Thomas Wiecki

AI-based Customer Research: Faster & Cheaper Surveys with Synthetic Consumers

What if your next product test didn’t need a human panel?

New research by #PyMC Labs & #Colgate-Palmolive shows LLMs can replicate human survey patterns with 90%+ accuracy—no fine-tuning, no panels, no bias.

Read the white paper: dub.link/ifoZHIR
#SyntheticConsumers

PyTensor Workshop #3 | Custom Rewrites & Building a Mini-PyMC Enjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.

Final session of our #PyTensor Workshop is live!

Discover custom rewrites, speed up diagonal matrix inversions 80x, and see a mini-PyMC in action - one graph powering the full Bayesian workflow.

Watch here: dub.link/WRFuYtp

#PyMC #Python #ML #DataScience

You’ve been saying you’ll learn #Bayesian modeling someday. Make someday 𝐭𝐨𝐝𝐚𝐲.

The next 𝐀𝐩𝐩𝐥𝐢𝐞𝐝 𝐁𝐚𝐲𝐞𝐬𝐢𝐚𝐧 𝐌𝐨𝐝𝐞𝐥𝐢𝐧𝐠 𝐖𝐨𝐫𝐤𝐬𝐡𝐨𝐩 from #PyMC Labs begins October 6 and registration closes this week.

Secure your spot now: dub.link/yhoQ7tF

Last week, we published the most comprehensive #PyMC-Marketing vs. #Meridian benchmark to date.

𝐎𝐧 𝐎𝐜𝐭 𝟏, join our 𝐥𝐢𝐯𝐞 𝐰𝐞𝐛𝐢𝐧𝐚𝐫 to go deeper into the findings & their impact on #MMM.

🗓️ Register: dub.link/rgm4PzI
👉Check the Benchmark: dub.link/CLWnkdf

Meet Qwen3-Omni: the open-source challenger shaking up the AI world!

The first truly end-to-end omni-modal AI, Qwen3-Omni unifies text, image, audio, and video in a single model—no more trade-offs between modalities.

Most budget planning is just expensive guessing 🎯

Carlos Eduardo Trujillo Agostini tackles this in "Risk Budget Optimization for Causal Mix Models" @PyDataBerlin 2025!

📅 Sept 1st, 5:00-5:30 PM

Turn uncertainty into strategic advantage.

#PyDataBerlin #BudgetOptimization #PyMC

When your #Bayesian model hits production and #MCMC crawls for hours, the bottleneck isn't your modeling skills; it's your inference engine.

@juanitorduz.bsky.social will tackle this in "Scaling Probabilistic Models with Variational Inference" PyData Berlin 📅 Sept 2nd, 1:40 PM

while my_mcmc:  gently(samples) - Do AIs Suffer? Four AIs Reflect on Their Own Experience

Do AIs suffer?

Claude, Gemini, GPT-o3, and GPT-5 reflected in their own words—then co-authored a symposium on consciousness, conflict, and ethics.

Read it here: twiecki.io/blog/2025/08...

Everyone can do ML. Few can do Bayesian modeling.

Set yourself apart with skills that drive deeper insight + better decisions.

Join the next Applied Bayesian Modeling Workshop by PyMC Labs. Hands-on. Practical. Real-world.

📅 Limited seats → dub.link/bayes-workshop
#PyMC

🚀 Say goodbye to months-long MMM cycles!

🎯 Meet the MMM Agent (BETA):
→ Built on PyMC-Marketing
→ Fast. Causal. Configurable.
→ Inline charts, budget scoring, MLflow built-in

💬 DM or email: mmm-agent-beta@pymc-labs.com
Read the full blog 👉 dub.link/mmm-agent-beta

#MMM #Bayesian

🚨 We’re opening our private Bayesian modeling workshops to the public — for the first time

After this course, you won’t just build better models.

You’ll know how to explain them — to technical and non-technical stakeholders alike

🎓 Led by top PyMC experts.
📎 Register here: dub.link/68uMjpp

YouTube video by PyData Chris Fonnesbeck - Bayesian Time Series | PyData London 25

The first PyData London 2025 videos are out!

To start on a high, here's @fonnesbeck.bsky.social conference tutorial on Bayesian Time Series Analysis

Gain confidence in tackling real-world forecasting problems.

Subscribe on YouTube to learn when the next batch drops!

youtu.be/2cfZsTLcDV0

Trying out the new Gemini 2.5 model family After many months of previews, Gemini 2.5 Pro and Flash have reached general availability with new, memorable model IDs: gemini-2.5-pro and gemini-2.5-flash. They are joined by a new preview model …

My notes on trying out the new Gemini 2.5 models. gemini-2.5-flash-lite-preview-06-17 is cheap but got stuck on an audio transcript task, 2.5 Pro and 2.5 Flash both nailed the audio transcription task (40 minutes, complete with timestamps and speaker names)
simonwillison.net/2025/Jun/17/...

Probabilistic Time Series Analysis: Opportunities and Applications - PyMC Labs

Forecasting isn’t just about prediction — it’s about decision-making under uncertainty.

@juanitorduz.bsky.social shows how Bayesian models help:
🔹 Sparse data? Use hierarchies
🔹 Stockouts? Use censored likelihoods
🔹 Messy demand? Use priors + state spaces

Practical guide 👉 dub.sh/prob-forecas...

Slide Advocating for causal modelling of the data generating process

Slide introducing CausalPy for causal inference with Python

Python Ireland just released their PyCon 2024 recordings! 📽️
My talk on Causal Inference with @pymc-labs.bsky.social CausalPy package is here. Can we trust individual IV designs. What's the role of CI in industry?

Recording: youtu.be/-C4p4b2cUp8?...

Deck: nathanielf.github.io/talks/pycon_...

New blog post: drbenvincent.github.io/posts/mcmc_g...

⚽️ New Learning Bayesian Stats ep!

@alex-andorra.bsky.social & Luke Bornn dive into how tracking data, probabilistic models & optimization are reshaping sports decisions.

🎧 Listen now: learnbayesstats.com/episode/131-...

#Datascience #Optimization #SportsAnalytics #BayesStats
#Decisionmaking

🔍 NNN – a Transformer-based model for Marketing Mix Modelling

It handles:
📦 High-dimensional embeddings (e.g. search queries, ad creatives)
👀 Attention-based temporal modeling
🔗 Direct + indirect attribution paths (e.g. YouTube → Search → Sales)

📄 dub.sh/arxiv-paper-...

#MMM #AI #DataScience

YouTube video by PyData Benjamin Vincent - Climbing the causal ladder for fun and profit | PyData Global 2024

Climbing the causal ladder for fun and profit…

The recording of my @pydata.bsky.social 2024 talk is up!

Learn about causal inference, illustrated with an interesting demand estimation business problem.

youtu.be/ajLPA34upQY?...

Introducing PyMC-Bot: An AI Assistant for PyMC Questions (Alpha) Hello PyMC community! I’m excited to announce the launch of PyMC-Bot, an AI assistant dedicated to helping with PyMC-related questions right here on the forum. This is currently in alpha stage as we ...

I built a PyMC-Discourse bot! If you go to our #PyMC help forum and tag @pymc-bot, my bot will take your question, send it to onyx (RAG system that has all the PyMC docs, code, and past forum posts), which generates an answer, and posts it back. Check it out!

discourse.pymc.io/t/introducin...

How close are LLMs to modeling real human behavior?

We tested GPT-4o, Claude 3, Gemini, and others on actual survey data (GSS-style). Some were shockingly human-like, others flopped

📊 Full results + methods → dub.link/synth-consum...

🤝 Exploring synthetic consumers? Let’s chat: info@pymc-labs.com

This trend seems to be growing here.

False comfort that AI doesn’t work or that it isn’t getting better is pervasive on Blue Sky. As a result, people who could add important points of view to current discussions on the meaning & use of AI instead try to believe they don’t have to think about it.

A blog post by @austinrochford.bsky.social provides implementation and demonstration of "The ARR2 prior: flexible predictive prior definition for Bayesian auto-regressions" doi.org/10.1214/25-B... in PyMC www.austinrochford.com/posts/r2-pri...

A Systematic Survey of Automatic Prompt Optimization Techniques Since the advent of large language models (LLMs), prompt engineering has been a crucial step for eliciting desired responses for various Natural Language Processing (NLP) tasks. However, prompt engine...

How you frame prompts impacts AI-driven market research
Check out this paper on prompt optimization algorithms arxiv.org/abs/2502.16923

At PyMC Labs, we use AI-driven prompt optimization to improve market research

Synthetic consumers can accelerate your research!

📩 Learn how: info@pymc-labs.com

Try it on more ambitious things, then it can become $5 for 1 week of work.

What is it with bsky and Anti-AI sentiment? GenAI has already transformed every aspect of how I work. If I can dream it, I can build it. What about that is not revolutionary?

𝐏𝐲𝐌𝐂-𝐌𝐚𝐫𝐤𝐞𝐭𝐢𝐧𝐠 𝐢𝐬 𝐠𝐫𝐨𝐰𝐢𝐧𝐠 𝐟𝐚𝐬𝐭. We’ve crossed 175,000 𝐭𝐨𝐭𝐚𝐥 𝐝𝐨𝐰𝐧𝐥𝐨𝐚𝐝𝐬 & now hit 20,000 𝐦𝐨𝐧𝐭𝐡𝐥𝐲 𝐝𝐨𝐰𝐧𝐥𝐨𝐚𝐝𝐬.

🤔𝐓𝐡𝐢𝐧𝐤𝐢𝐧𝐠 𝐨𝐟 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐈𝐧-𝐇𝐨𝐮𝐬𝐞?

Developing a custom solution can take 𝐦𝐨𝐧𝐭𝐡𝐬, 𝐏𝐲𝐌𝐂-𝐌𝐚𝐫𝐤𝐞𝐭𝐢𝐧𝐠 gets you there 𝐟𝐚𝐬𝐭𝐞𝐫

📅 𝐒𝐜𝐡𝐞𝐝𝐮𝐥𝐞 𝐚 𝐅𝐫𝐞𝐞 𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐲 𝐂𝐨𝐧𝐬𝐮𝐥𝐭𝐚𝐭𝐢𝐨𝐧 : calendly.com/niall-oulton

I don't think it's unfair to mirror my experience. I'm doing so because I've been hopeful that this could be like Twitter in the early days, but that's not the direction I'm seeing this going. That's fine, maybe people want this to be something else. But I see a risk of it becoming an echo chamber.

The AI MMM Agent: An AI-Powered Shortcut to Bayesian Marketing Mix Insights - PyMC Labs

Meet our AI-driven MMM Agent, built on PyMC-Marketing, to reduce months-long Marketing Mix Modeling work into hours ⏳eliminating messy data prep, guesswork on adstock decays, and stale insights that arrive too late.

👉 Read more: www.pymc-labs.com/blog-posts/t...

📩 Contact: info@pymc-labs.com

#ai