Read more about it here: deepmind.google/discover/blo...

This year, the advanced Gemini model operated end-to-end in natural language, producing rigorous mathematical proofs directly from the official problem descriptions – all within the 4.5-hour competition time limit.

At IMO 2024, AlphaGeometry and AlphaProof required experts to first translate problems from natural language into domain-specific languages, such as Lean, and vice-versa for the proofs.

This achievement is a significant advance over last year’s result.

An advanced version was able to solve 5 out of 6 problems.

The result? Saves tons of time, money, and builds super reliable voice assistants that have undergone a rigorous evaluation process. No more guesswork! 📈
Full details + code here: towardsdatascience.com/let-ai-tune-...

GOOD NEWS: I built an #AutomatedPromptEngineering (APE) pipeline specifically for voice AI! 🤖✨ My new @towardsdatascience blog post dives deep.
What it does:
✅ Creates diverse audio tests
✅ Automates performance eval
✅ LLM optimizes your prompts! 👇

Thanks for sharing @towardsdatascience.com 🤗

Tom's Guide: www.youtube.com/watch?v=Pt3q...

WSJ interview: www.youtube.com/watch?v=NTLk...

Building a v1 GenAI app on an existing platform while overhauling the foundation for a better 'V2' is a common strategy. But explaining this to everyday consumers is challenging. These kinds of interviews really help communicate that effectively. What did you think?

To give an example, right out of the gate she asks, "𝙇𝙖𝙨𝙩 𝙮𝙚𝙖𝙧 𝙮𝙤𝙪 𝙖𝙣𝙣𝙤𝙪𝙣𝙘𝙚𝙙 𝙖 𝙨𝙢𝙖𝙧𝙩𝙚𝙧 𝘼𝙄-𝙙𝙧𝙞𝙫𝙚𝙣 𝙎𝙞𝙧𝙞. 𝙒𝙝𝙚𝙧𝙚 𝙞𝙨 𝙨𝙝𝙚?"
From a developer's point of view, Apple's answers made a lot of sense: a 'V1' worked, but didn't meet their high quality/reliability standards when users went 'off the beaten path'.

This year, however, Craig Federighi and Greg Joswiak were interviewed by other outlets, including Tom's Guide, TechRadar, and The Wall Street Journal. I particularly liked Joanna Stern's interview and her style: direct, concise, and challenging.

While it's not fair to characterise Gruber as an "Apple fanboy," I consistently found his questions too long-winded and too softball. By the end, it often felt (to me, at least) like just a few folk were a bit too cosy on stage.

i definitely hear you on that one 😅 out of curiosity - what are the benefits you are looking to gain from an agent framework (in general)?

Not perfect by any means, but much better already than "traditional" voice assistants, and we are only at the beginning of this journey.

You can try it yourself with the Developer Guide for Gemini's Multimodal Live API 🤗

github.com/heiko-hotz/g...

I believe that multimodal AI models have the potential to change that. They allow me to speak much more freely about what I want them to do and oftentimes they understand and execute in the way I expected them to.

But soon I realised that these voice assistants still require a rigid syntax: I would have to phrase commands in a very specific way for the voice assistant to understand what I meant.

To me it was a magical moment when I got my first Amazon Echo in 2015 and could just shout words into the air and got a response.