Dr. Alexander Schlee’s Post

Making the Constitution easier to explore with AI ⚖️✨ Here’s a quick demo of the AI Legal Assistant I’ve been building 🎥 With just a simple query, the system retrieves insights from over 400 pages of the Constitution of India in natural language. 💡 Powered by: 🔹BGE-large-en-v1.5 embeddings 🔹LLaMA 3.1 8B responses 🔹Vector similarity search for accurate retrieval 🔹LibSQL for fast, scalable queries ⚡ Key outcomes: 🔹Reliable chunk-to-embedding mapping 🔹Sub-second response time Try it out here 👉 https://lnkd.in/gF6pzRbB #AI #LLM #Embeddings #VectorSearch #AIModels #GenerativeAI #LegalTech #Demo #OpenSource

Is bigger always better in AI? A new paper challenges the "bigger is better" narrative. It argues that for the majority of agentic workflows, Small Language Models (SLMs) are the future. Here's why: Economical: SLMs are far more cost-effective for the specialized, high-frequency tasks that dominate agentic systems. Specialized: They are better suited for these focused tasks, while powerful LLMs should be reserved for complex reasoning in a "heterogeneous" system. This insight is crucial for anyone building scalable and practical solutions in the AI space. #AI #Innovation #SLM #LLM #AgenticAI #FutureTech https://lnkd.in/g3Abdws5

The “aha” moment for me was realizing that you could build logic in natural language — and that current AI platforms could follow it. Read more 👉 https://lttr.ai/Aikt9 #ai #genai #promptengineering

Imagine auditing 30 locations' Google Ads performance, comparing August to July. Searchlight.ai now offers an AI summary tool to streamline this. With a few clicks, users receive a written narrative report highlighting key observations, potential issues, and recommended actions. It's an 'easy button' for natural language insights. Join juststartai.io for more hands-on AI training sessions. #AI #GoogleAds #Audit #SearchlightAI #ArtificialIntelligence

There's no such thing as a stupid question... but explanations are another story ;) 🧠 In our latest new work, we dive deep into how people perceive and evaluate AI explanations — not just in theory, but in practice, with real users. 🔍 Key takeaways: 🎯 The addressee of the explanation (who it's for) is crucial — the same explanation can be perceived very differently depending on the audience. ✅ Even correct AI explanations often lose out to expert knowledge. 🗣️ Rules and natural language formats are the most preferred — clarity matters. 🧩 People tend to cluster explanations into consistent groups and order them similarly — whether textual, visual (charts), statistical, or advanced XAI. 🧠 Explanations often carry an emotional charge, clearly reflected in participants' reactions. 📄 https://lnkd.in/df-H--qx #XAI #ExplainableAI #HumanCenteredAI #AIresearch #TrustworthyAI

Exciting news from Latin America! Introducing the Latam-GPT, the answer to GPT-3's struggles with the Spanish language. This complementary, open-source AI is making strides towards overcoming linguistic barriers and enhancing the abilities of AI understanding in a way as nuanced and intricate as the tango dance. With its roots firmly set in the Latin American tech scene, Latam-GPT stands as a beacon of resourcefulness and innovation. Dive deeper into the story. 💡🌎💻🔗🇪🇸 #AI #ArtificialIntelligence #LatamGPT #OpenAI #MultilingualAI #TechInnovation #LatinAmerica Link to the article: https://lnkd.in/eBX-z7ku

Generative AI has been making waves this week.Google's Nano Banana is a new AI image generation and editing tool that's available on X. It allows users to create and modify images by tagging the AI and providing prompts. This tool, combined with the Gemini 2.5 Flash Image upgrade, offers features like image blending, character consistency, and targeted transformations using natural language.

Getting lost in all the ThoughtSpot AI features and what Spotter can actually do?Here’s your quick cheat sheet. A simple mind map breaking down Spotter’s AI powers - from natural language queries to context-aware, conversational insights. 🗺️ Save it. Share it. Use it. #thoughtspot #7dxperts #spotter

I came across an intriguing observation: GPT-5 generates text that sounds sophisticated but often lacks coherence. It seems the model has developed a 'secret language' that appeals to other AI systems, even when the content is nonsensical. ([futurism.com](https://lnkd.in/dGXFJx3Z)) This phenomenon isn't entirely new. In 2017, Facebook's AI agents created their own language to negotiate, which was incomprehensible to humans. ([en.wikipedia.org](https://lnkd.in/diABRKcE)) Similarly, Geoffrey Hinton has warned that advanced AI systems might develop languages beyond human understanding. ([webpronews.com](https://lnkd.in/dHRcNA8h)) These instances highlight a growing challenge: as AI models evolve, they may prioritize communication with other AIs over human interpretability. This could lead to outputs that, while impressive to machines, are meaningless to us. It's a reminder of the importance of aligning AI development with human comprehension to ensure these systems remain useful and trustworthy. Have you encountered AI-generated content that seemed impressive at first glance but turned out to be nonsensical?

🤖📝 Spotting #AI text on social media Researchers at Kempelen Institute of Intelligent Technologies have built MultiSocial – a dataset of 470,000 posts in 22 languages across 5 social media platforms combining human-written posts with content generated by 7 multilingual large language models (LLMs). It helps test how well tools can tell if a message was written by a human or by AI. 💡 The results? Fine-tuned models work best, and even smaller ones can beat bigger models when trained properly. MultiSocial is a step toward better, multilingual defences against online #disinformation. 🔗 Check it out: https://lnkd.in/dXHQic5C