2022 AI Wrapped: the good, the bad, the funny

The year when artificial intelligence became visible

Hi everyone! It's Calin and I'm back to wrap up the year with a special edition.Artificial intelligence has been around for years, powering many of today's products. Until recently, however, AI was a "shadow operator" - it was difficult to tell if a product or service was powered by AI or not.

2022 will be remembered as a landmark year for technology. For the first time, AI became visible. We began interacting with AI, and it produced results of a quality that we had previously thought only humans could achieve.In this edition, I cover some of the most important events of the year and explain why I think they are noteworthy.

I’m ready to be teased

This year, we had the opportunity to witness AI creating art and even winning competitions! Not everyone was happy about it. We observed an AI model convincing an engineer it was sentient, sparking conversations about unfriendly AGIs (Terminator memes have resurfaced).

We saw two generative AI companies reach unicorn status, and our new interactions with AI made previous interactions with Siri and Alexa seem like conversations with cavemen.

Finally, if you managed to escape the frenzy of creating magical, mystical, and sexy avatars of yourself, congratulations. I caved.

It started with a koala dunking a basketball

It was April 6th, when OpenAI announced DALL·E 2, the smarter, younger brother of DALL-E. This model was significantly better than its predecessor, which was released in January 2021 and introduced avocado chairs to the world.

Let’s pause for a second to take stock of what was happening. I remember having to explain this technology to my tennis coach (who’s renounced technology), and it wasn’t easy. Basically, it’s technology that generates images from a text prompt, I told him. You tell it what to draw and it draws exactly that (well, more or less).

Twitter was exploding. The Demagorg from Stranger Things had its moment, Nosferatu was on RuPaul’s Drag Race and Gollum was devouring… watermelons. Humans really are creative creatures.

Initially in a closed trial, DALL·E 2 was released to everyone in late September. In the meantime, contenders started making waves.

The koala is released into the wild and it starts stripping

On August 22, Stability AI and CompVis released Stable Diffusion 1.4, an image synthesis model similar to OpenAI's DALL-E 2. But while DALL·E was released in a closed beta with heavy restrictions, Stable Diffusion was released as an open source project. Suddenly, everyone could build applications on top of it.

Not everyone was happy about the open source release. Critics were concerned about misuse. With the code being out in the wild, developers could circumvent protection against propaganda content, pornography, deep fakes or violent imagery. The Stable Diffusion license officially forbids many of these use cases, but can humans really be trusted?

AI-Generated porn🌽 was born in August 2022. It started with Unstable Diffusion, a community of people that attempted to use Stability to create pornographic content. When the results were unsatisfactory, they added more training data to improve them.

Today, Unstable Diffusion is a growing community (100,000+ users) and despite their recent Kickstarter campaign being shut down, they continue to find ways to finance their efforts. Matthew covered the topic in more detail in our previous edition.

There’s a chance people will be playing with… prompts next year.

AI Generated content wins a competition and artists flip

In early August, Colorado resident Jason Allen entered three AI-generated images into the Colorado State Fair fine arts competition. Late in the month, he announced that one piece, Théâtre d'Opéra Spatial, had won the top prize in the Digital Arts/Digitally Manipulated Photography category. The news of the victory sparked a fierce debate among people about the nature of art and what it means to be an artist.

The incident reflects a wider cultural debate surrounding AI-generated art. In the red corner we have progressive technologists who argue we need to accept this as an inevitable and positive technological step. In the blue corner, we have artists arguing their entire livelihood is at stake.

The debate continues and is unlikely to be resolved soon.

Personally, I believe we are witnessing a restructuring of the creative value chain. Creativity will be less focused on downstream output production and more on upstream thought and idea generation. Ethical concerns, however, are likely to become more pressing in the coming year.

It's either sentient or giving off a genuine human vibe

In early July, the Washington Post reported that Google engineer Blake Lemoine had been placed on paid leave due to his belief that Google's LaMDA (Language Model for Dialogue Applications) was sentient and deserved rights equal to those of a human.

Lemoine started talking to LaMDA about religion and philosophy and thought he could detect real intelligence in the text. "I can tell when I'm talking to a person," Lemoine told the Post. "It doesn't matter if they have a physical brain or a billion lines of code. I talk to them and listen to what they have to say. That's how I decide who is and isn't a person.”

Google replied that LaMDA was only telling Lemoine what he wanted to hear and that LaMDA was not, in fact, sentient. Grateful you clarified that for all of us, Google.

Today's powerful neural networks generate impressive results that appear to mimic human speech and creativity, thanks to all the data they’ve been trained on. But these outputs are based on pattern recognition, not wit or candor.

Most scientists are convinced that humans cannot create sentient AI, and many do not believe that ‘artificial general intelligence’ (AGI) is just around the corner.

But some believe AGI isn’t that far away. Oculus’s former CTO, John Carmack left a legless Zuckerberg behind and moved his focus away from VR to start a company focusing on AGI, raising $20m from Sequoia and others.

I hope Arnold has another Terminator movie in him; we might need it.

We invest in the creator economy, web3, metaverse… GenAI

On Sep 19th, Sequoia published a blog article under the title ‘Generative AI: A Creative New World’. It sent a signal that Generative AI is a major disruptive force in tech, with the potential to create new markets.

Soon after, Sonya from Sequoia published the Generative AI map, which listed the various players in the space. My startup, broadn, was featured on this list for our work in the text and audio space. AI wrote this last sentence; I wouldn't dare to self-promote.

Stability AI became a unicorn, and the next day Jasper AI followed suit. GenAI was generating unicorns.

Investment sentiment for this space remains positive, but there are still questions about moats, defensibility, technical limitations of LLMs, and ethical concerns. The conversation has shifted, however, from "whether" to "how" - no longer a question of if, but of how.

ChatGPT: generate the future

On the 30th of November, OpenAI launches ChatGPT, a sister model of InstructGPT with a conversational chatbot interface. No one, including OpenAI anticipated what came next.

ChatGPT became the quickest consumer tool to reach 1 million users and soon the world was putting it to test to challenge search, write and debug code or generate essays. You can use it to generate lyrics, write covers letters and even help you negotiate bills!

Whilst the model that powers it, GPT 3.5, is more powerful than the previous GPT 3 model, it’s really the conversational interface that got the world excited.

As much as I’m personally excited about it, I’m also deeply concerned about its limitations. GPT3, despite its 175 billion parameters, is only 60% truthful. It can be worrying when a machine offers inaccurate information four out of ten times, especially when it is confident in its response.

2023: High entropy in Generative AI

Entropy can be seen as the amount of disorder in a system. For example, a messy room has high entropy and requires energy to be transformed into a clean one.

Generative AI is an exciting, but a messy space. Whilst the technology is bound to continue to improve (GPT4 whaaat?), we’ll need to put a lot of energy to create order in this space. The effects of this technology have yet to truly impact the world, but big changes are on the horizon.

Old jobs will vanish, and new ones will emerge. Cultural norms will be disrupted, and entire industries will be revolutionized. The effects will be far-reaching, and new tech giants will emerge.

I’m personally excited by the impact generative AI can have on education and at broadn, we’ve got something exciting baking in the oven.

But I’m also excited to keep watching and showcasing what’s happening across the entire space. We’ve been witnessing an incredible transformation this year and I can’t wait to see what 2023 will bring to the table.

Wishing you a peaceful end of the year and leaving you with some ChatGPT New Year wishes.

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

Before you go

Rest assured, your job is safe - for now.

Signing off for the year.

— Calin Drimbau (@calindrimbau)