Insights and updates from our agency

DeepMind introduced AlphaCode, investigating code generation in competitive programming tasks.

OpenAI described InstructGPT, emphasising human feedback and instruction following rather than scale alone.

DeepMind published research on Gopher and retrieval-enhanced language modelling, examining scale, capabilities and access to external information.

UNESCO adopted its Recommendation on the Ethics of Artificial Intelligence, setting out principles for responsible development and use.

Microsoft and NVIDIA described Megatron-Turing NLG, highlighting the infrastructure behind very large language models.

Stanford's foundation-model research helped frame how general-purpose models could be adapted to many downstream tasks. The underlying report appeared in August.

OpenAI introduced Codex, a model capable of translating natural-language instructions into code.

DeepMind and EMBL-EBI launched the AlphaFold Protein Structure Database, widening access to predicted structures.