The future of data science is here. The AI Data Scientist runs tests, builds models, and delivers insights in minutes—not days. From raw data to clear recommendations, fully autonomous and precise.
—> Autonomous LLM-powered agent for end-to-end data science
—> 6 Subagents: Cleaning, Hypothesis, Preprocessing, Feature Engineering, Model Training, Call-to-Action
—> Runs hypothesis testing, builds models, and outputs business-ready insights
—> Model-agnostic: works across GPT-4, LLaMA-3, PHI-4, etc.
—> Outperforms baselines, ablations show Hypothesis + Feature stages are critical
—> Converts results into plain-language strategies with KPIs & timelines
—> Example: found churn drivers like “customers with <2 products have 25% higher exit rates”
- A groundbreaking framework blending AI with telecom infrastructure. - It features device edge cloud collaboration, scalable familial models, real-time multimodal inference, draft-verify LLM decoding, and emergent intelligence through connectivity. - The future of AI is fast, adaptive, and everywhere.
GLM‑4.1V‑Thinking is here from Zhipu AI & Tsinghua University!
—> 9B-parameter vision-language model —> Excels in multimodal reasoning (STEM, long docs, UI, video) —> Beats or matches GPT‑4o & Qwen‑VL‑72B —> Trained with scalable reinforcement learning (RLCS) —> Now available on Hugging Face
China unveils Meteor‑1, the world’s first highly parallel optical computing chip, delivering a theoretical 2,560 TOPS at 50 GHz - challenging Nvidia’s top GPUs amid US export controls on advanced semiconductors.
How ChatGPT Affects Your Brain: MIT Reveals the Cognitive Cost of AI Writing :
▶️ MIT’s new study shows that using ChatGPT to write essays may come at a cognitive cost. ▶️ Researchers recorded brain activity (EEG), memory recall, and essay quality.
📉 Key Findings: ▶️ Lowest brain activity in ChatGPT users (weaker alpha/beta EEG signals) ▶️ Poor memory recall—83% couldn’t quote their own essays after using ChatGPT ▶️ Reduced authorship—participants felt the essays weren’t truly theirs ▶️ High satisfaction despite low engagement—AI makes people feel productive ▶️ Persistent cognitive disengagement even after stopping ChatGPT use ▶️ More generic writing—ChatGPT essays lacked novelty and originality
📚 Conclusion: ChatGPT helps with speed—but over-reliance may damage memory, creativity, and critical thinking over time. The study calls this “cognitive debt.”
đź’ˇ Best practice: Write first with your brain. Use AI to polish, not to replace your thinking.
Microsoft open-sourced Data Formulator - a no code data analysis tool!
Key features: • Drag-and-drop UI for visualisations • AI-powered insights and analysis • Creates new relevant fields automatically • Generates visuals beyond the original dataset
BREAKING: Apple researchers just dismantled the myth that AI “reasoning” models (like Claude, DeepSeek-R1, and o3-mini) can actually think. Turns out, they don’t reason- they just look like they do.
Paper : The Illusion of Thinking – Apple AI Research (June 2025)
Key Findings:
✅ Models tested: Claude 3.7, DeepSeek-R1, o3-mini ✅ Claim: These models can think — but Apple proves they can’t reason. ✅ Tested on puzzles like Tower of Hanoi, River Crossing, Blocks World. ✅ Collapse Point: All models fail at higher complexity — 0% accuracy. ✅ Even with the correct algorithm, models couldn’t follow instructions. ✅ Overthinking: On easy problems, they find correct answers then keep going and get it wrong.
Thinking effort drops as tasks get harder — they give up faster on harder problems. Conclusion: Current "thinking" AI's just memorise patterns — not real reasoning.
🚀 Exciting Insights from NVIDIA GTC Live Paris 2025!
Today, I had the privilege of meeting Jensen Huang in person, capturing a moment with a 📸. Beyond his technical brilliance, what truly stood out was his down-to-earth and approachable nature, showcasing true leadership that has propelled NVIDIA to its current status.
🔬 Evolution of Supercomputing:
- Modern supercomputers are now "AI factories," emphasizing efficiency over raw power. - Quantum computing applications are on the horizon within 2 years, not decades! - NVIDIA systems lead the Green Top 500, demonstrating the coexistence of sustainability and performance.
🏠Industrial AI Transformation:
- Siemens & Volvo showcase innovative practices: - Digital factory construction using Omniverse before physical build, saving significant downtime. - Adoption of digital wind tunnels over physical testing. - Robots programmed with natural language, eliminating the need for coding.
🌍 Europe's AI Strategy:
- Strength lies in world-class industrial data and software talent. - Challenge: Acquiring sovereign hardware capabilities. - Opportunity: Spearheading responsible AI development.
🤖 Open Source Momentum:
- European startups harness open models for global competitiveness. - Transitioning from AI prototypes to production remains a significant challenge. - The future envisions every developer evolving into an AI developer.
đź’ˇ Bold Predictions for 2025:
- Robots serving coffee at conferences. - Quantum computing addressing real-world challenges. - AI workflows scaling up for production. - Mainstream adoption of Edge AI in manufacturing.
In summary, Europe is not just keeping up but defining a path focused on industrial applications, open collaboration, and responsible innovation. What AI transformations are shaping your industry?
Mervin Praison
🚀 NVIDIA AI Free Workshop: Which Certification Should You Pick?
👉 Register now: nvda.ws/42O9oNi
Thinking about Generative AI certs? Free webinar today (Tue, Oct 7) with live Q&A, exam prep tips, and a special exam promo code.
You’ll learn:
• Which exam fits your role (Associate + NEW Professional)
• Practical prep strategies + sample questions
• Agentic AI & LLM certs overview
Live sessions:
• EMEA/APAC: 9:00 a.m. CET / 12:30 p.m. IST / 3:00 p.m. SGT
• Americas/LATAM: 9:00 a.m. PT / 1:00 p.m. BRT
Speakers: Gary Fluitt, Aeshah Hadges, Behzad Torkian
#NVIDIA #GenerativeAI #Certification #LLM #AgenticAI #Webinar #AIJobs #Developers
7 months ago | [YT] | 4
View 0 replies
Mervin Praison
The AI Data Scientist is here!
The future of data science is here. The AI Data Scientist runs tests, builds models, and delivers insights in minutes—not days. From raw data to clear recommendations, fully autonomous and precise.
—> Autonomous LLM-powered agent for end-to-end data science
—> 6 Subagents: Cleaning, Hypothesis, Preprocessing, Feature Engineering, Model Training, Call-to-Action
—> Runs hypothesis testing, builds models, and outputs business-ready insights
—> Model-agnostic: works across GPT-4, LLaMA-3, PHI-4, etc.
—> Outperforms baselines, ablations show Hypothesis + Feature stages are critical
—> Converts results into plain-language strategies with KPIs & timelines
—> Example: found churn drivers like “customers with <2 products have 25% higher exit rates”
Check this link for detailed breakdown: x.com/MervinPraison/status/1960286364076843189
Paper: arxiv.org/abs/2508.18113
8 months ago | [YT] | 78
View 5 replies
Mervin Praison
🚨 China Telecom’s TeleAI introduces AI Flow
- A groundbreaking framework blending AI with telecom infrastructure.
- It features device edge cloud collaboration, scalable familial models, real-time multimodal inference, draft-verify LLM decoding, and emergent intelligence through connectivity.
- The future of AI is fast, adaptive, and everywhere.
For detailed breakdown check this link : x.com/MervinPraison/status/1948080225587523989
đź“– Read the paper: arxiv.org/abs/2506.12479
10 months ago | [YT] | 57
View 0 replies
Mervin Praison
GLM‑4.1V‑Thinking is here from Zhipu AI & Tsinghua University!
—> 9B-parameter vision-language model
—> Excels in multimodal reasoning (STEM, long docs, UI, video)
—> Beats or matches GPT‑4o & Qwen‑VL‑72B
—> Trained with scalable reinforcement learning (RLCS)
—> Now available on Hugging Face
Check this link for detailed breakdown : x.com/MervinPraison/status/1940689732163301834
Paper : arxiv.org/abs/2507.01006
10 months ago | [YT] | 75
View 3 replies
Mervin Praison
Couldn’t believe that I am tending at No. 1 on GitHub for developing PraisonAI
Do try PraisonAI if you haven’t . Loads of features recently got released
10 months ago | [YT] | 254
View 20 replies
Mervin Praison
China unveils Meteor‑1, the world’s first highly parallel optical computing chip, delivering a theoretical 2,560 TOPS at 50 GHz - challenging Nvidia’s top GPUs amid US export controls on advanced semiconductors.
For detailed breakdown : x.com/MervinPraison/status/1937739840012865588
Read more : www.scmp.com/news/china/science/article/3315496/li…
11 months ago | [YT] | 58
View 1 reply
Mervin Praison
How ChatGPT Affects Your Brain: MIT Reveals the Cognitive Cost of AI Writing :
▶️ MIT’s new study shows that using ChatGPT to write essays may come at a cognitive cost.
▶️ Researchers recorded brain activity (EEG), memory recall, and essay quality.
📉 Key Findings:
▶️ Lowest brain activity in ChatGPT users (weaker alpha/beta EEG signals)
▶️ Poor memory recall—83% couldn’t quote their own essays after using ChatGPT
▶️ Reduced authorship—participants felt the essays weren’t truly theirs
▶️ High satisfaction despite low engagement—AI makes people feel productive
▶️ Persistent cognitive disengagement even after stopping ChatGPT use
▶️ More generic writing—ChatGPT essays lacked novelty and originality
đź“š Conclusion:
ChatGPT helps with speed—but over-reliance may damage memory, creativity, and critical thinking over time. The study calls this “cognitive debt.”
đź’ˇ Best practice:
Write first with your brain. Use AI to polish, not to replace your thinking.
▶️ Detailed breakdown : x.com/MervinPraison/status/1935521474103898268
đź“„ Full paper: arxiv.org/abs/2506.08872
11 months ago | [YT] | 61
View 5 replies
Mervin Praison
Microsoft open-sourced Data Formulator - a no code data analysis tool!
Key features:
• Drag-and-drop UI for visualisations
• AI-powered insights and analysis
• Creates new relevant fields automatically
• Generates visuals beyond the original dataset
Perfect for analysts, PMs & no-code builders!
Check this out : github.com/microsoft/data-formulator
11 months ago | [YT] | 48
View 1 reply
Mervin Praison
BREAKING: Apple researchers just dismantled the myth that AI “reasoning” models (like Claude, DeepSeek-R1, and o3-mini) can actually think. Turns out, they don’t reason- they just look like they do.
Paper : The Illusion of Thinking – Apple AI Research (June 2025)
Key Findings:
âś… Models tested: Claude 3.7, DeepSeek-R1, o3-mini
✅ Claim: These models can think — but Apple proves they can’t reason.
âś… Tested on puzzles like Tower of Hanoi, River Crossing, Blocks World.
✅ Collapse Point: All models fail at higher complexity — 0% accuracy.
✅ Even with the correct algorithm, models couldn’t follow instructions.
âś… Overthinking: On easy problems, they find correct answers then keep going and get it wrong.
Thinking effort drops as tasks get harder — they give up faster on harder problems.
Conclusion: Current "thinking" AI's just memorise patterns — not real reasoning.
For detailed breakdown check this out : x.com/MervinPraison/status/1933395023116333103
11 months ago | [YT] | 57
View 14 replies
Mervin Praison
🚀 Exciting Insights from NVIDIA GTC Live Paris 2025!
Today, I had the privilege of meeting Jensen Huang in person, capturing a moment with a 📸. Beyond his technical brilliance, what truly stood out was his down-to-earth and approachable nature, showcasing true leadership that has propelled NVIDIA to its current status.
🔬 Evolution of Supercomputing:
- Modern supercomputers are now "AI factories," emphasizing efficiency over raw power.
- Quantum computing applications are on the horizon within 2 years, not decades!
- NVIDIA systems lead the Green Top 500, demonstrating the coexistence of sustainability and performance.
🏠Industrial AI Transformation:
- Siemens & Volvo showcase innovative practices:
- Digital factory construction using Omniverse before physical build, saving significant downtime.
- Adoption of digital wind tunnels over physical testing.
- Robots programmed with natural language, eliminating the need for coding.
🌍 Europe's AI Strategy:
- Strength lies in world-class industrial data and software talent.
- Challenge: Acquiring sovereign hardware capabilities.
- Opportunity: Spearheading responsible AI development.
🤖 Open Source Momentum:
- European startups harness open models for global competitiveness.
- Transitioning from AI prototypes to production remains a significant challenge.
- The future envisions every developer evolving into an AI developer.
đź’ˇ Bold Predictions for 2025:
- Robots serving coffee at conferences.
- Quantum computing addressing real-world challenges.
- AI workflows scaling up for production.
- Mainstream adoption of Edge AI in manufacturing.
In summary, Europe is not just keeping up but defining a path focused on industrial applications, open collaboration, and responsible innovation. What AI transformations are shaping your industry?
#AI #NVIDIA #GTCParis #GTC #Paris
11 months ago | [YT] | 193
View 7 replies
Load more