AI news you might’ve missed so far in 2024

From AI poker to ecoacoustic machine learning, here are the new models & algorithms that may have slipped past your feed this year

Just as I was about to hit send on recent AI news… BOOM! We get the leak of Mistral’s new model.

But don’t get too excited. According to Mistral CEO Arthur Mensch the model is an old version retrained from Llama 2. That still didn’t keep the internet from blowing up about it.

In this edition:

  • 🃏 Why AI sucks at poker, but is great at chess.

  • 🩻 Doctors judge Retrieval LLMs’ skills in clinical medicine

  • 📲 Five new trending AI Apps to check out!

  • 🔉 Using AI to transform children’s voices into adult voices.

  • 🥬 ML, Ecoacoustics, and analyzing forests

  • 🎓 Full Stanford lecture on Retrieval-Augmented Language Models

  • 🐦 Twitter: Forbes and Mustafa Suleyman’s latest AI updates

  • 🔍 Beam Search, Latent Dirichlet Allocation, and other AI glossary entries!

Thanks for letting us crash your inbox; let’s party. 🎉

Oh yeah, and while you may be familiar with Deepgram’s speech-to-text API, you might want to check out our upcoming text-to-speech technology as well 🥳

🧑‍🔬 Shockingly Underrated New AI Research

A Survey on Game Theory Optimal Poker - Games like chess rely solely on skill, while games like roulette rely purely on luck. AI can learn the skill-based games, but not the luck-based ones. Somewhere in the middle of this skill-to-luck spectrum is poker, where strategy won’t guarantee victory, but rather maximizes your chances of winning. This paper goes over the role of ML in poker and other half-skill/half-luck games.

Almanac — Retrieval-Augmented Language Models for Clinical Medicine - We tasked a panel of eight board-certified clinicians and two health care practitioners with evaluating Almanac, an LLM framework augmented with retrieval capabilities from curated medical resources for medical guideline and treatment recommendations. The results lie within.

Have you seen our piece of the internet dedicated to showing off AI Apps? Here are the top 4 trending apps over the past week.

Spark Engine - Put the future of AI conversations in your hands with full customizability.

SparkGPT, meanwhile, is an AI chatbot created by Spark Engine that allows unprecedented customization and control over large language models like GPT-3.5 16k and GPT-4. It represents the next evolution in conversational AI.

Toma - Automate phone calls for your business with AI

Toma is an AI voice platform designed to automate both inbound and outbound business calls.Toma's AI agents can handle customer calls, schedule appointments, qualify leads, and respond to inquiries around the clock, offering an efficient solution for customer communication.

Zappr - Train customized AI chatbots in minutes with no coding using their revolutionary platform

Heyday - Get actionable insights from your information, automatically

Heyday is used by modern professionals to create meeting notes, glean ideas from research, and compose articles based on previously read material and interactions.

Heyday is also part of the Deepgram Startup Program.

🏇 AI Voice and Ears: Journeying into auditory models

Child-to-Adult Voice Style Transfer: A Case Study in Auditory AI - Machine Learning Developer Advocate Jose Francisco walks through a Stanford research project of his in which he attempted to use AI to transform children’s voices into adult voices. Audio files and architectures are in the link above!

Listening to Forests with Machine Learning & Algorithms- This article is essentially a case-study in ecoacoustics. Brad Nikkel, also from Stanford, takes an in-depth look at machine learning’s role in analyzing elephant vocalizations in a deep, dense forest. And the audio that the scientists picked up isn’t exactly what you’d expect…

🎥 Stanford’s Lecture on Retrieval Augmented Language Models

RAG was all the rage a couple months ago. So much so that Stanford actually released a lecture on retrieval-augmented language models in January (though the original lecture was recorded in early December 2023).

🐝 Social media buzz

🤖 Glossary Pages

If you like AI Minds and want to expand your knowledge even further, check out the Deepgram AI Glossary! We update the pages every week to give you the most up-to-date and relevant information on a wide variety of terms. Here are a few:

Beam Search Algorithm - When you have a decision tree with too many branches—whether you’re building a chess bot or a recommendation system—this algorithm can help!

Mistral - Want to learn about one of the newest multimodal AI models out there? Check out this page on Mistral!

Latent Dirichlet Allocation- LDA is a probabilistic generative model that analyzes documents to discover latent topics and themes that are present across a collection of texts. To see how it’s (surprisingly) used, check out the link!