Timeline

2026

3 stories

OpenAI reports that its researchers are using coding agents to run more experiments and carry out longer research tasks under human direction.

OpenAI releases GPT-6 Astra and reports that it reaches the company’s Critical cybersecurity capability threshold.

Anthropic releases Claude Fable 5.1 and describes it as a safeguarded version of the underlying Mythos 5.1 model.

2025

8 stories

Google begins releasing Gemini 3 Pro in preview across its app, developer tools, and AI Mode in Search, alongside the Antigravity development platform.

OpenAI launches GPT-5 as a system that routes between faster answers and deeper reasoning depending on the task.

Anthropic introduces Claude Opus 4 and Sonnet 4, emphasizing coding, reasoning, and AI agents.

Stanford's 2025 AI Index reports $33.9 billion in global private investment in generative AI during 2024, up 18.7% from 2023.

OpenAI announces $40 billion in new funding at a $300 billion post-money valuation, in partnership with SoftBank Group.

OpenAI introduces deep research, which browses and combines information from multiple online sources into a report.

The DeepSeek-R1 paper describes training reasoning abilities with reinforcement learning and shares results and model details.

OpenAI announces Stargate, with an intention to invest $500 billion over four years in US AI infrastructure. Initial equity funders include SoftBank, OpenAI, Oracle, and MGX.

2024

6 stories

Anthropic introduces the Model Context Protocol: an open way for AI applications to connect to data sources and tools.

The Nobel Prize in Chemistry recognizes Demis Hassabis and John Jumper for protein-structure prediction, alongside David Baker for computational protein design.

OpenAI introduces o1-preview, trained to work through problems before producing an answer.

The European Union's AI Act enters into force, establishing a risk-based framework with obligations phased in over time.

NVIDIA introduces Blackwell at GTC, a platform designed for large-scale AI training and inference.

OpenAI previews Sora, a text-to-video model, with access initially limited to testers and selected creators.

2023

5 stories

Paris-based Mistral AI releases Mistral 7B, a small language model with downloadable weights under the Apache 2.0 license.

Amazon announces a strategic collaboration with Anthropic, with AWS becoming its primary cloud provider for key workloads.

Meta releases Llama 2 model weights under a license allowing many research and commercial uses.

OpenAI announces GPT-4, a model that accepts text and image inputs, with image access initially limited.

Microsoft and OpenAI announce a new multiyear, multibillion-dollar phase of their partnership.

2022

3 stories

ChatGPT puts a language model inside a simple chat box. You ask, it answers, and you can ask again.

Stability AI announces the public release of Stable Diffusion, with model weights people can download.

DeepMind's Chinchilla research finds that, for a given training budget, model size and training data need to be balanced.

2021

4 stories

GitHub launches a technical preview of Copilot, which suggests code inside a developer’s editor.

LoRA proposes adapting a language model by training a much smaller set of additional numbers while keeping the original weights fixed.

Anthropic announces $124 million in Series A funding. Led by Dario and Daniela Amodei, the team sets out to build more reliable and steerable AI.

DALL·E demonstrates image generation from text descriptions, including unusual combinations of familiar ideas.

2020

4 stories

AlphaFold’s results in the CASP14 assessment show a major advance in predicting protein structures.

A Berkeley research team presents a diffusion approach that learns to turn noisy images into clear ones.

GPT-3 shows that a large language model can attempt new tasks from examples in its prompt, without retraining for each one.

OpenAI researchers describe how language-model performance changes with model size, training data, and compute.

2019

2 stories

Microsoft announces a $1 billion investment in OpenAI and a partnership to build new Azure AI supercomputing technology.

OpenAI introduces GPT-2 but initially withholds the full model, citing concerns about misuse.

2018

2 stories

Google researchers train a model to use the words before and after a missing word to work out what fits.

Oak Ridge unveils Summit: a scientific supercomputer combining IBM processors with thousands of NVIDIA GPUs.

2017

1 story

“Attention Is All You Need” introduces the transformer, a design that learns relationships between parts of a sequence.

2016

1 story

DeepMind’s AlphaGo wins its five-game match against Lee Sedol, four games to one.

2015

5 stories

OpenAI introduces itself as a nonprofit research company, led by a team including Ilya Sutskever and Greg Brockman, with Sam Altman and Elon Musk as co-chairs.

ResNet introduces connections that let information skip over some layers in a neural network.

TensorFlow gives developers a shared set of tools for building and training neural networks.

In the replies, Musk points to V7.1 when asked about getting out and letting the car park itself in a garage.

Context & original posts

Researchers try the demo, make music with it, and debate what the generated text actually proves.

Context & original posts
NO BACKGROUND NEEDED

The basics.

A few ideas that make everything else click.