Topics

Breakdowns of work on autonomous agents: planning, tool calls, multi-agent systems and why they fall apart on long tasks.

Models that write, read and fix code — from autocomplete to working through a task in a repository on their own.

How models arrive at an answer: chains of thought, planning, self-checking and what long deliberation costs.

What a model remembers between requests and within one: long-term memory, long context, forgetting.

Models that see: images, video, generation and visual understanding.

How model quality is measured and why benchmarks keep lying.

Hallucinations, alignment, robustness to attack, and what happens when models are trusted further than they should be.

RLHF, GRPO, DPO and other ways to teach a model what the training data does not contain.

Models driving physical devices: from manipulators to the gap between seeing and doing.

Text, sound, image and video in a single model.

Retrieval-augmented generation and deep research: getting the right knowledge in front of the model at the right moment.

Deployment, process change and economics: what happens to a company when AI stops being an experiment.

Data engineering in the model era: preparing data, analytics without SQL, and why text-to-SQL keeps breaking.

Consciousness, AGI, trust, and what models do to the people who use them.

Internal representations of reality: world models, simulation, and the move from prediction to understanding.

Models beyond the screen: wearables, the internet of things and new ways to talk to a machine.

Getting more out of a model: prompting, picking the right mode, and the mistakes people make using it.