• Modern Chaos
  • Posts
  • MC.23: CEO Sam Altman Teases GPT-5 "Better at Everything"

MC.23: CEO Sam Altman Teases GPT-5 "Better at Everything"

Discover OpenAI's strategic move, the intriguing prospects of GPT-5, Ollama's compatibility with OpenAI, the unique capabilities of AI assistant Mindy, and the arrival of multilingual AI model Aya.

modern chaos issue 23

Modern chaos is a newsletter exploring tech and AI through the journey of a dev agency turning into a startup studio. We share our notes, analysis, and experiments, plus a bit of Emacs tips.

Hello everyone,

The trend towards AI agents is accelerating. Lindy.ai, which raised $50 million to date, has gone public 2 weeks ago. Mindy is also entering the market, lots of research is being done, and OpenAI is not at rest, as you will discover in the updates section.

Last week, I asked how you feel about giving your credentials to AI agents to work on your behalf. I'm keeping the poll open this week to gather more answers, and we will debrief the results next week.

What's your stance on AI agents having access to your accounts?

How comfortable are you with this idea, especially if it could amplify tenfold your productivity on a specific task?

Login or Subscribe to participate in polls.

This issue is massive with 12 updates, more than my usual 7... the pace is going strong in this space.

Enjoy,

In this issue:

🤖 OpenAI: The New AI Battleground for Software and Automation
🧠 OpenAI: CEO Sam Altman Teases GPT-5 "Better at Everything"
🔌 Ollama: OpenAI Compatibility Now Available
💻 Mindy: The AI Assistant You Need
🤝 Artisan AI: Advanced Digital Workers
🌍 Cohere For AI: Aya, A Global Initiative for Multilingual AI
⚡ Reka Flash: An Efficient and Capable Multimodal Language Model
🎤 ElevenLabs: AI that Transforms Your Voice
🌎 NVIDIA: CEO Calls for Sovereign AI for Every Country
⚖️ Humans VS LLMs: The Rise of Legal Language Models
👥 RESEARCH: More Agents Is All You Need
🔀 RESEARCH: Large Language Models Self-Compose Reasoning Structures

Updates & tools

OpenAI: The New AI Battleground for Software and Automation 
They are moving towards creating AI agents that can automate tasks and operate devices. One type can transfer data between documents, fill out forms, or manage expense reports. The other type is web-centric, designed for tasks like collecting public data or booking travel. Rumors suggest that OpenAI may transform ChatGPT into an advanced personal assistant. This AI assistant can compose emails or documents in the employee's style. It remains unclear if this assistant will be a standalone product or part of a more comprehensive software suite. MORE

OpenAI: CEO Sam Altman Teases GPT-5 "Better at Everything" 
At the recent World Government Summit, OpenAI CEO Sam Altman shared some enticing details about GPT-5. He stated that the new model will be "smarter" and probably faster and more multimodal. What caught our attention was his emphasis on the model's generality - it's going to be "a little bit better at everything," not just one particular task. While we don't have a release date yet, the anticipation is palpable. MORE

Ollama: OpenAI Compatibility Now Available 
Ollama has recently built-in compatibility with the OpenAI Chat Completions API. This move allows you to use a range of tools and applications that work with OpenAI, directly with Ollama. You can now test your workflows locally on your computer, giving you more control and accessibility. MORE

Mindy: The AI Assistant You Need 
Looking for a tool to help you with everything from complex research to shopping for great deals or organizing your meetings? Meet Mindy. What's unique about Mindy is that you can interact with it using your regular email client. Considering that AI agents often take a while to perform their actions, an asynchronous channel like email seems like a great fit. MORE

Artisan AI: Advanced Digital Workers 
Artisan AI enters the game with Ava, a digital sales rep who specializes in outbound sales. Ava can be ready to go in just 10 minutes. As AI agents become more common, it's interesting to speculate how they might change the face of sales, and more broadly, how people will respond to these changes. MORE

Cohere For AI: Aya, A Global Initiative for Multilingual AI 
Over 3,000 independent researchers from 119 countries have come together under Cohere For AI to create Aya. This state-of-the-art model and dataset pushes the boundaries of multilingual AI, covering 101 languages. Language models often overlook a vast part of the world due to language barriers. Aya, however, makes a big stride in inclusive AI by thinking beyond English. MORE

Reka Flash: An Efficient and Capable 21B Multimodal Language Model 
Reka AI delivered a smackdown with their new model, Reka Flash. This model is a "turbo-class" offering, trained from scratch, and optimized to its limits. It's not only efficient and quick, but it also matches the quality of heavyweights like Gemini Pro and GPT-3.5. Although it's not open source, the potential applications are exciting. MORE

ElevenLabs: AI that Transforms Your Voice 
A new tool that allows you to change your voice into another character's with just a click. Not only can you alter the voice, you control the emotion and delivery too. This AI technology will certainly be a game-changer for indie game developers and content creators alike. MORE

NVIDIA: CEO Calls for Sovereign AI for Every Country 
In a recent statement at the World Governments Summit in Dubai, NVIDIA founder and CEO Jensen Huang emphasized that every country should produce its own intelligence. While it's an excellent stance for a company selling the tech, the message rings true. Countries must comprehend and make decisions concerning AI, or others will make those decisions for them. MORE

Humans VS LLMs: The Rise of Legal Language Models 
Imagine legal contract reviewers working at lightning speed and a fraction of the cost. That's what Large Language Models (LLMs) bring to the table. A recent study put human lawyers up against LLMs in evaluating legal documents and found LLMs matched or even exceeded the accuracy of junior lawyers. LLMs also worked between 91.63% to 99.64% faster than their human counterparts. The legal industry stands on the brink of disruption, heralding a new era of efficiency and accessibility in legal services. MORE

RESEARCH: More Agents Is All You Need 
A new study proposes that the performance of Large Language Models (LLMs) improves with the increase in the number of agents. This simple sampling-and-voting method is a fresh approach to enhancing LLMs, and the level of improvement seems to be linked to the task difficulty. This is the first study I've come across that explores the scale laws concerning AI agents. It's an exciting revelation that, similar to LLMs, more computation could be a good initial step. MORE

RESEARCH: Large Language Models Self-Compose Reasoning Structures 
Meet SELF-DISCOVER, a general structure for LLMs to self-identify task-centric reasoning structures to tackle complex reasoning problems that typical prompting methods find challenging. The heart of this framework is a self-discovery process where LLMs select various atomic reasoning modules and compose them into an explicit reasoning structure. It has significantly improved the performance of GPT-4 and PaLM 2's on challenging reasoning benchmarks. It's amazing how we keep finding new ways to improve our interactions with LLMs using prompting frameworks. MORE


Reply

or to participate.