<aside> <img src="/" alt="/" width="40px" />

Practical activities to explore during or after the workshop.

Link to slides: https://unioxfordnexus.sharepoint.com/:p:/s/ITServices-AIMLCompetencyCentre/EfzvpDzEGbBLgFLg-7_RTm4BPy-b301gjuxmPJZ2z1ubBw?e=ACXOOK

Link to workshop details: AI for researchers - AI and ML Competency Centre

</aside>

Part 1: Exploring the cognitive baseline

Core activities

Activity 1: Learn LLM semantic ability and failure modes

Go to ai.dev; ask Gemini to write a short text, then start a new chat, paste it in and interrogate the text to see where it succeeds and fails.

Activity 2: Compare factual recall across Gemini variants

In ai.dev, ask Gemini 2.5 Pro, Flash, and Flash Lite the same factual questions and compare consistency and citations.

Activity 3: Explore model‑size effects with Gemma

In ai.dev, pick Gemma at multiple sizes and ask identical questions to observe capability vs size trade‑offs.

Activity 4: Test reasoning toggles and budget sensitivity

Using Gemini 2.5 Flash in ai.dev, run tasks with reasoning ON/OFF and vary the reasoning budget; note outcome differences.

Activity 5: Use Code execution for arithmetic/counting

In ai.dev, turn on Code execution and ask quantitative tasks (e.g., 12434 * 312; word counts) to validate tool use.

Activity 6: Run simple data analysis with synthetic data

In ai.dev, enable Code execution, ask Gemini to generate sample data, save it, and perform a basic analysis.

Further exploration

Activity 7: Cross‑model and orchestration comparison on identical tasks

Repeat Activities 1–6 in ChatGPT, Claude, and Qwen; record differences in accuracy, speed, and explainability.

Activity 8: Multimodal conversation about a video/screenshare

In ai.dev, hold a conversation with Gemini about an uploaded video or screenshare; probe for semantic understanding.