Mistral releases Mistral 7B
On September 27, 2023, France's Mistral AI released Mistral 7B under Apache 2.0, a 7.3B model that beat Llama 2 13B on every benchmark.
GPT-4V gives ChatGPT eyes
OpenAI's October 2023 GPT-4V system card detailed adding image input to GPT-4, bringing vision to a mainstream chatbot.
The G7 Hiroshima AI Process Code of Conduct (2023)
On 30 October 2023 the G7 published the Hiroshima Process International Code of Conduct, voluntary guidelines for organizations developing advanced AI systems.
US Executive Order 14110 on Safe, Secure, and Trustworthy AI (2023)
President Biden signs Executive Order 14110, the most sweeping US federal action on AI of the period, which was later revoked in January 2025.
The Bletchley Declaration
At the first global AI Safety Summit in November 2023, 28 countries and the European Union signed the Bletchley Declaration on managing frontier AI risk.
Machine learning completes the last Beatles song
The Beatles released Now and Then in 2023 after AI audio tools isolated John Lennon's vocal from a hissy 1970s home demo.
The UK AI Safety Institute (2023)
Established around the November 2023 Bletchley summit, the UK AI Safety Institute was the first state-backed body for testing and evaluating advanced AI.
Meta requires AI disclosure in political ads (2023)
Meta required advertisers to disclose when social-issue, election, or political ads use AI to create or alter content.
GraphCast: machine-learning weather forecasting beats the leading system
DeepMind's GraphCast (2023), a machine-learning weather model, beat the leading conventional forecasting system on most metrics while running in under a minute.
Student of Games unifies perfect and imperfect games
Student of Games was the first single algorithm to play both perfect-information games like Go and imperfect-information games like poker well.
A-Lab: an autonomous laboratory for materials synthesis
Berkeley's A-Lab ran for 17 days and autonomously made dozens of new inorganic compounds, guided by AI and robotics.
GNoME predicts 2.2 million new crystals for materials discovery
DeepMind's GNoME deep-learning tool predicted 2.2 million new crystal structures, including about 380,000 stable candidates worth synthesizing in the lab.
Google Cloud TPU v5p
Google's TPU v5p doubled FLOPS over v4 and scaled to 8,960 chips per pod, its most powerful AI accelerator at launch.
Google launches Gemini
Google launches Gemini, a natively multimodal model family built jointly by Google DeepMind, in three sizes - Ultra, Pro, and Nano.
Mistral releases Mixtral 8x7B, an open mixture-of-experts model
On December 11, 2023, Mistral released Mixtral 8x7B, an open-weight sparse mixture-of-experts model that matched or beat Llama 2 70B and GPT-3.5.
OpenAI and Axel Springer sign first major news-publisher deal
OpenAI's deal with Axel Springer let ChatGPT summarize and cite Politico, Business Insider, Bild and Welt, launching the publisher-licensing wave.
FunSearch makes new mathematical discoveries with an LLM
DeepMind's FunSearch paired an LLM with an evaluator to find genuinely new results on the cap set problem and bin packing.
ISO/IEC 42001, the First AI Management System Standard (2023)
ISO and IEC publish 42001, the world's first certifiable management-system standard for governing how organizations build and use AI.
FTC bans Rite Aid from facial recognition for five years
In December 2023 the FTC barred Rite Aid from using AI facial recognition for five years after false matches led to wrongful accusations.
Coscientist: autonomous chemical research with LLMs
A GPT-4-driven agent that designed, planned, and ran chemistry experiments on real robotic hardware, reported in Nature in 2023.
Suno turns text prompts into full songs
Suno, a Cambridge AI music startup, came out of stealth offering to generate complete original songs with vocals from a short text prompt.
The New York Times Sues OpenAI and Microsoft
On December 27, 2023, The New York Times sued OpenAI and Microsoft for copyright over its articles used in AI training, putting training data before the courts.
CrewAI launches role-based multi-agent orchestration framework
CrewAI released an open-source Python framework for composing role-playing AI agents into coordinated crews, enabling software development pipelines where specialised agents - lead developer, code reviewer, test engineer - collaborate through explicit handoffs.
Mobile ALOHA learns household tasks from cheap teleoperation
Stanford's Mobile ALOHA learned to cook and clean from just 50 demonstrations each on a low-cost whole-body teleoperation rig.
IMF: AI to Affect 40 Percent of Jobs
An IMF staff note estimated AI would affect about 40 percent of jobs worldwide, rising to 60 percent in advanced economies.
OpenAI's 2024 worldwide-elections policy
OpenAI published election rules in 2024: no impersonation or campaigning bots, image provenance, and pointing voters to official info.
AlphaGeometry solves olympiad geometry near gold-medal level
DeepMind's AlphaGeometry solved 25 of 30 olympiad geometry problems, near the gold-medalist average; successors later reached silver-medal level at a full IMO.
Neuralink implants its brain-computer interface in a human
In January 2024 Neuralink implanted its wireless brain-computer interface in a human for the first time, later letting a paralyzed man move a cursor by thought.
Limbic's AI self-referral chatbot widens NHS mental-health access
A 2024 Nature Medicine study of 129,400 patients found Limbic's self-referral chatbot raised NHS therapy referrals, most for minority groups.
Vesuvius Challenge Grand Prize awarded
A student team used machine learning to read text inside an unopened Herculaneum scroll, winning the $700,000 Vesuvius Challenge Grand Prize.
FCC makes AI-generated voices in robocalls illegal (2024)
The FCC ruled that AI-cloned voices in robocalls are 'artificial' under the TCPA, making them illegal without prior consent.
USPTO inventorship guidance for AI-assisted inventions
The US Patent Office said AI-assisted inventions are patentable if a human made a significant contribution, even though AI cannot be a named inventor.
Microsoft and OpenAI report state-affiliated threat actors using LLMs (2024)
Microsoft and OpenAI disclosed and disabled accounts of Russian, North Korean, Iranian, and Chinese state-linked actors using LLMs.
Google launches Gemini 1.5 with a million-token context window
Google announced Gemini 1.5 Pro in February 2024 with a context window of up to one million tokens, tested as far as ten million in research.
OpenAI introduces Sora text-to-video
OpenAI unveiled Sora in February 2024, a model that generates video from text prompts, then released it to the public roughly ten months later.
AI Elections Accord signed at Munich (2024)
Twenty tech companies signed a voluntary accord to fight deceptive AI content in the 2024 elections.
Reddit licenses its data for AI training and goes public
Reddit's 2024 IPO filing disclosed data-licensing deals to train AI, including a reported $60M-a-year Google agreement, formalizing user content as an AI asset.
Figure raises $675M and signs an OpenAI collaboration
In 2024 humanoid startup Figure raised $675M at a $2.6B valuation and partnered with OpenAI to build AI models for humanoid robots.
Anthropic releases the Claude 3 model family
Anthropic launches Claude 3 in three models - Opus, Sonnet, and Haiku - which it says set new industry benchmarks across cognitive tasks and add vision.
MethaneSAT launches to map methane leaks from space
In March 2024 an environmental nonprofit launched MethaneSAT, using machine learning to map oil and gas methane leaks worldwide.
Cerebras WSE-3: The Largest AI Chip
Cerebras's third wafer-scale engine packed 4 trillion transistors and 900,000 cores onto a single dinner-plate-sized chip.
OpenAI partners with Le Monde and Prisa Media
OpenAI's deals with France's Le Monde and Spain's Prisa Media extended its publisher-licensing program into French- and Spanish-language news.
xAI open-sources the Grok-1 weights
xAI released the weights of Grok-1, a 314-billion-parameter mixture-of-experts model, under the Apache 2.0 license in March 2024.
Blackwell and the AI chip race
Nvidia unveils its Blackwell GPU platform at GTC 2024 as H100 demand explodes, turning AI compute into a capital-intensive race and lifting Nvidia to trillions.
YouTube requires disclosure of altered or synthetic content (2024)
YouTube began requiring creators to label realistic videos made with AI or other altered or synthetic media.
Google's AI predicts extreme floods in ungauged rivers
Google's 2024 Nature paper showed an AI flood model giving five-day warnings in ungauged basins, powering free forecasts in over 80 countries.
Tennessee's ELVIS Act protects voice against AI cloning
In March 2024 Tennessee enacted the ELVIS Act, the first US law to make voice a protected property right against AI voice cloning.
OMB Issues Federal AI Governance Guidance M-24-10 (2024)
The US Office of Management and Budget directs federal agencies to appoint Chief AI Officers and apply risk safeguards to rights- and safety-impacting AI.