Skip to content
Google · OpenAI · GoEpisode 057 · 11 May 2025 · 42:21

Voice Assistants Still Make Mistakes, While Scammers Already Use AI at Full Scale

Central question

Why do voice assistants still make basic mistakes while fraudsters already use generative AI at full scale?

What you take away

Compare the maturity of legitimate voice assistants and fraudulent uses through accuracy, speed of adaptation, data access, and the user’s ability to opt out.

Main threads

What to watch for

1Compare “Collective action for Apple Intelligence” with “About Alexa Plus: plus and minus”: they provide different criteria for judging the same issue.
2Test the conclusion from “Cybercrimes in AI” in your own use case—what actually changes in the process and what remains a promise.
3Before choosing a product or approach, record the constraint identified in “UAE: federal laws, with AI”.
4Define the owner of the outcome and the quality metric for the situation described in “Why is "referring" more important than repeat‐founder experience”.
Signals to track afterwards
Watch for actions by Apple and Google that confirm or challenge the episode’s central claims.
Compare new launches and policy changes with “Cybercrimes in AI”: have access, quality, price, or constraints changed?
Check whether the scenario in “Why is "referring" more important than repeat‐founder experience” becomes repeatable practice rather than a one-off demonstration.
Most useful for
AI usersProduct teamsEntrepreneursExecutives and managersContent creatorsDesigners

Key takeaways

00:00The market tests it through use: in this issue

The discussion of “In this issue” yields a practical test: the important signal is not one funding number: the next round, available runway, and closure rate show whether a company can survive the new cost of capital.

04:00The boundary between value and constraint: gemini Metriki

The boundary of the “Gemini Metriki” case is defined by this point: the important signal is not one funding number: the next round, available runway, and closure rate show whether a company can survive the new cost of capital.

05:55Apple Intelligence promised features that were delayed or did not work as expected

The practical meaning of “Collective action for Apple Intelligence” is that this section clarifies the mechanism behind the topic and preserves a constraint that would otherwise be lost in an overly simple conclusion.

07:00Amazon describes the Alexa Plus problem honestly: the accuracy of complex multi-step commands may remain at a level that is unacceptable for a household assistant

The discussion of “About Alexa Plus: plus and minus” yields a practical test: this section clarifies the mechanism behind the topic and preserves a constraint that would otherwise be lost in an overly simple conclusion.

10:59Why context matters more than one metric: chatGPT experience

For the “ChatGPT experience” scene, the decisive point is this: the important signal is not one funding number: the next round, available runway, and closure rate show whether a company can survive the new cost of capital.

14:07How the issue moves from news to product: transpa Decree: AI‐Chunglasm K‐12 throughout the country

The discussion of “Transpa Decree: AI‐Chunglasm K‐12 throughout the country” yields a practical test: the important signal is not one funding number: the next round, available runway, and closure rate show whether a company can survive the new cost of capital.

16:00Scammers do not need that level of reliability

The practical meaning of “Cybercrimes in AI” is that this section clarifies the mechanism behind the topic and preserves a constraint that would otherwise be lost in an overly simple conclusion.

20:52The UAE is going to the opposite extreme by officially using AI to analyze and draft legislation

The “UAE: federal laws, with AI” issue should be assessed with one constraint in mind: the key test is enforceability: who must prevent the risk, who records the violation, and who is accountable for the consequences.

22:13What determines the outcome: startap in defence: 21-year-old faunder attracted +$185 million

In the context of “Startap in defence: 21-year-old faunder attracted +$185 million,” this criterion applies: the important signal is not one funding number: the next round, available runway, and closure rate show whether a company can survive the new cost of capital.

23:55OpenAI and Anthropic assembled a rare concentration of talent before the mass boom

The “Why is "referring" more important than repeat‐founder experience” issue should be assessed with one constraint in mind: new startups find that hard to reproduce, so execution speed matters more than a repeat founder’s biography. Devices—glasses, pendants, and voice assistants—also add continuous data collection. The market matures only when it shows not the promise, but the accuracy, the rules for training on data, and the ability to opt out.

What this episode is about

Apple faces a lawsuit over unfulfilled promises, Amazon acknowledges low accuracy in multi-step agents, and ChatGPT and Meta are testing voice. While large companies refine their interfaces, cybercriminals use generation without any quality requirement—they only need to fool a small percentage of people.

Apple Intelligence promised features that were delayed or did not work as expected. The class-action lawsuit matters not because of the amount, but because of the principle: the company sold a device through future AI, while the user paid in the present. Summaries in Mail are useful, but they do not replace the promised intelligent Siri.

Amazon describes the Alexa Plus problem honestly: the accuracy of complex multi-step commands may remain at a level that is unacceptable for a household assistant. If an agent sometimes orders the wrong thing or fails to carry out an action, the person stops trusting it altogether. ChatGPT’s voice mode sounds more natural, but it also requires verification.

Scammers do not need that level of reliability. They can generate thousands of emails, voices, and messages; a one-percent success rate is enough. Cybercrime therefore captures the benefit before the ordinary user does—the criminal does not have to build a stable product or maintain a reputation.

The UAE is going to the opposite extreme by officially using AI to analyze and draft legislation. A system can find contradictions more quickly, but political choice and responsibility cannot be delegated to it. The higher the level of the decision, the more important it is to see the underlying data and a human approval.

OpenAI and Anthropic assembled a rare concentration of talent before the mass boom. New startups find that hard to reproduce, so execution speed matters more than a repeat founder’s biography.

Devices—glasses, pendants, and voice assistants—also add continuous data collection. The market matures only when it shows not the promise, but the accuracy, the rules for training on data, and the ability to opt out.

The voice-assistant market will mature only when accuracy, training-data rules, and the ability to opt out matter more than promises.

Episode transcript

The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 108 segments: 71 identified, 3 mixed, 13 marked with ✓, and 21 unresolved.

Loading…