Skip to content
AI Daily · Asia/Shanghai · 2026-11-20 · Friday· Issue 58· About 12 minutes to read

GPT-Live-1.1 Now Powers Voice in Developers' Apps

OpenAI says GPT-Live-1.1, which has powered voice in ChatGPT Plus and Pro since October 15th, can now power developers' apps. OpenAI says developers can use it to build voice agents that understand details such as letters and numbers, follow instructions, and delegate complex workflows to a backend model of their choice.
Time machineChronicle

OpenAI Developers · XRead the storyOriginal

01

Model releases and updates

Latent Space

Google DeepMind Launches Gemini 4 Neon, Its Fourth Gemini 4 Model

Google DeepMind released Gemini 4 Neon, the fourth Gemini 4 model in about seven weeks. Neon is positioned as a midtier model and is faster and cheaper than Carbon, though it outperformed Carbon on some measures despite being smaller. In independent results, it matched Carbon’s Intelligence Index score at high effort and placed second on the Vals Index, while scoring below Carbon and Hydrogen on Code Migration.

02

Product releases and updates

Josh Woodward · XReported by 3 other sources

Gemini for Mac Can Now Perform Tasks on Your Computer

Josh Woodward says the Gemini app on Mac can now do tasks on a user's computer, calling it a top feature request. The quoted GeminiApp post says users can hand off work across folders, apps, and the web while keeping their screen and cursor. It shows Gemini clicking, scrolling, and typing in a live background window to turn a sketch into a rigged 3D model in Blender.

AWS Machine Learning BlogFirst-party

Amazon Polly Makes Voice Cloning Generally Available as a Gated Release

Amazon Polly voice cloning is now generally available as a gated release, letting customers create a synthetic voice from recordings of a consenting speaker and use it with the generative engine. Voice creation typically takes 5–10 minutes, and cloned voices support seven languages while speaking only the language recorded. AWS deletes its copy of the recordings after creation; customers control access and can delete the voice.

AWS Machine Learning BlogFirst-party

Amazon Bedrock AgentCore Harness Adds Strands Defaults and Lifecycle Hooks

AWS is upgrading Amazon Bedrock AgentCore harness with Strands harness defaults, built-in tools, and lifecycle hooks; the rollout began on November 11 for new and existing harnesses. Managed defaults include planning, task delegation, and context management, while hooks let teams run synchronous Lambda checks before tool calls and receive notifications through SNS or EventBridge. Existing configurations remain in place, and no customer action is required.

Google Blog: AIFirst-party

Google Launches Tool Retrieval in the Gemini Interactions API

Google has launched Tool Retrieval in preview in the Gemini Interactions API. Agents can keep deferred tools' names and descriptions available while loading their parameter schemas on demand; server-side retrieval completes search and function calling within one request, with no separate charge. The feature also supports remote MCP servers and works with Gemini 3.6 and later models, including all Gemini 4 models.

03

Industry

Ars Technica · AI

Teens Sue School and Police After Doritos Bag Triggers AI Gun Alert

Four Black students are suing Baltimore County Public Schools, Baltimore County police officers, and Omnilert after a gun detection alert triggered a police response in October 2025, even though school staff had cancelled it within about two minutes. An inspector general found that a principal later contacted a school resource officer without seeing the cancellation; the students seek an injunction barring Omnilert use until staff and police are adequately trained.

NVIDIA BlogFirst-party

NVIDIA and Microsoft Expand Accelerated Computing From Azure to Windows PCs

NVIDIA and Microsoft are expanding their collaboration across Azure, enterprise model customization, edge systems, and Windows PCs. Microsoft Foundry is adding customer access to NVIDIA open models and customization options, while Azure Local gains support for Nemotron 3.5 and NVIDIA GPUs. The companies also describe scientific AI integrations and Microsoft’s use of NVIDIA software and GPUs in its own models and services.

04

Research

Google ResearchFirst-party

Google Measures Quantum Contextuality on Processors up to 105 Qubits

Google Quantum AI reports measuring contextuality on its superconducting quantum processors in a paper published in Science. The processors won the magic square game 98.30% of the time and exceeded the classical bound on a contextuality test; the team also ran hidden-linear-function instances on grids of up to 105 qubits with success probabilities above one-half. The researchers say these tasks can reveal subtle coherent errors that random-circuit benchmarks may miss.

Aravind Srinivas · X

Perplexity Shares Research on Self-Improving Skills in Computer

Aravind Srinivas shared research on Perplexity Computer's Brain, which uses past session trajectories to propose personalized agent skills when it finds recurring patterns with enough evidence. In an eight-week randomized test on production traffic, 72.9% of users received at least one new personal skill, compared with 6.5% who added one manually. Tool call failures fell 11% and dissatisfied threads fell 7.6%, with no measurable change in cost or latency.

05

Tips and perspectives

(End of issue)

e/acc DailyAutomatically edited from public sources. Every story links to its original · Daily archive

Tomorrow at 08:00, a briefing like this arrives in your inbox.

The same kind of briefing as the one above: a few stories worth reading. Free. Mail starts after you confirm, and every letter has a one-click unsubscribe. Read a past issue · 2026-11-19