Your AI Advantage Weekly

Voice AI Just Had a Big Week

PLUS: Privacy and safety changes at Anthropic and OpenAI worth knowing.

Riley

Riley

— 5 min read
Voice AI Just Had a Big Week
Just the Signals · Edition #13

It's Friday again!

I've been using voice dictation more and more lately.

Turns out we can speak much faster than we can type.

I measured it and I'm ~2x faster at talking (138 words per min vs 65 when typing).

My tool of choice has been Wispr Flow, but I've noticed speaking to ChatGPT has also gotten better.

This isn't the sad experience you've had with Siri or Alexa. AI has made voice capture more accurate and intuitive.

For me, talking has become an easier way to give AI feedback, casually explore an idea, or braindump thoughts into notes.

Based on this week's news, it seems like others are shifting towards voice too.

Here's the news you need to know this week:

  1. AI Settings Updates to Know About
  2. Spotlight on AI Voice Models
  3. Other Signals Worth Knowing

1/ AI Settings Updates to Know About

What happened: Anthropic and OpenAI made updates to their privacy and safety measures which could impact you.

Source: Anthropic

Anthropic Will Train Claude on User Convos

Anthropic will now train Claude models on your conversations, unless you opt out in your privacy settings by September 28. If you allow data sharing, Claude will retain your information for 5 years.

This applies to all consumer accounts, free and paid, with the exception of business and enterprise tiers.

The timing isn't a coincidence. Anthropic closed a $13B funding round this week, tripling its valuation to $183B. They'll need massive datasets to compete with OpenAI and Google.

Worth Noting: Gemini, ChatGPT, and Grok all default to training on user data for their consumer products unless users opt out. Claude was the lone holdout after Gemini announced a similar update this past month.
Source: OpenAI

OpenAI Adds Teen Safety Controls to ChatGPT

OpenAI is rolling out parental controls for teen users and will route sensitive conversations to more careful reasoning models.

Parents will be able to create linked accounts for kids 13+, set how ChatGPT responds to their child with behavior rules, manage memory and chat history settings, and get notifications if their teen is in distress.

Beyond controls for teens, ChatGPT will route all conversations it flags as potentially distressing to reasoning models that are designed to think strategically before responding.

What this means for you:

  • Check your privacy settings on your personal AI accounts. Make sure they align with your preferences.
  • Know the enterprise difference. Business accounts do not use chat data for training, but consumer accounts do by default.
  • Expect more safety controls. The industry is starting to take safety and harmful outputs more seriously.

2/ Spotlight on AI Voice Models

What happened: Voice AI took major leaps this week. Microsoft launched their first in-house speech models, OpenAI moved their voice API out of beta with image support, Google released a 70+ language conversation partner.

Source: Microsoft

Microsoft's New Voice Model & Independence Play

Microsoft launched MAI-Voice-1 and MAI-1-preview, their first speech models built entirely in-house.

MAI-Voice-1 generates natural speech in less than one second and is already integrated into Copilot Daily and Podcasts. MAI-1-preview is in testing on LM Arena (a model testing and ranking site).

The story here isnt just it's speech model. Instead, it's its independence from OpenAI. Microsoft spent years relying on OpenAI's tech in a messy partnership. Now they're building their own AI stack.

Source: OpenAI

OpenAI's Voice Upgrade

OpenAI moved its speech-to-speech model Realtime API out of beta with some big upgrades.

The new gpt-realtime model detects nonverbal cues like pauses and tone changes, switches languages mid-conversation, and handles image input during voice conversations.

Speech accuracy jumped from 65.6% to 82.8% on audio reasoning benchmarks.

Source: Google

Google's Translation Partner

Google released a bilingual AI conversation partner in beta, supporting over 70 languages with real-time translation.

Source: Cohere

Cohere's Enterprise Translation Model

Canadian AI org Cohere also launched a translation model: Command AI Translate, geared towards enterprise use.

They've claimed top scores on translation benchmarks with enterprise-grade security and GDPR compliance for business comms.

What this means for work:

  • Voice is becoming a default interface. Real-time translation brings new possibilities for international work.
  • Global collaboration gets easier. Business accounts do not use chat data for training, but consumer accounts do by default.
  • Customer service will change quickly. Voice agents are getting better at understanding nuanced convos.

3/ Other Signals Worth Your Attention

  • Amazon got an AI Shopping Upgrade. Amazon's AI Lens now includes real-time image search and live shopping features. Point your camera at something and Rufus helps you find and buy it immediately.

Signals to Act On

1. Check your AI privacy settings.

Make sure that data-sharing and training selections align with your preferences. If you use Claude personally, you have until Sept 28 to opt out of training.


2. Start testing AI voice capabilities.

Test out AI voice features available in your chat interface or dedicated software (Wispr Flow, Granola, OtterAI) that you can use to help dictate across applications.

Try speaking over typing. It can be a little weird at first, but if you keep it up, I think you'll like it.

Struggling to separate hype from opportunity?
Start with the 5-Day AI Advantage Challenge to build the AI fundamentals and critical thinking skills you'll need.
Join the 5-Day Challenge →
Not a subscriber yet?
Join Your AI Advantage Weekly for practical insights on AI, strategy, and the changing workplace.
Subscribe Free →
Found this useful? Forward it to a teammate who’s figuring out AI too.

Read more