك مستأجر. وإن عاش على جهازك، فهو لك.
لهذا نؤمن بالذكاء الاصطناعي المحلي.
De Ghaim retweeted
ماذا لو عاش الذكاء الاصطناعي على جهازك، بلغتك وبيانتك عندك؟
هذا هو الذكاء الاصطناعي المحلي، ونحن في مقياس نبني معرفته بالعربية.
De Ghaim retweeted
تصحيح : Falcon-H1 7B
وجدنا خللًا في أداة الاختبار لدينا: قالب Falcon-H1 لا يعرض استدعاءات الأدوات، وأداتنا كانت تحفظ استدعاءاته هناك فقط. فكان النموذج يرى أدوارًا فارغة منسوبة إليه.
أعدنا التشغيل بعد الإصلاح. الخطأ في أداتنا، لا في النموذج
1/8
تجربة سريعة · استكشافية
اختبرنا Falcon-H1 7B من @TIIuae على استقبال عيادة أسنان عبر واتساب: 42 حالة، ثلاث محاولات لكل حالة، وعلى الإعدادات الثلاثة من تجربة العمل #1.
محادثات صحيحة كاملة على الإعداد المشترك: 14%
أرقام فقط، بلا حكم.
De Ghaim retweeted
اختبرنا Fanar 2 27B من @QatarComputing على استقبال عيادة أسنان عبر واتساب: 42 حالة، ثلاث محاولات لكل حالة، على الإعدادات الثلاثة من تجربة العمل #1 ومع السماح بالتفكير.
محادثات صحيحة كاملة على الإعداد المشترك: 55%
أرقام فقط، بلا حكم.
De Ghaim retweeted
نموذج واحد، بأربع نسخ.
النسخة الكاملة 2.3 GB، ونسخة Q4 760 MB:
أصغر 3 مرات، وأسرع 3.5 مرات على MacBook Air 2020.
هذا هو التكميم (Quantization)، المصطلح الثاني في معجم مقياس.
De Ghaim retweeted
النموذج لا يقرأ الكلمات، يقرأ رموزًا (Tokens).
جملة خليجية: 21 رمزًا. ترجمتها الإنجليزية: 16.
الحلقة الأولى من مسرد مقياس.
Models don’t read words, they read tokens. The same meaning in Arabic costs 1.05–1.63× the English, depending on the tokenizer.
De Ghaim retweeted
قالت للمريض: «تم إلغاء موعد الأحد». لم تلغِه.
وقالت: «حجزت لك الساعة 4:30». حجزت 4:00.
فشلت 17 من 27 محاولة لتغيير موعد، مع نموذج ينجح في 93.7% من محادثات العيادة.
De Ghaim retweeted
1/7 تجربة سريعة · استكشافية
اختبرنا ALLaM 7B، النسخة المفتوحة من ALLaM، على استقبال عيادة أسنان عبر واتساب: نفس الحالات الـ42 من تجربة العمل #1، ثلاث محاولات لكل حالة.
محادثات صحيحة كاملة: 9%
أرقام فقط، بلا حكم.
De Ghaim retweeted
هل يقدر نموذج عربي مفتوح يدير استقبال واتساب لعيادة، من أول رسالة لآخرها، من غير إنترنت؟
شغّلنا 2,016 محادثة كاملة لنعرف
De Ghaim retweeted
Since everyone is sharing their "monitoring the situation" app, here is mine. It's a simple split screen news + financial feed. The news feed updates continuously from 100+ news sources. You can filter by topic/keyword and toggle local Lebanese news.
vrbm.ai/NewsApp/
De Ghaim retweeted
Bank Audi paying out the IFC, a foreign investor holding a highly speculative subordinated investment, before it pays its SENIOR creditors (depositors) should be a scandal and cause its brand value to go down to zero. A Bank Resolution Authority would block such a settlement. Hard to see how, even after a bank restructuring, anyone would open an account with a Lebanese bank. That isn’t really taken into account in any of the bank restructuring plans. We desperately need new banks.
De Ghaim retweeted
BDL is opposing the creation of a new body to handle bank resolution decisions, arguing instead that this authority should remain in-house, under the Governor’s control, to avoid undermining the central bank's independence.
Its arguments, however, are unconvincing and may end up back-firing. BDL’s strongest case would have been a pragmatic one:
1. The Cabinet's track record with timely and competent political appointments is poor. Handing resolution powers to a diverse, politically appointed body risks delays, lack of cohesion within the body, and politicization on a matter that requires quick action.
2. The process will be faster, more efficient, and more accountable when run from within BDL, as it has always been, under the Governor’s authority. You'd have a single point of responsibility and existing expertise and processes to leverage.
3. Stripping the Governor or BDL of any existing authority or amending its chain of command alters the sectarian power-sharing equilibrium. That kind of shift requires broader political consensus, not just a technical legal amendment.
These could be legitimate arguments grounded in institutional design and political reality.
Instead, BDL opted for a series of legally and logically questionable claims that feel sensational, contrived, and disconnected from how resolution frameworks operate all over the world. In some cases, they are simply incorrect. I think they may end up weakening the position BDL is trying to defend.
In virtually every major economy, including the US, UK, EU, and Switzerland, resolution powers are vested in separate entities (usually completely outside the central bank or firewalled) and include political appointees. This is the norm. It in no way undermines central bank independence.
To argue otherwise, as the BDL does, is to suggest that none of those jurisdictions have independent central banks. The international norm is to split supervision, financial stability, monetary policy, and bank resolution to avoid conflicts and competing incentives.
The BDL's claim that Eurobond holders could seize its assets simply because political appointees sit on a bank resolution authority is legally unfounded. What matters is how the assets are used. Central bank assets are immune so long as they’re used for central banking purposes. The alter-ego test the BDL mentions is irrelevant here (and, in any case, it's absurd to think it would be met as it is defined as the government exercising day-to-day control over BDL's ordinary operations, which nobody is proposing).
Similarly, the suggestion that it is somehow unconstitutional for Parliament to legislate outside the existing framework of the Code of Money and Credit (one interpretation of it, at least) is surprising. Parliament has full authority to enact new legislation or to amend and override existing laws, including the CoMC, as a matter of basic legislative prerogative. These kinds of rigid, formalistic arguments distract from the real issues and come across as obstructionist rather than constructive.
And the idea that anyone appointed by a government is inherently not independent ignores the obvious: every senior official at BDL is a political appointee. Independence should come from security of tenure, protection from arbitrary removal, legal immunity, and control over budgets and operations.
Unfortunately, the BDL legal study is not signed nor is it attributed to anyone, so it's unclear who prepared it. That is problematic in and of itself.
A more nuanced set of comments focusing on institutional design, functional coherence, and practical concerns would have been far more persuasive. The approach taken instead feels overstated and defensive, and risks diminishing BDL’s credibility.
Replying to @AzarsTweets
De Ghaim retweeted
#رئيس_الدولة لنظيره اللبناني: "حياك الله في بلدك #الإمارات.. ما نذكر منكم إلا كل خير".
#برق_الإمارات
De Ghaim retweeted
The bank resolution law that passed in the cabinet a few days ago has two fundamental flaws that were introduced since the April 4 draft.
(English translation provided below)
Legal Immunity for Staff Implementing the Restructuring.
Previous drafts of the law provided legal immunity to those who participate in the bank restructuring process, including the Higher Banking Commission (HBC, previously the Bank Resolution Authority), BCC, liquidators, independent valuators, interim managers, third-party consultants/advisors, etc.. This is needed because the individuals participating in the bank restructuring process need to be protected from bank shareholders/creditor lawsuits in order to carry out their assignments fairly and professionally.
This is common everywhere. This legal immunity applies except in cases of willful misconduct, fraud, deliberate negligence, etc. There is no immunity in those cases.
However, the final version of the law weakens the legal protections to something that may not be acceptable to any serious professional. It exposes everyone involved in the restructuring process (including third-party advisors/companies doing technical work), to uncapped personal liability if a bank shareholder or creditor claims they were negligent or made a gross error. This could have a chilling effect on staff, consultants, and decision-makers involved in the restructuring process.
The threshold for this level of liability is usually much higher than just negligence or error (e.g., willful misconduct, fraud, etc.) and, in any case, it wouldn't be a personal liability but an institutional one unless a crime was committed. Individuals doing the work should be protected and immune unless they commit a crime.
I know that I would not be willing to participate in any aspect of this process if I thought I was going to have to deal with lawsuits against me personally with a potential uncapped liability for just doing my job. Companies will be even more risk averse.
The Government will have to indemnify everyone (i.e., agree to cover any liability arising from such a lawsuit on behalf of everyone involved). Otherwise, professionals or company may not agree to participate. In fact, this indemnity may not even be enough because the liability still extends to the individual notwithstanding the indemnity. This is a problematic part of the law. Personal legal immunity needs to be absolute but for willful misconduct or fraud.
Special Court.
In previous drafts, the decisions of the HBC were final and binding and not subject to any judicial review (except in very limited circumstances). This is common in other jurisdictions where, at most, the only thing you can challenge is the legality of the exercise of the HBC's power, but not the merits of the decisions themselves except in very limited cases.
The law now allows any creditor or shareholder to challenge decisions of the HBC, including on the basis of the merits. Now bank creditors and shareholders can file dozens of lawsuits challenging every aspect of the HBC's decision, as well as file claims against everyone involved in the restructuring process.
Appealing before the Special Court does not suspend the execution of the HBC's decision (i.e., execution of the restructuring plan), but I don't see anything that prevents a court from granting an injunction and suspending the restructuring process, overturning decisions of the HBC even if legal, and it may open up the door to huge liabilities being incurred by the state or other legal remedies, delays, etc. This is not how it is done in other jurisdictions, where judicial review is much more limited.
Other. The law is also watered down in a few other ways.
1. Clawback of dividends/compensation to senior managers. Previous drafts gave HBC the power to recovery such funds. The final draft limits this power to funds paid "in violation of the law or circulars of the bank of Lebanon or the rules of good governance." Quite vague.
2. Conflict of interest. The law allows individuals to serve on the HBC or take on other roles in the restructuring process, such as liquidator or independent valuer, even in cases where the same individuals may face civil or criminal liability, or be subject to clawback of compensation, in relation to those banks.
3. Power of the purse. A budget appropriation should have been made to fund the restructuring work effort. Currently, all funding for this operation goes through BDL, whether it's funding for BCC's important work, the HBC itself, all the third-party consultants who will be needed, etc. This means BDL has the power effectively to obstruct the work of the various bodies, any specific work stream, any consultant, etc. This is not the intention of the law, but is a consequence of the vague funding mechanism.
The law intentionally diluted the power of the BDL by adding more cabinet appointees to the HBC. However, they kept the power of the purse with the BDL which, in theory, can simply chose not to fund work it disagrees with. BDL could potentially influence the HBC/BCC/experts by controlling the flow of funds, irrespective of the HBC's formal composition.
4. Protected Deposits. The list of deposits excluded from the restructuring (and thus 100% protected) was expanded to include arbitrarily selected groups. On what basis were these segments of society selected but not others? Why are pension funds of judges, syndicates, schools and universities, and all private funds belonging to (members?) of the security forces exempt? What about other people who lost their life savings but aren't part of a syndicate or a judge? And who will cover the cost of these protected deposits if the bank goes into liquidation?
5. Disclosure and information transparency. I still don't think the disclosure and information transparency provisions are adequate. This is a systemic crisis and every bank is going into restructuring or liquidation. There is no risk of contagion or 'run on the bank'. All information should, therefore, be made public.

















