😸 Your 24/7 language coach

PLUS: Google dropped 3 new models.

Welcome, humans.

Any fans of the Pixar movie Up out there? This video DEFINITELY reminds us of Up, except what if the friendly old couple was living their best life, going on dates and dancing the night away at jazz clubs.

This video is so adorable, and it’s not even made with Midjourney to Dream Machine, Kling, or Gen 3 Alpha. Speaking of those videos… they’re getting a little, ahem, out of control…

Here’s what you need to know about AI today:

People shared a lot of crazy Advanced Voice Mode demos.
Google released three new open-source AI models for local device use.
Taco Bell announced it will go ahead with AI drive-thrus.
A study found “AI” in product descriptions lowered trust and purchases.

Advertise in The Neuron here.

People are getting Advanced Voice Mode access, and the things it can do are pretty crazy.

Ever wanted to learn a new language from a native speaker? Y'know, ask loads of questions, like "how’d you pronounce this", "how’d you say that", "hang on, repeat it slower" ...again and again...without driving your tutor up the wall?!

Now, learning a new language is a 10/10 use case with ChatGPT’s new Advanced Voice Mode (talking AI)—if you can get access, that is!

Don't mind us, just waiting our turn, internally weeping 'cause we're not on OpenAI's cool early access list.

To fuel our FOMO, we’ve turned to social. First take? It’s fast, zero lag, and quite entertaining. It gets pauses too (it adds commas for breath breaks in its transcripts) and is way better about not interrupting you while you talk.

Folks on X and Reddit are sharing demos—here are 8 examples of what people are already using Advanced Mode for:

Helping you learn a language (like Spanish) across multiple accents (#1, #2, #3).
Cracking jokes at different volumes and moos (link).
Reciting a text while adapting to live feedback (#1, #2).
Telling you a story with background sound effects (link).
Playing both roles in a scene (link).
Belting out a tune with a specific vibe (#1, #2)— although sometimes it refuses.
Rattling off the fine print at the end of drug ads (link).
Get it to yelling—with echo! (link).
Count super fasting—IDK why, but cool to know! (link).

It even beatboxes!

What you can’t do:

Use it with custom GPts.
Type to it and then have it talk back
Make sound effects—unless it's “performing” them (or should we say, PURR-forming them?)

How well does it pull all of this off? It's kinda hit-or-miss.

On accents, Grant’s Venezuelan friend Raul assures us the Venezuelan accent is only ~70% right—to him, it's more like a neutral Mexican accent from a Spanish dub.

On a second listen, it seems Advanced Mode only really gets local words, not actual accent.

Plus, the OG voice actor's accent sticks around, no matter what tweaks you ask for.

Hopefully OpenAI licenses more actors to pick from, like from other AI voiceover startups, as this could boost its accent game. Oh, and for those keeping score—there was no ScarJo voice.

It’s also funny that ChatGPT says it has to breathe before speaking. This is probably a good idea, cause Advanced Voice is by far and away the most human-like, talking AI we’ve seen yet. It really does feel like there’s a human on the other side of the screen, even though it definitely (and sadly) isn’t ScarJo.

So when do the rest of us get access to it? Looks like later this fall, but we’ll keep you posted.

FROM OUR PARTNERS

Want to know which AI model hallucinates the most? Check out this report.

If you want to master RAG and build AI apps you can trust, you need to know which AI models actually tell the truth.

Galileo’s latest LLM Hallucination Index spills the tea:

22 popular models are put to the test.
3 key RAG tasks compared under the microscope.
Some surprising results that might blow your mind.

Want to know which ones shine on accuracy and which ones tend to make stuff up?

Spoiler alert: Turns out, bigger models aren't always better.

Don't risk your live AI deployments on unreliable AI. Get the facts. Make smarter choices. Read the report.

Around the Horn.

SAM 2 can automatically identify and outline objects in both images and videos, making it easier to edit or track specific items across frames (download here, or try it on the web).

Google released a trio of new open-source AI models (Gemma 2 2B, Gemma Scope, and ShieldGemma) that can run locally on your devices.
Taco Bell announced it will move ahead with AI ordering in drive-thrus, even after McDonalds decided to stop its own AI drive-thru pilot.
The U.S. Copyright Office proposed an “urgent” new federal law to provide protection against the unauthorized creation + distribution of AI deepfakes.
A new study found that mentioning AI in product descriptions lowers emotional trust and reduces purchasing decisions, especially for “high risk” products.

Treats To Try.

*Future-proof your career in minutes a day. Join 10M professionals learning AI with Brilliant's bite-sized, interactive lessons. Start your free 30-day trial here.
Bex is a Slack integration that automatically converts team messages into a searchable knowledge base and wiki.
Hooper automatically tracks basketball stats and generates highlight clips from game footage recorded on your phone.
Toby is an desktop app that provides real-time speech translation for video calls across multiple languages and platforms.
Simply Draw offers personalized drawing tutorials and feedback to help users learn and improve their drawing skills.
Axle automates compliance operations tasks, promising to increase efficiency and reduce backlogs for financial institutions.
Imagetocaption.ai generates customized social media captions from uploaded images and videos.

*This is sponsored content. Advertise in The Neuron here.