https://lucafiaschi.com/blog/ Insights on AI strategy, Bayesian methods, data science leadership, and GenAI from Luca Fiaschi. en-us https://lucafiaschi.com/blog/posts/intent-taste-masters-data-ai/ https://lucafiaschi.com/blog/posts/intent-taste-masters-data-ai/ On the Masters of Data, BI & AI podcast, I discussed why pricing is the most powerful data problem, what happens when AI eats software, and why intent and taste are the only competitive advantages left for humans. Tue, 31 Mar 2026 00:00:00 GMT https://lucafiaschi.com/blog/posts/synthetic-buyers/ https://lucafiaschi.com/blog/posts/synthetic-buyers/ A summary of my SAMI talk on synthetic consumers: using LLM-simulated buyer panels to pre-screen products, creatives, and messaging before spending real money. Wed, 25 Mar 2026 00:00:00 GMT https://lucafiaschi.com/blog/posts/data-ai-teams-benchmark-2026/ https://lucafiaschi.com/blog/posts/data-ai-teams-benchmark-2026/ I analyzed LinkedIn data on 213 tech companies to benchmark data and AI team composition. Role mix matters more than team size. Sat, 14 Mar 2026 00:00:00 GMT https://lucafiaschi.com/blog/posts/diverse-ideas-experiments/ https://lucafiaschi.com/blog/posts/diverse-ideas-experiments/ A Wharton paper found that Chain-of-Thought prompting nearly matches human idea diversity. I tested what happens when you scale from 1 agent to 22. The results changed how I build AI brainstorming tools. Sat, 14 Mar 2026 00:00:00 GMT https://lucafiaschi.com/blog/posts/bayesian-autoresearcher/ https://lucafiaschi.com/blog/posts/bayesian-autoresearcher/ Adapting Karpathy's autoresearch to Bayesian causal modeling, where the treatment effect has no ground truth and the agent must optimize a proxy. Wed, 11 Mar 2026 00:00:00 GMT https://lucafiaschi.com/blog/posts/llm-personality-survey/ https://lucafiaschi.com/blog/posts/llm-personality-survey/ Running the IPIP Big Five personality inventory across OpenAI, Google, xAI, and open source models shows that RLHF gives each family distinct behavioral fingerprints. The patterns are real, but the statistics are trickier than they look. Mon, 23 Feb 2026 00:00:00 GMT https://lucafiaschi.com/blog/posts/skills-registry-landscape/ https://lucafiaschi.com/blog/posts/skills-registry-landscape/ The AI agent skills ecosystem grew explosively. Security scanning followed. But nobody is checking whether skills actually work. That's the gap. Sun, 22 Feb 2026 00:00:00 GMT https://lucafiaschi.com/blog/posts/bayesian-hyperparameter-tuning-mmm/ https://lucafiaschi.com/blog/posts/bayesian-hyperparameter-tuning-mmm/ Combining Optuna's Bayesian optimization with PyMC-Marketing and CRPS to systematically tune Media Mix Model hyperparameters, instead of guessing. Sat, 20 Dec 2025 00:00:00 GMT https://lucafiaschi.com/blog/posts/bayesian-thinking-for-business/ https://lucafiaschi.com/blog/posts/bayesian-thinking-for-business/ Why traditional statistical methods often fail in business contexts, and how Bayesian approaches can help you make better decisions under uncertainty. Wed, 12 Nov 2025 00:00:00 GMT https://lucafiaschi.com/blog/posts/agentic-systems-bayesian-mmm/ https://lucafiaschi.com/blog/posts/agentic-systems-bayesian-mmm/ My Databricks Data + AI Summit talk on how agentic AI is moving marketing science from manual processes to collaborative decision-making, with two production systems we built on Databricks. Sun, 01 Jun 2025 00:00:00 GMT