> ## Content Index
> Fetch the complete content index at: https://www.oliphonic.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# It's getting harder to tell who (or what) is real...
- URL: https://www.oliphonic.com/its-getting-harder-to-tell-who-or-what-is-real/
- Published: 2026-10-02T11:54:14.000Z
- Updated: 2026-10-02T11:54:14.000Z
- Author: Oliphonic

*Progress on improving the realism of AI-generated humans is rapid & increasing. The results are equal measures impressive & deeply worrying.*

Tavus, an AI video startup, has previewed **Griffin** \- what it calls a "Human Interaction Model". Unlike a conventional chatbot, Griffin does not simply wait for you to finish speaking, transcribe your words, think for a bit and then respond with a tell-tale slightly clunky paragraph. It watches and listens continuously. It can nod while you are talking, react to something visible on your screen, interrupt, pause, change its expression and respond to the rhythm of the conversation. In other words, it attempts to behave in a 'human' way.

[www.tavus.io](https://www.tavus.io/?ref=oliphonic.com)

The 'scientific' results are certainly not bad. In a one-minute video-call study involving 54 participants, 26 — 48% — believed they had been speaking to a real person. Their website rather boldly claims that it is the first AI to pass the Turing test... though the statistician in me can't help but cringe at the sample size there.

**So let's be upbeat for a second** \- there are positives to this type of technological advancement! AI personas that are better able to pick up on visual cues, faster at understanding & responding to the incredible complexities of voice intonation & pitch, better companions, better listeners. AI agents that are more 'human' and simply nicer to interact with.

Alas there are also clear potentials for misuse. An AI insurance assistant that watches & digitises every tweak of your facial expressions & intonation as you recount a claim or circumstance, feeding this back into centralised probability models to accept or reject claims. An AI social media companion that is learning precisely which emotional buttons to press to keep you coming back. The **perfect friend** tuned to observe & react in precisely the right way to establish deep emotional & social connection.

![](https://storage.ghost.io/c/9c/fa/9cfabe9c-bcc0-471f-a6fa-271bee166fa4/content/images/2026/10/AI-Robot_-Help-and-Hype.png)

And of course the most obvious misuse which would be the user not even realising that they are actually talking to an AI. AI can now convincingly impersonate a voice after sampling just 10 seconds of spoken word. Imagine one of your friends or relatives facetiming you and being **absolutely convincingly them** \- except they're not.

AI companies need to be pushing forward responsible & ethical use of AI systems just as much as they're pushing forward technological advancement. Sadly, I'm just not that convinced that that's what is really happening. Regardless, the tech companies can't be the single point of failure - it will always require regulatory input, government legislation and (as usual) the user being as savvy as possible when engaging with anything that may indeed be.... artificial.

Regardless, Tavus has clearly demonstrated that the technology is moving quickly. We just need to make sure we can still tell which bits are 'tech' and which are human...