chatGPT isn’t really all that "smart", you know?


The collapse of SVB is the topic on everyone's mind. It has sent the tech world into a tailspin. I hope you haven't been significantly affected by it. I aim to offer some respite from the SVB news, with the freshest AI developments in anticipation of the GPT-4 announcement, which is allegedly happening this week!

What do we have for you today?

  • 🧠 chatGPT isn’t all that “smart”. Noam Chomsky argues chatGPT is just a very smart “auto-complete” platform and that the future of artificial intelligence needs to start resembling more how humans think

  • 👀 GPT4 is coming. Are you ready to face it? Multimodal AI, faster speed, maybe even the ability to generate videos? All eyes peeled for the announcement

  • 👨‍🏫 AI and Education. You know we’re passionate about the ability of AI to supercharge learning. Last week we’ve been featured by Sifted as one of the companies to watch in this space. I venture into a call to arms encouraging fellow builders to pick up their shovels and start building

chatGPT isn’t all the smart, you know?


It looks like Noam Chomsky, Ian Roberts, and Jeffrey Watumull aren't big fans of language models like ChatGPT. They're basically saying that these programs are fundamentally flawed and will never lead to the dawn of artificial general intelligence, which is when machines become smarter than humans.

Apparently, models like chatGPT are good at certain things, like computer programming, but they're not great at reasoning and using language like humans do. The authors argue that human language ability comes from an innate, genetically installed operating system that's completely different from what machine learning programs use. So basically, these programs are stuck in a pre-human phase of cognitive evolution. Ouch.

Their deepest flaw is the absence of the most critical capacity of any intelligence: to say not only what is the case, what was the case and what will be the case — that’s description and prediction — but also what is not the case and what could and could not be the case. Those are the ingredients of explanation, the mark of true intelligence.

The current models also don't use causal mechanisms or physical laws like humans do. Instead, they just use brute correlations among data points to come up with the most probable response.

The authors believe that ChatGPT and other language models won't ever reach true intelligence because they lack the ability to be creative and come up with explanations.

GPT-4 is coming: here’s what we know


GPT-4 is allegedly being released this week. Not many things are confirmed yet, but here’s what we know:

Last week, Microsoft Germany CTO Andreas Braun announced that the company will be introducing GPT-4 this week

  • What exactly did he say: We will introduce GPT-4 next week, where we have multimodal models that will offer completely different possibilities – for example, videos.

How is this being interpreted?

  • GPT-4 will likely have multimodal capabilities, which means users will be able to engage through multiple modes such as text, images, videos, and sounds. The previous generation was only capable of producing texts

  • As a multimodal language, users may be able to manage the input and output of data in different languages

  • Some believe the new model will be able to generate videos from prompts

What else can we expect?

  • The next-generation language model is expected to give out answers much more quickly and in a more human-like manner

AI and Education: calling the bold and the brave


Since chatGPT has been released, everyone has been talking about its impact on education. While some schools have banned access to it, others are accepting AI tools to become part of the educational system. In a recent study on how AI will affect industries and occupations, it surfaced that the most common profession “exposed” to this new wave of technology is that of the teacher.

But in the startup world, there are very few players focusing on building innovative AI-powered solutions. In an intelligence report by Sifted on generative ai startups, broadn, the company we’re building was featured as one of the top companies to watch. From 110 startups, there’s a total of only 4 included in the Education and Training section.

Why? Problems in education are complex and historically, edtech has been a very unattractive space for investments, so very few companies dare to build in this space. Also, the technical limitations of LLMs around truthfulness require smart solutions to circumvent them.

There is no doubt AI will change education, from lifelong learning to learning in the workplace and traditional education. For this to happen, we need more builders in this space, investors that are taking a long-term view, and great products that reimagine the way we learn. We couldn’t be more excited to continue building!

Gen AI Deals that make your eyes (and mouth) water 💰

What caught our eye? 👀Anthropic’s AI chatbot, Claude, is in closed beta mode, but in a paper detailing its goals, it is expected to combat harmful prompts by explaining why they are dangerous or misguided. This follows previous investment from Google which put down $400m for a 10% stake.

What caught our eye? 👀This financing will enable the company to develop its AI and drug discovery capabilities, in addition to launching Iktos Robotics; a unique, end-to-end, drug discovery platform which combines artificial intelligence and automation of chemical synthesis to significantly accelerate drug discovery timelines.

New developments to spam your #random Slack channel 💬

Things to learn when you need a raise

Want access to the Generative AI investors database? Share and subscribe to AI In the Middle (below) to receive it! ⬇️

Before you go

The power of Midjourney v5

And scene. That’s all for today! We’re going to one post per week for a while - hope you won’t miss us too much! Thank you for always reading, sharing, and subscribing. Want to share something with us? Slide in our DMs, they’re open.

— Calin Drimbau (@calindrimbau)