Mind-Bending Capabilities in New Models

The new Claude 3.5 is sweet, but it's about more than that

Thank you so much to those who have invested a $30 paid subscription. This blog is a lot of work, so thank you so much for the support, and thanks to your investment my mix of free and paid posts can continue.

Over the last couple of days, there has been a spat of new model releases. Today’s release focused on the release of Claude Sonnet 3.5 and their claim that it marginally exceeds ChatGPT-4o on common benchmarks. Perplexity also claimed it’s the top model based on their internal benchmarks.

While this is certainly a beneficial improvement, what I think is more significant is its new capabilities. Watch this demo on how to use Claude to help you prep useful course materials, not just write a lesson plan.

Practically speaking, it’s powerful.

On June 17, Google released V2A, which auto-generates audio based on any video scene.

Of course, as mentioned, this isn’t this week’s only news. Just yesterday, MIT released an open-source model, including the data set, that allows text-to-voice translation for 7,000 languages!

And today, we saw the release of an AI math tutor that may actually work. Getting AI models to reason through math problems has always been a challenge. Let’s see if this one is any better.

Things are moving quickly, and applications are powerful, AGI or not.