Gamma Presentations, Intel's Gaudi 3, Canada and Korea Spend $B's
Tools and news that are (will) changing our AI work.
Morning y’all!
We’re quickly working our way through April and I’m grateful to have everything operating as expected, especially my health. And as I get older I appreciate this more and more as it doesn’t take much for any of us to be badly benched due to illness.
So keep your mind, body, and spirit in a good place and you’ll be happier too. Have a good one folks!
※\(^o^)/※
— Summer
Gamma is a free-to-use AI tool that helps users create beautifully designed presentations. You don’t even need any design or code skills, too.
Intel is trying to reverse course and AI, of course, is one of their solutions. Their new Gaudi 3 was unveiled for helping companies deploy generative AI as scale, promising 50% faster training and inference than NVIDIA’s H100 chips. .
Another internal post via GitHub on how they are using Copilot in their own work, even helping them learn new skills:
Twenty-three minutes and nine seconds later, he successfully produced a functional version written in Rust, despite having no prior experience with the language.
That’s pretty neat.
Google opens up Gemini Code assist and can run in popular IDEs, generating new code, explaining existing code, and more. But not everyone is optimistic.
Canada is trying to secure itself for and against AI and Justin Trudeau announced a $2.4b package to invest more into the space. It’s fascinating to watch nation-states invest into industries like this and their behavior will decide a lot of where our money will go. South Korea is dropping $7b too.
LLM training in simple, pure C/CUDA. That’s it. That’s the post. A big win for folks who wanted to optimize.
Ferret UI is some research by Apple that makes multimodal models understand user interfaces. Makes sense since there’s so much information that can be gleaned by simply understanding what the heck is on our screen at any given moment. Siri anyone?
Morph is an open source project that is essentially an answer engine with a generative UI. It’s a clean looking project, that’s for sure.
Ideatum creates color schemes with font pairings for design ideas. Not terrible.
Private LLM allows you to run local models for Apple silicon.
Melodisco is a music discovery AI platform. That’s it. That’s the post.
Anthropic shares some research on “persuasiveness” and how they measure them. You won’t be surprised by the findings:
Within each class of models (compact and frontier), we find a clear scaling trend across model generations: each successive model generation is rated to be more persuasive than the previous.
I’d hope that’s the result. But, what does this say about our use? I can imagine a future where many of our (shopping) engagements are being managed by AI.
Instructor is a python library that helps get structured data from LLMs.
Economists doubt that AI are creating productivity gains (non-paywall).
I really like their overview (and name):
R2R, short for RAG to Riches, provides the fastest and most efficient way to provide high quality RAG to end users. The framework is built around customizable pipelines and a feature-rich FastAPI implementation.
It’s an open source project too.
And that’s it! Have a great day folks!
\( ﾟヮﾟ)/
— Summer