Skip to content
OpenAI · China · GoogleEpisode 056 · 4 May 2025 · 37:55

AI Now Helps People Pass Interviews and Check Candidates—So Trust in Tests Is Collapsing From Both Sides

Central question

How is AI undermining trust in interviews from both the candidate's and the company's side?

What you take away

Understand how AI helps both the candidate and the evaluator—and why old tests no longer prove skill without a new verification process.

Main threads

What to watch for

1Compare “GPTZero: the AI-content detector industry” with “Google: the advertising monopoly trial”: they provide different criteria for judging the same issue.
2Test the conclusion from “DeepMind: scientists form the first AI union in the UK” 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 “A social network from OpenAI: rumor or reality?”.
4Define the owner of the outcome and the quality metric for the situation described in “Perplexity's own browser”.
Signals to track afterwards
Watch for actions by Apple and DeepSeek that confirm or challenge the episode’s central claims.
Compare new launches and policy changes with “DeepMind: scientists form the first AI union in the UK”: have access, quality, price, or constraints changed?
Check whether the scenario in “Perplexity's own browser” becomes repeatable practice rather than a one-off demonstration.
Most useful for
EntrepreneursInvestorsExecutives and managersLegal professionalsProduct leadersAI users

Key takeaways

03:05The practical meaning of the issue: cluely: how AI-assisted cheating works in job interviews

The “Cluely: how AI-assisted cheating works in job interviews” scene leads to a working conclusion: candidates really do use AI in interviews and tests — a hint or code can be produced unnoticed; employers respond with new, equally probabilistic checks, which erodes trust on both sides.

07:54Where the promise meets reality: euro-Use Law: AI in Education and Work

The “EU law: AI in study and at work” scene leads to a working conclusion: a rule on AI use in study and work works only if it is enforceable — who must prevent the violation, who records it, and who answers for the consequences; without that the norm stays a declaration.

11:38AI-content detectors promise to determine whether a text was written by a person

The decision in “GPTZero: the AI-content detector industry” depends on one criterion: detectors promise to tell whether a person wrote a text, but models change faster than the signals, and a nonbreaking space or a polished style is not proof; in education and hiring such a mistake can decide someone's fate.

16:45Why an announcement is not enough: a social network from OpenAI: rumor or reality

The decision in “A social network from OpenAI: rumor or reality?” depends on one criterion: OpenAI does not want to remain a model provider inside other people's platforms — a feed of its own gives data, advertising, and a direct relationship with the user; Perplexity is building a browser for the same reason.

21:07The market tests it through use: your browser from Perplexity

The practical meaning of “Perplexity's own browser” is that a browser of its own gives Perplexity direct access to the user and their data, but the announcement matters only when it actually changes how a person searches and works, not merely when it is declared.

24:04A platform, however, is more than distribution

In the context of “Google: the advertising monopoly trial,” this criterion applies: google has been found to hold a monopoly in online advertising, while YouTube and Netflix spent years building moderation and recommendation systems. OpenAI would have to answer for content, people, and abuse—not only for the quality of an answer.

28:19Who owns the outcome: elon Musk's largest AI rate: $20 billion

The decision in “Elon Musk's biggest AI bet: $20 billion” depends on one criterion: twenty billion shows the scale of the bet on xAI, but the sum alone does not prove a product — durability is decided by repeatable revenue, servicing cost, and the ability to reach the next stage of growth.

30:24China is developing Huawei chips, DeepMind employees are forming a union, and researchers are running into visa restrictions

The “DeepMind: scientists form the first AI union in the UK” scene leads to a working conclusion: Huawei chips, a DeepMind union, and visa restrictions on researchers show the race is as much for people and data as for models — and verifiable provenance of the work becomes the central value.

What this episode is about

GPTZero looks for traces of machine-written text, candidates use AI during interviews, a border officer checks a profession through ChatGPT, and OpenAI is considering a social network of its own. As generation becomes ubiquitous, platforms want to own not only the model, but the data that can tell them who is in front of them.

AI-content detectors promise to determine whether a text was written by a person. But models change faster than the signals, and an accidental nonbreaking space or a polished style is not proof. In education or hiring, such a mistake can decide someone’s fate even though the tool cannot produce a reliable expert judgment.

Candidates, on the other hand, really do use AI during interviews and tests. They can receive a hint, write code, or prepare an answer without being noticed. Employers respond with new checks, some of them equally absurd: a border officer or recruiter asks ChatGPT how to identify a real programmer and receives yet another probabilistic scenario.

Rumors of an OpenAI social network make sense in this context. The company does not want to remain a model provider inside other people’s platforms. A feed of its own would provide data, advertising, and a direct relationship with the user. Perplexity is building a browser for the same reason.

A platform, however, is more than distribution. Google has been found to hold a monopoly in online advertising, while YouTube and Netflix spent years building moderation and recommendation systems. OpenAI would have to answer for content, people, and abuse—not only for the quality of an answer.

China is developing Huawei chips, DeepMind employees are forming a union, and researchers are running into visa restrictions. All of this shows that the race is as much for people and data as it is for models. In a world where any text can be generated, verifiable provenance becomes the central value: who did the work, in what context, and with what responsibility.

Lower execution cost does not eliminate work; it raises the value of choosing the task, supplying context, and recognizing an error.

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 80 segments: 55 identified, 3 mixed, 20 probable, and 2 unresolved.

Loading…