RedPajama Reproduces the LLaMA Training Data
An open collaboration rebuilt LLaMA's 1.2-trillion-token training set from public sources so anyone could train comparable models without Meta's private data.
Stability AI releases StableLM
On April 19, 2023, Stability AI released StableLM, open 3B and 7B language models under a permissive license, trained on 1.5 trillion tokens.
Harvey raises a $21M Series A for legal AI
Legal-AI startup Harvey raised a 21 million dollar Series A led by Sequoia in April 2023, backed by the OpenAI Startup Fund, to build GPT-4 tools for lawyers.
Microsoft Designer opens its DALL-E-powered design app
Microsoft expanded its DALL-E-based Designer app in April 2023, dropping the waitlist for a tool that generates social graphics from a text prompt.
Inflection launches Pi, a chatbot built to be kind
In May 2023 Inflection AI released Pi, a 'personal AI' tuned for warm, supportive conversation rather than task completion, before the team moved to Microsoft.
MosaicML releases MPT-7B
On May 5, 2023, MosaicML released MPT-7B, an open commercially licensed model trained on 1 trillion tokens for about 200,000 dollars.
Google introduces Duet AI for Google Workspace
At Google I/O 2023 Google announced Duet AI, embedding generative help across Gmail, Docs, Slides, Sheets, and Meet.
Power and Progress (Acemoglu and Johnson)
A book arguing that technology's benefits are not automatic but depend on who controls its direction and who shares the gains.
Massively Multilingual Speech, recognition for 1,100+ languages
Meta's 2023 MMS project brought speech recognition and synthesis to over 1,100 languages and language ID to nearly 4,000.
Adobe brings Generative Fill to Photoshop
Adobe added Firefly-powered Generative Fill to Photoshop beta in May 2023, letting users add, remove, or extend image content from a text prompt.
NVIDIA announces ACE for Games for AI NPCs
NVIDIA announces ACE for Games, a generative-AI toolkit giving non-player characters real-time spoken conversation.
The Statement on AI Risk
A one-sentence May 2023 Center for AI Safety statement put extinction risk from AI alongside pandemics and nuclear war, signed by top researchers and lab CEOs.
AlphaDev finds faster sorting algorithms
DeepMind's AlphaDev framed code optimization as a single-player game and found sorting routines now used trillions of times a day.
OpenAI Adds Function Calling to Its API
OpenAI's June 2023 API update let GPT models output structured JSON to call developer-defined functions on demand.
The Grammys rule that only human creators can win
The Recording Academy ruled that only human creators are eligible for Grammy Awards, while music using AI elements can still qualify.
IBM brings generative AI commentary to Wimbledon
In 2023 IBM and Wimbledon launched watsonx-generated audio commentary for tennis highlight clips.
Databricks acquires MosaicML for $1.3 billion
In June 2023 Databricks agreed to buy generative-AI startup MosaicML for about $1.3 billion to let enterprises build and own their own large language models.
Insilico's INS018_055, a generative-AI drug, enters Phase II trials
In 2023 Insilico Medicine dosed Phase II patients with INS018_055, which it calls the first drug with both an AI-discovered target and AI-designed molecule.
MetaGPT introduces multi-agent software company simulation
MetaGPT published a framework simulating a full software company with role-playing LLM agents - product manager, architect, engineer - that collaborate to produce user stories, architecture docs, and working code from a single natural-language requirement.
Ollama makes running local LLMs a one-line command
Ollama, an open-source tool first released in 2023, wrapped local LLM inference behind a simple command and API, popularizing private on-device AI.
New York City requires bias audits of AI hiring tools
From July 2023 New York City's Local Law 144 required employers to bias-audit AI hiring tools and notify candidates before using them.
Pangu-Weather beats numerical forecasting with a 3D neural network
Huawei's Pangu-Weather, in Nature in 2023, was the first AI model to beat leading numerical weather prediction while running thousands of times faster.
Med-PaLM and large language models encode clinical knowledge
Google's Med-PaLM, published in Nature in 2023, scored 67.6% on USMLE-style questions and was the first LLM tuned to answer medical questions at scale.
OpenAI licenses the Associated Press news archive
OpenAI agreed to license the AP's text archive back to 1985 in one of the first formal news-content deals between a major publisher and an AI firm.
Meta releases Llama 2, free for commercial use
Meta released Llama 2 in July 2023, an open large language model that, unlike Llama 1, was licensed for commercial use, with Microsoft as launch partner.
Worldcoin launches the Orb and iris-based proof of personhood
In 2023 Sam Altman's Worldcoin began scanning people's irises with the Orb to issue a World ID proving they are a unique human.
Meta and academics publish the 2020 US election feed experiments
In 2023 Meta and outside academics published experiments that altered real feeds and found little effect on polarization.
MASAI trial tests AI mammography in a randomized study
Sweden's MASAI trial randomized 80,033 women and found AI-supported mammography safe while cutting reading workload 44 percent.
Alibaba open-sources the Qwen model family
On August 3, 2023, Alibaba Cloud open-sourced Qwen-7B, the start of a model family that became one of the most-used open-weight series worldwide.
NASA and IBM release the Prithvi geospatial foundation model
In 2023 NASA and IBM open-sourced Prithvi, a foundation model for satellite imagery built on Landsat and Sentinel-2 data.
OpenAI's GPTBot and the robots.txt fight over training data
OpenAI's 2023 GPTBot crawler let websites block AI training scraping via robots.txt, triggering a wave of publishers disallowing AI bots.
China's Interim Measures for Generative AI Services (2023)
Effective 15 August 2023, China's Interim Measures for Generative AI Services became the first national law specifically governing generative AI.
Thomson Reuters buys Casetext and its CoCounsel AI assistant
Thomson Reuters completed a $650 million cash acquisition of Casetext, maker of the GPT-4-powered legal assistant CoCounsel.
ElevenLabs brings high-fidelity AI voice cloning to the public
In 2023 ElevenLabs left beta with a multilingual speech model and voice cloning convincing enough to spark a wave of audio deepfake concern.
SeamlessM4T, a single model for speech and text translation
Meta's SeamlessM4T, released in 2023, is one model that translates speech and text across nearly 100 languages in multiple directions.
A brain implant decodes speech and drives a talking avatar
In 2023 UCSF researchers decoded a paralyzed woman's attempted speech from brain signals into text, audio, and an animated avatar at 78 words per minute.
The Pentagon launches the Replicator initiative
In August 2023 the US announced Replicator, an effort to field thousands of cheap attritable autonomous systems within 18 to 24 months.
Harvard's CS50 adds an AI rubber-duck tutor
In 2023 Harvard's CS50 deployed cs50.ai, an AI rubber-duck debugger that guides students toward answers while banning outside chatbots.
The Copyright Office refuses to register an AI-made prize-winning artwork
The US Copyright Office Review Board refused to register Theatre D'Opera Spatial, the Midjourney image that won the 2022 Colorado State Fair.
Google requires AI disclosure in election ads (2023)
Google began requiring verified election advertisers to disclose synthetic or digitally altered content in their ads.
Amazon requires authors to disclose AI-generated books
Amazon's Kindle Direct Publishing began requiring authors to disclose AI-generated content and capped uploads at three new titles a day.
Stability AI launches Stable Audio for text-to-music
Stability AI released Stable Audio, a latent diffusion model that generates 44.1kHz music and sound from text prompts.
AlphaMissense classifies 71 million genetic variants
DeepMind's AlphaMissense, adapted from AlphaFold, predicted the effect of 71 million possible human missense mutations.
Anthropic's Responsible Scaling Policy
Anthropic's 2023 policy tied model deployment to AI Safety Levels, committing to pause scaling until safeguards catch up to capabilities.
Authors Guild v. OpenAI: novelists sue over training data
The Authors Guild and 17 novelists filed a class action accusing OpenAI of copying their books to train GPT without permission.
CFPB says AI lenders must give specific reasons for denials
In 2023 the CFPB ruled that lenders using complex AI models must still give borrowers specific, accurate reasons for credit denials.
Amazon invests up to $8 billion in Anthropic
Amazon agreed to invest up to $4 billion in Anthropic in 2023, then doubled the commitment to $8 billion in 2024, with AWS as cloud and training partner.
Hollywood strikes write AI into the contract
The 2023 WGA and SAG-AFTRA strikes ended with the first major union contracts to set explicit terms on generative AI, making AI protections negotiated terms.