Generating for draw.io used to work well (GPT-4 Turbo, Claude 3.5 Sonnet, …). Wrote a walk-through (€, 🇩🇪) here:
www.heise.de/ratgeber/Pro...
Posts by Nils Durner
... pitfalls to avoid. A select collection of Q&A is here: www.linkedin.com/pulse/qa-ai-... and I keep an updates document there: github.com/ndurner/Bild.... (2/2)
Article out: "Process Visualization with Generative AI: A Practical Test" (🇩🇪,€). I present AI-assisted workflows that build on BYOK/API-based LLM chat frontends, including a feature matrix, cost considerations, vendor availability and ... (🧵 1/2)
heise.de/-10266093
So this is less about summarization in the NLP sense of the word and more about RAG poisoning - and boils down to which search engines treat non-factual pages as sources of facts. Prompting ChatGPT with Search enabled and Perplexity, neither confirms Encanto 2, but Perplexity confirms Shawshank 2. 🤔
Those wishing to try the new Amazon Nova foundation models outside their Playground may find Amazon Bedrock setup non-intuitive, so I have compiled a step-by-step guide: ndurner.github.io/amazon-nova. It‘s 14+ steps!