Mohamad
Amin Al Hajj

✍️ 𝐎𝐟𝐟𝐢𝐜𝐢𝐚𝐥: 𝐂𝐨𝐧𝐭𝐫𝐚𝐜𝐭 𝐬𝐢𝐠𝐧𝐞𝐝. Delighted to share that I’ve officially taken on the role of 𝗔𝗜 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 𝗠𝗮𝗻𝗮𝗴𝗲𝗿 within SKF Group’s 𝑨𝑰 𝑪𝒆𝒏𝒕𝒆𝒓 𝒐𝒇 𝑬𝒙𝒄𝒆𝒍𝒍𝒆𝒏𝒄𝒆, continuing also as 𝐒𝐞𝐧𝐢𝐨𝐫 𝐀𝐈 𝐒𝐩𝐞𝐜𝐢𝐚𝐥𝐢𝐬𝐭 within 𝑫𝒊𝒈𝒊𝒕𝒂𝒍 𝑬𝒏𝒈𝒊𝒏𝒆𝒆𝒓𝒊𝒏𝒈 𝒂𝒏𝒅 𝑰𝒏𝒏𝒐𝒗𝒂𝒕𝒊𝒐𝒏. As a football player at heart, the shirt’s on the table, the ink's dry, and it is time to score AI goals. #AI #Leadership #Innovation #MachineLearning #DigitalEngineering #CareerGrowth #SKF #TeamSKF #ArtificialIntelligence #AIDevelopment #FutureOfWork #CareerMilestone #NextChapter #BuildingTheFuture #Teamwork

Every Sunday, together with an amazing team of volunteers, we cook meals at Café Averechts — where every euro goes directly to charities around the world. Just good food for a good cause. It’s the simple things: chopping vegetables together, the quiet focus, the shared laughter, and the sense that we’re doing something that matters. It reminds me that kindness doesn’t need grand gestures — just community and heart. And every time I walk out of that kitchen, I feel lighter, grateful, and a little more connected to the world. Here’s a glimpse into our Sundays. #Volunteering #Charity #Community #GivingBack #SocialGood #Impact #Utrecht #FoodForGood #Sustainability #MakeADifference

𝘛𝘩𝘦 𝘝𝘢𝘭𝘶𝘦 𝘰𝘧 𝘈𝘐 𝘌𝘹𝘱𝘦𝘳𝘵𝘪𝘴𝘦 𝘣𝘳𝘦𝘢𝘬𝘴 𝘥𝘰𝘸𝘯 𝘵𝘰 𝘰𝘯𝘦 𝘸𝘰𝘳𝘥. 𝙉𝙊 - 𝘓𝘦𝘵 𝘮𝘦 𝘦𝘹𝘱𝘭𝘢𝘪𝘯 𝗡𝗼, 𝘁𝗵𝗶𝘀 𝗶𝘀 𝗻𝗼𝘁 𝗮𝗻 𝗔𝗜 𝗽𝗿𝗼𝗷𝗲𝗰𝘁. 𝗡𝗼, 𝘄𝗲 𝗱𝗼𝗻’𝘁 𝗻𝗲𝗲𝗱 𝗮𝗻 𝗟𝗟𝗠 𝘁𝗼 𝘀𝗼𝗹𝘃𝗲 𝘁𝗵𝗶𝘀. 𝗡𝗼. 𝗡𝗼. 𝗡𝗼. If 𝘈𝘐 is the answer to all your questions, then your “AI experts” might not be what they seem to be. Being an AI specialist isn’t just about building models — it’s about knowing 𝘄𝗵𝗲𝗻 𝗻𝗼𝘁 𝘁𝗼. Here is a blog post of me explaining more: https://lnkd.in/eME-DTNF Let me give you two real examples. 📦 𝗘𝘅𝗮𝗺𝗽𝗹𝗲 1 — 𝗖𝗼𝘂𝗻𝘁𝗶𝗻𝗴 𝗕𝗼𝘅𝗲𝘀 𝗼𝗻 𝗮 𝗣𝗮𝗹𝗹𝗲𝘁 Someone once came to us with a “visionary” idea: “Let’s use AI to count the boxes on each pallet!” They had seen it in a video and wanted to “bring AI to production.” Sounds fancy — but totally unnecessary. We sat down with the people on the floor, grabbed a coffee, and simply asked: “How do you currently count them?” Turns out, they already had the answer. Weigh the pallet. Weigh one box. Divide. Simple math. No AI, no cameras, no cloud, no waste. ☕ 𝗘𝘅𝗮𝗺𝗽𝗹𝗲 2 — 𝗧𝗵𝗲 𝗖𝗼𝗳𝗳𝗲𝗲 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 A coffee machine checks if a cup is under the nozzle before pouring. Should we use computer vision to detect the cup? No. Just place a basic sensor under the dispenser. If it detects something, pour. Done. That simple “𝗡𝗼” is not so simple. It saves time, money, compute, and most importantly — 𝘧𝘰𝘤𝘶𝘴. Because every “No” to the wrong use case is a “Yes” to scaling the right one. True AI expertise is knowing that 𝗔𝗜 𝗶𝘀 𝗻𝗼𝘁 𝘁𝗵𝗲 𝗮𝗻𝘀𝘄𝗲𝗿 — it’s just one of the tools. 💬 𝘞𝘩𝘦𝘯 𝘸𝘢𝘴 𝘵𝘩𝘦 𝘭𝘢𝘴𝘵 𝘵𝘪𝘮𝘦 𝘺𝘰𝘶 𝘴𝘢𝘪𝘥 “𝘕𝘰” 𝘵𝘰 𝘢𝘯 𝘈𝘐 𝘱𝘳𝘰𝘫𝘦𝘤𝘵 — 𝘢𝘯𝘥 𝘸𝘦𝘳𝘦 𝘨𝘭𝘢𝘥 𝘺𝘰𝘶 𝘥𝘪𝘥? #AI #MachineLearning #DigitalTransformation #Leadership #Innovation #ResponsibleAI #Experience

Hallucinations are Dead — well, sooner that you think. They’re fixable incentives and information gaps (apparently) Two fresh takes worth your time: OpenAI: Our training & evaluation norms reward confident guessing. Let models abstain and penalize wrong answers more than “I don’t know.” Hassana Labs (Leon Chlon, PhD): Turn hallucination risk into a controllable knob. Measure information sufficiency (ISR) per query and answer only when the evidence clears a threshold. I wrote a short blog briefly explaining both and how to apply them (together) in practice. https://lnkd.in/eTQNpZbu Sidenote: Reading more about Hassana Labs' story and their work pushes me even more to giving and #Ai4Good Hugging Face Looking forward to having these tools added ! #LLM #AI #Hallucinations #Evaluation #RAG #TrustworthyAI

Why does ChatGPT sometimes act like it rolls dice — even when you give the same input ?👉 Probably some desired randomness. But why does the model behind still give different responses when we remove this by design randomness ? I just wrote a blog unpacking this — what’s myth, what’s real, and why reproducibility in LLMs is harder than it seems. 👉 [https://lnkd.in/epHCdm2G] In short: Most explanations blame “complexity” or “randomness in AI,” but that’s not the full story. LLMs are deterministic at the math level. The surprises come from: Designed randomness: temperature, top-p, top-k sampling. (You can control if you do API calls) System nondeterminism: GPU kernels, floating-point math quirks, and batch inference optimizations. (You can not control unless you are designing the API or running the LLMs locally) A recent paper by Thinking Machines shows how batching and kernel choices can break determinism even when sampling is turned off. More AI insights coming up — stay tuned! Oh yeah and this is a picture of me sailing some time ago.

Yesterday, I pushed myself to my first 10 kilometers run through the beautiful streets of Utrecht, proudly carrying the Lebanese flag on my back. 🏃‍♂️🇱🇧 With each step, my thoughts were with my homeland, facing challenging times amidst war, a way to channel my hopes and stand in solidarity with those back home. I'm inspired to continue this journey and embrace running more often. It's incredible how a simple run can bring so much clarity and connection. How do you find strength and stay connected during tough times? I'd love to hear your stories. PS: If you know someone from that region, reach out to them. #Lebanon #RunningForHope #Utrecht #Community #Peace #Run