Slavko Kovačević

co-founder @deepmark_me (YC F26)

[Slavko Kovačević @slavzilla]

Today we launch @deepmark_me (@ycombinator F26). We put an inaudible ID inside an AI agent's voice, so the person on the other end of a phone call can verify who is really speaking. Here is me doing it live, on one phone.

[Slavko Kovačević @slavzilla]

@ericmigi I see it working as a secondary phone, but not as a primary (and only) device

[Slavko Kovačević @slavzilla]

@nubsvault 1. No, their ToS does not let them do that, so we didn't sign up for it.

2. Two wrongs don't make a right.

[Slavko Kovačević @slavzilla]

This is a massive L take.

We open-source most of our research, as we love sharing knowledge. But we deserve to be credited if someone uses it.

I would be pretty mad if someone wrote a prompt, "audio watermarking Claude help pls" and it spat out our proprietary models.

[nubs (spirit/acc) @nubsvault]

if you seriously are upset about someone "stealing" your research as their own you have lost the plot entirely

why cant we just be happy we made this advancement and keep moving forward?

everyone is doing everything i doubt anthropic directly stole this too

its like the whole navier stokes shit all over again

[Slavko Kovačević @slavzilla]

@nubsvault No, that doesn't make any sense. Claude is a tool, not a shadow library we all fill without consent. I reserve the right to own my IP.

[Slavko Kovačević @slavzilla]

@benhylak sadly, the entire internet is mostly dead... humans are just scrolling, and AI is posting slop endlessly

[Slavko Kovačević @slavzilla]

@jack Does it carry a signal telling the receiving end that it is AI? If not, it might get you in trouble with the upcoming regulations.

[Slavko Kovačević @slavzilla]

We love open-sourcing! And the name is no coincidence! Kosta's wife's name is Ana, and so is my wife's name 🙂

[Slavko Kovačević @slavzilla]

Decades of signal processing progress, but the best way to compare two audio samples was by listening.

Not anymore.

DeepMark is proud to announce the Auditory Nerve Noticeability Index (ANNI).

Beats SOTA standards, including ViSQOL and PESQ.

pip install deepmark-anni

[Slavko Kovačević @slavzilla]

@FourVork 🍍pineapple🍍attack🍍still🍍works🍍?🍍

[Slavko Kovačević @slavzilla]

@vibeistani @VeezaAI

[Slavko Kovačević @slavzilla]

@NikhilKMurthy The ultimate interface needs an ultimate trust layer. Right now, any voice can claim to be anyone. There's also a compliance angle, and the deadline is in a month. Happy to help!!

[Slavko Kovačević @slavzilla]

@blocks Applied but silence. @blocks 👇

[Slavko Kovačević @slavzilla]

Starting November 7, xAI will add text watermarking to Grok API models by default. This means every major LLM is now watermarked.

[Slavko Kovačević @slavzilla]

@CharlesFLehman Trusting tools like this one is basically believing in fairy tales.

[Slavko Kovačević @slavzilla]

OpenAI researched text watermarking for years but didn’t put it into production. When compliance deadlines were near they let Anthropic go first and take all the hate, haha. You gotta respect that, very smart 😆

[Slavko Kovačević @slavzilla]

@_junaidkhalid1 This is completely fair and your open weights will have to do the same. Bad take

[Slavko Kovačević @slavzilla]

Ladies and gentlemen, a new record has been set in the AI voice fraud market.

One hacked voice and $108M departed Italy's biggest bank. Poof!

https://t.co/pQqpWlXbDD

[Slavko Kovačević @slavzilla]

@videhps endless source of papers nobody will ever read, but they’ll get cited by other equally important papers - it’s like sudoku

[Slavko Kovačević @slavzilla]

@kwindla Wel done!!

[Slavko Kovačević @slavzilla]

The title is my takeaway from a new study by security researchers at Brno University of Technology, published in August. They tested how well people catch AI voices by ear. Spoiler, it doesn’t go well! https://t.co/ZNGrTcG6KC