This is TechCrunch.

This episode is brought to by Modulate.

Speech is more than just text.

It's all about how you say something.

With Modulate's AI model Velma, you can decode intent,

emotion, stress, behavior, and authenticity in messy

multilingual audio.

You get human-level voice intelligence made for real-world conversations.

I mean, Velma was built on 21 billion minutes of audio, so you know what that means.

Velma is fast, cheap, and accurate.

I think you'll really like Velma because it's different from typical AI.

Go beyond transcripts and see what a voice native AI model can really do.

Go to Modulate's live, ungated preview of Velma and preview..commodulate.aI.

That's preview.modulate.com to see why Velma ranks number one on leading benchmarks for No crumbs on your keyboard or sauce on your dashboard. Just all the flavor of sweet chili chicken, barbecue and bacon chicken and new teak chicken. Rapt up neat.

Served after 11 a.m. Tika chicken wrap available until the 13th July 26. Fies apply to delivery orders subject to availability.

Amazon's ring canceled their partnership with Flock, a network of AI cameras used by ICE,

feds, and police. I'm I'm Miron-Chake, and your Daily Crunch for Saturday starts right now.

Starting Friday, Open AI will cease providing access to five legacy chat GPT models, including the popular but controversial

GPT-40 model. The 40 model has been at the center of a, let's say, number of lawsuits

concerning user self-harm, delusional behavior, and AI psychosis. It remains open AI's highest

scoring model for sycophancy for crying out loud.

In addition to GPT-40, the GPT5, GPT4.1, GPT4.1 mini, and OpenAIO4 mini models have also been deprecated.

...