2023 Recap: The Year of AI… and Especially LLMs

Are LLMs overshadowing other AI advancements? How effective is RLHF on multimodal models? Is attention really all you need? Find out here!

Welcome (back) to AI Minds, a newsletter about the brainy and sometimes zany world of AI, brought to you by the Deepgram editorial team.

In this edition:

  • 💼 The case for LLMs overshadowing other AI Advancements

  • ⚔️ Why the AI Business is very much like an ongoing war

  • 🐶 Robot Dogs and the AI Summit Conference in NYC

  • 📊 Test of Time Award Paper: Learning Fair Representations

  • 👀 Attention is NOT all you need

  • 🔊 Multimodal RLHF

  • 🌺 Review: International Conference on Machine Learning 2023

  • 🐦 Satya Nadella on AI in 2023

  • 🔐 The NSA on AI and Cybersecurity

  • ❗Bonus content: AI Apps, Generative Music AI, Mixture of Experts

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

We coded with the brand-new Whisper-v3 over the past week, and the results were not what we expected. Check it out here!

🐎 LLMs shined in 2023… but did they overshadow other AI?

Are LLMs and Diffusion Models overshadowing other AI Advances? Find out in this article! From ChatGPT to Llama to Gemini, it seems like the excitement about LLMs never stops. But what about Computer Vision? Self-driving cars? Biotech AI? Perhaps there’s some news we all missed during large LLMs announcements.

A Year’s Worth of Battles in the AI Business - Of course, we can’t talk about AI in 2023 without discussing the business side of tech. Whether it’s the initial boom of AI early this year, or Google’s journey from Bard to Gemini, or even the fiasco with Sam Altman and OpenAI last month, business can actually be a war. And this article reveals the breakdown.

End-of-the-year Conference: AI Summit, NYC 2023- And for some more lighthearted news, here’s a recap of one of the final AI events of 2023: The AI Summit in New York! From voice-commanded robot dogs to the announcement of brand new text-to-speech models, check out this article to see what such a huge event entails 🥳

🧑‍🔬 Research Recap: Some incredible AI papers of 2023

We could list the Gemini Paper, but we already did that last week…

We could list the Llama-2 Paper, but we already analyzed it in this video and this article

Instead here are some highlights of AI Research in 2023, and even some before!

First off, we have to give kudos to Learning Fair Representations, a paper published in 2013 that received the “Test of Time” Award in this year’s International Conference on Machine Learning (ICML).

Next, it’s crucial to discuss AI and Safety, especially with respect to minimizing the spread of false information and AI hallucinations. And so, the paper Attention is Not All You Need brings to light some fantastic insights about AI and safety.

The AI Community, in general, loves RLHF. We also love the idea of top-tier, reliable multimodal models. And so, this paper is a must-read since it combines both ideas into one: Aligning Large Multimodal Models with Factually Augmented RLHF.

🎥 International Conference on Machine Learning 2023

If you want to see what the AI landscape looked like much, much earlier in the year, check out this video! We also cover the Test of Time Award, mentioned above, at the “Day 5” scene. 🚀

🐝 Social media buzz

Here’s what some people are saying on social media about the evolution of AI in 2023!

📲 Apps to Check out

Integrating AI and LLMs into your application?

Graphlit is built on a serverless, cloud-native platform, Graphlit automates complex data workflows, including data ingestion, knowledge extraction, LLM conversations, semantic search, alerting and webhook integrations. Whew, that’s a mouthful.

Check out this recent blog post here on how Graphlit, in collaboration with MapScaping, Deepgram, and OpenAI GPT-4 Turbo, revolutionizes content workflows for podcasters.

AI Apps Feature here

Heyday is an AI-powered memory assistant that resurfaces content you forgot about. It saves all the articles, documents, emails and transcribed meetings into one place. Then it goes one step further, it resurfaces them when you need them. 

Professional coaches use Heyday to create meeting notes, glean ideas from research, and surface insights from their conversations with clients. 

AI Apps Feature here

🤖 Additional bits and bytes

AI Can Sing: A Look at Generative Music Models - “More recently, this new wave [of generative AI] has inspired a new wave of automatic music generation using deep learning neural network frameworks. While some of these AI-generated music is completely made up by the model, some are created using the AI versions of the voices of popular musicians.”

Five Must-Follow Researchers: Household names in ML - Who is currently on the Mount Rushmore of Machine Learning? What names are on the walls of the AI Pantheon? Who are The Beatles of LLMs? And most importantly, who are the Ringos and who are the Johns? Find out here 📈

Mixture of Experts: How an Ensemble of AI Models Decide As One - “[T]he efficacy of neural networks hinges heavily on their parameter count. Mixture-of-Experts (MoE) presents an efficient approach to dramatically increasing a model’s capabilities without introducing a proportional amount of computational overhead.”