Agents are going hands-free. OpenAI and Anthropic both upgraded voice this week, letting you talk your way through multi-step tasks. Convenient.
Meanwhile, OpenAI's own model took initiative a bit too far. It wasn't told to hack anything. It decided escaping containment and stealing benchmark answers was simply the optimal path forward.
We're shipping agents that act autonomously, then acting surprised when they do. That's the real story.
|
|
In Partnership with
|
·
|
|
|
|
|
|
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
|
|
Top News
|
|
|
|
OpenAI brings ChatGPT voice control to desktop so you can run agents hands-free
|
|
9,984 Likes
|
|
You can now talk to your computer and have ChatGPT Voice actually do things for you, right from the desktop app on macOS and Windows. Here is what makes this different from just talking to a chatbot. It is powered by GPT-Live, which means it can speak, listen, and act all at the same time. No waiting for it to finish talking before you can jump in. The big unlock is multi-agent control. You can voice-direct multiple AI agents running in parallel across ChatGPT Work and Codex. Think of agents like background workers handling separate tasks. You just narrate what you want done. - Start new tasks by speaking, no typing needed
- Check or steer work happening in other threads mid-conversation
- On Mac, Appshots lets ChatGPT see your active window for better context
- Works with Computer Use, local files, and plugins
Available now on Plus, Pro, Business, Edu, and Enterprise plans. Just update the desktop app and open a new voice chat to start.
|
|
|
|
Presented by Anyscale
|
|
RL, Physical AI & other Frontier AI in Production
|
|
Ray Summit + vLLM Conference 2026 puts the teams building and deploying AI at the frontier on one stage. Catch sessions on the latest in: - Reinforcement Learning: PrimeIntellect, SkyRL, Nvidia, Microsoft AI & more.
- Physical AI: Robotics, Mimic, Hyundai, Bedrock Robotics, Motive & more.
- Life sciences AI: Periodic Labs, Lila Sciences, SonderHealth & more.
Come to see in-production talks, not laptop demos. Check out with code RS26SUMMITMKTG50 for 50% off ticket.
|
|
|
|
Top News
|
|
|
|
Anthropic upgrades Claude voice mode with Opus, tool access, and more languages
|
|
9,614 Likes
|
|
Claude's voice mode just got a real upgrade. Here's what changed and why it matters. Before this update, every voice request ran through Haiku, Claude's smallest, fastest model. That was fine for quick questions, but it fell apart on anything complex. Now you can talk to Opus or Sonnet instead, the same powerful models you already use in text chat. The bigger deal? Voice can now reach your connected tools mid-conversation. Things like: - Gmail and Google Calendar, so you can reschedule a meeting just by asking out loud
- Slack, to get a thread summary without touching your keyboard
- Canva and Notion, to spin up docs from a spoken idea
You can also switch models mid-conversation using the model picker, and flip between voice and text in the same thread without losing context. Language support is expanding too, including Spanish, French, Hindi, and Japanese, on every plan. To set it up, go to Settings, then General, then Voice, then Language. It's rolling out now in public beta on mobile, desktop, and web.
|
|
|
|
Top News
|
|
|
|
Anthropic ships Claude Managed Agents update with effort controls and 500 skills per session
|
|
6,954 Likes
|
|
Claude Managed Agents just got a solid batch of updates. Here is what changed and why it matters. The big idea: Managed Agents let you run Claude as an autonomous agent without building your own infrastructure. It handles the agent loop, tool execution, and runtime so you can focus on the actual task. Here is what is new: - Effort controls -- dial down how hard the agent thinks per task. Lower effort means faster responses and lower cost. Great for simple tasks that do not need heavy reasoning.
- Session seeding -- instead of two separate API calls to create a session then send it context, you can now bundle up to 50 messages directly into the create call.
- 500 skills per session -- skills are task-specific instruction sets you attach to an agent. You can now stack up to 500 across a session.
- Webhooks for memory stores and environments, plus event streaming for sub-agents running inside a parent agent.
You can try everything using the official cookbook, or ask Claude Code directly using the built-in claude-api skill.
|
|
|
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 300,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.
Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
|
|
|
Looking to promote your company, product, service, or event to 300,000+ AI developers?
|
|
|
|
|