Investigating with AI

Course material for the 2026 edition. Each session links to its Google Slides deck.

Syllabus on SciencesPo

Day 1: 0908 Welcome, intro, course scope, logistics;

• Intro, presentations, logistics, whatsapp channel

• brainstorming session: https://ai-opinion-retro.onrender.com/

• LLMs: we break down the meaning of Large Language Models as Models + Language Models + Large.

• LLMs: 3M models on Huggingface, a closer look at Kimi3 and Poolside:Laguna

• 1. Intro and Welcome

• 2. LLMs

Day 2: 0915 - Data, Working with AI

• News: AI, a threat to humanity
• class representative election
• Tech: Data; scraping; enriching with LLM ; JSON; prompting and context window;
• Society: Cognitive biases when working with AI; centaurs and cyborgs; jagged frontier

• Slides
• AI brainstorm
• Group projects form

Day 3: 0922 The web, APIs; AI and Jobs

• Group projects : pitch

• News : new platforms and models : Jev, Muse, grok bot

• The OpenAI - Huggingface incident

• Tech: the web as API; API keys; API caps; some APIs (wikipedia, NYT, duckduckgo)

• Hands on : Wikipedia API, build a simple dataset

• Hands on : NYT API, get a developper key, build a simple dataset

• Society: AI and Jobs; WEF 2025 report on jobs

• slides

Reading :

• WEF 2025 future of jobs
• Dario Machine of Loving grace

Day 4: 0929 Tokens, embeddings and parameters; Regulating AI

• News: Jev

• Tech: Tokens, embeddings and LLM parameters

• slides on embeddings and tokens

• Hands on : Claude, Codex, Antigravity, Kimi Code, CLI, OpenCode

• slides hands-on

If we have time:

• Society: AI-act and regulations; sovereignty;
• slides on regulation
• Actors: OpenAI and Anthropic, Sam Altman, Dario Almodei

Day 5: 1006 - Benchmarks and reasoning emergence; AI on Science

• News

• Tech: LLM Benchmarks; simple predictors to reasoning models

• Society: AI and Science

• Demo: Building apps with AI, and Lovable

• Opencode

• Actors: Chinese labs and models

• Group projects

• Next : prepare the workshop

• slides

• reasoning in LLMs

Day 6: 1013 - workshop

This is a special session where we work on the EU parliament debate

We will

• constitute the dataset by scraping / downloading the verbatim of the debates for the past 12 months
• define study questions
• enrich and expand the dataset
• design artefacts : applications, web sites,

Day 7: 1020 - LLM to agents, harnesses + Education

• News
• Tech: LLM to agents, harnesses
• Society: Education & loss of skills
• Actors: Nvidia
• Demo: tbd
• Actors:
• Group projects

Readings

• Unesco

Day 8: 1103 -

• News
• Tech: Guardrails and abliteration
• Society: Environment
• Demo: tbd
• Actors:
• Group projects

Day 9: 1110 -

• News
• Tech: Voice AI & Personas
• Society: Democracy and disinformation
• Demo: tbd
• Actors:
• Group projects

Day 10: 1117 -

• News
• Tech: Image and Video
• Society:
• Demo: tbd
• Actors:
• Group projects

Day 11: 1124 -

• News
• Tech: Multi agent collaboration; Monitoring
• Society:
• Demo: tbd
• Actors: Langfuse; Langsmith
• Group projects

Day 12: 1201 - Project presentations