playgroup

AI learning community, UK

We're a diverse group of thinkers — developers, researchers, strategists — who've been using LLMs for years. We push these tools to the edge and watch them break, then figure out what they're actually good at.

We meet in person for full-day events, typically at zero cost, to work together on a well-specified problem. Between events, we talk on Slack.

No marketing blather. No hype. Just practitioners figuring out what's really going on.

Playgroup working session Discussing code at playgroup Playgroup at the pub afterwards

You're an excellent and motivated explorer — not necessarily a coder — who wants to understand both how and why these machines work. You're not afraid to ignore the hype and figure out what's really going on.

Contact Ian on LinkedIn and say you're interested in joining.

Join via LinkedIn

Coming up

Run your own event

We discuss future topics on Slack. If you want to run a private event or sponsor us to work on your problems — a little like the FullFact day from March 2026 — contact Ian.

Past events

March 2026
FullFact

We built tools that sped up fact-checking workflows and enabled new investigations, helping FullFact correct more mis-truths in our post-truth society.

Read write-up →
February 2026
PDF Fact Extraction and TheAgentCompany Agentic Benchmark

Visual LLMs have reasonably solved heterogeneous financial PDF extraction, but frontier models did very poorly on scripted white-collar office tasks.

Read write-up →
February 2026
Training and Fine-Tuning nanoGPT

We spent the day learning how parts of a Transformer work, training our own GPT and fine-tuning GPT-2 on rented H100s.

Read write-up →
November 2025
Can We Make ChatGPT Funny?

With two comedians and a benchmark dataset we tried to make ChatGPT produce original jokes to automate comedic material generation.

Read write-up →
August 2025
ARC AGI 2025

Building on earlier work, we explored ways to make LLMs solve hard logical tasks they'd never seen before in this updated $1M challenge.

Read write-up →
June 2025
ARC AGI 2024

Drawing on a year of personal research, we tried to teach LLMs to solve hard logical games they'd never seen before in this $1M challenge.

Read write-up →
Playgroup group watching a demo Playgroup presentation and laptops ChatGPT Offline Version Magic 8 Ball

About the organiser

Ian Ozsvald

Internationally-renowned data scientist and community builder. Co-founder of the 15,000-member PyData London meetup and co-author of O'Reilly's High Performance Python (3rd edition).

Over 25 years of experience in AI, including with Man Group, Groupe Mutuel, Hotels.com, Guidewire, Accel Partners, ITV, Channel 4, and the NHS.

Ian also leads the private RebelAI Data Science leadership group.