November

10 events

Editors' summary

A coup erupts at a booming OpenAI

On 1 and 2 November, the UK hosted an AI Safety Summit. Twenty-eight countries including the US and China signed the Bletchley Declaration.

On the 6th, at its first DevDay, OpenAI announced GPT-4 Turbo and GPTs. Developers cheered.

On the 15th, Cambridge Dictionary named "hallucinate" its word of the year — the sense, an AI producing plausible falsehoods, had been added to the dictionary only that year.

Then on the 17th, OpenAI's board abruptly fired CEO Sam Altman. Microsoft, its largest outside investor, offered to take him in. Around 700 employees signed a letter demanding the board resign, and on the 22nd Altman returned as CEO. Two interim CEOs came and went in five days, and the board was replaced.

This block is written by the editors. It is kept separate from the sourced record below.

Record10 events

1
PolicyUnited Kingdom / European Union
First international AI Safety Summit issues the Bletchley Declaration
The United Kingdom hosted the first international AI Safety Summit over two days at Bletchley Park, and the Bletchley Declaration was agreed on the opening day. Signed by 28 countries — including the United States, China and Japan — and the European Union, it committed signatories to building a shared scientific understanding of the substantial risks posed by frontier AI and to risk-based policies in their own countries. The declaration states that organisations developing frontier AI carry a "particularly strong responsibility" for the safety of those systems, including through safety testing and evaluations. Executives from the major AI companies were among the attendees, and follow-up summits were set for South Korea and France.
6
ModelOpenAI / Sam Altman
First DevDay unveils GPT-4 Turbo and GPTs
OpenAI held its first developer conference, DevDay, announcing GPT-4 Turbo with a 128K context window and sharply lower prices, the Assistants API for building agent-like apps, and APIs for DALL·E 3 and text-to-speech. It also introduced GPTs, custom versions of ChatGPT anyone can create for specific purposes, along with plans for a GPT Store, underscoring OpenAI's evolution into a platform company.
15
CultureCambridge Dictionary
"Hallucinate" is named word of the year
Cambridge Dictionary named "hallucinate" its word of the year for 2023, in the sense of an AI producing plausible but false information — a definition added to the dictionary only that year. As ChatGPT spread, the word became, within a single year, the everyday term for talking about AI's limits. The phenomenon itself made repeated headlines, including a US lawyer who cited fictitious cases invented by ChatGPT in a court filing. Dictionary.com chose the same word as its word of the year. Japanese adopted the term as-is, transliterated as "ハルシネーション".
15
ProductMicrosoft
Microsoft unveils its own Maia 100 and Cobalt 100 chips
At Ignite, Microsoft announced two chips designed in its own silicon lab: the Azure Maia 100 AI Accelerator and the Azure Cobalt 100 CPU. Maia 100 was an AI accelerator for running cloud training and inference for workloads such as OpenAI's models, Bing, GitHub Copilot, and ChatGPT, with 105 billion transistors — among the largest chips on a 5nm process. Microsoft said it had shared the Maia 100 design with OpenAI to confirm it was optimized for large language model work. Cobalt 100 was a 128-core, 64-bit Arm CPU that Microsoft said delivered 40% better performance than existing Azure Arm chips. Both were fabricated at TSMC and were to reach Microsoft's data centers in the first part of the following year. With this, Microsoft joined Google's TPUs and Amazon's Trainium in building its own silicon to reduce dependence on NVIDIA.
17
GovernanceDisputedOpenAI / Sam Altman / Greg Brockman / Mira Murati
OpenAI board fires CEO Sam Altman
The board of OpenAI's nonprofit abruptly announced that Sam Altman would depart as CEO and leave the board, stating he "was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities" while offering no specifics, fueling rampant speculation. CTO Mira Murati became interim CEO, and chairman Greg Brockman, removed as board chair, announced he was quitting the same day. It was the start of a crisis that shook the company at the center of the generative AI boom.
20
GovernanceOpenAI / Microsoft / Emmett Shear / Sam Altman
Emmett Shear named interim CEO as some 700 employees demand board resign
After talks to reinstate Altman collapsed, OpenAI's board named former Twitch CEO Emmett Shear as the new interim CEO. Hours later, Microsoft CEO Satya Nadella announced that Altman and Greg Brockman, together with colleagues, would join Microsoft to lead a new advanced AI research team. In response, more than 700 of OpenAI's roughly 770 employees signed an open letter demanding the board resign and reinstate Altman, threatening to move to Microsoft otherwise. Signatories included Ilya Sutskever, who had taken part in the ouster and said he regretted his involvement.
21
ModelAnthropic
Claude 2.1 arrives with 200K tokens and tool use
Anthropic released Claude 2.1, doubling the context window to 200K tokens — about 150,000 words, or 500-plus pages — and reporting a 2x reduction in false statements versus Claude 2.0, a 30% drop in incorrect answers on complex documents, and a three-to-fourfold reduction in misreading a document's claims. It also introduced tool use in beta, letting Claude call external APIs, searches, database queries, and calculators, plus support for system prompts to shape responses. The full 200K context window was reserved for Claude Pro subscribers. The release landed in the middle of OpenAI's board crisis.
22
GovernanceOpenAI / Sam Altman / Bret Taylor / Larry Summers
Sam Altman returns as CEO with reconstituted board
Five days after the firing, OpenAI announced an agreement in principle for Altman to return as CEO with a new initial board of Bret Taylor (Chair), Larry Summers, and Adam D'Angelo; Ilya Sutskever, Helen Toner, and Tasha McCauley, who had taken part in the ouster, left the board. On November 29, OpenAI formally announced Altman's return as CEO, Mira Murati's return as CTO, and Greg Brockman's return as President, with Microsoft taking a non-voting observer seat on the board. The extraordinary firing-and-return saga highlighted the challenges of AI governance worldwide.
22
ResearchReportedOpenAI / Sam Altman
Reuters reports "Q*" breakthrough linked to Altman ouster
Reuters reported, citing two people familiar with the matter, that ahead of Altman's ouster several OpenAI researchers had sent the board a letter warning of a powerful AI discovery they said could threaten humanity. An internal project called Q* (Q-star) had reportedly become able to solve grade-school math problems, which some inside saw as a breakthrough toward AGI. Never verified, the report fueled worldwide speculation about progress toward AGI, and Q* became an enduring symbol of the saga.
28
ProductPika / Lightspeed Venture Partners
Pika ships Pika 1.0, taking video generation out of Discord
Pika released Pika 1.0 and disclosed $55 million in funding alongside it. What had been a Discord bot was rebuilt as a web service: as well as producing video in styles from 3D animation to anime to live action, it could extend a clip, widen the aspect ratio, and swap out what was in the frame. The money was largely a Series A led by Lightspeed Venture Partners, with individuals including Quora's founder and GitHub's former CEO taking part. The company had been started that year by two people who had been PhD students at Stanford. At the announcement it put its community at 500,000 people generating millions of videos a week. Video generation at the time was still largely something you tried through a bot on Discord; products that stood on their own on the web were few.

↑ 2023