A few themes for 2024

The obligatory predictions piece

Dec 30, 2023

This is a weekly newsletter about the business of the technology industry. To receive Tanay’s Newsletter in your inbox, subscribe here for free:

Hi friends,

Hope everyone is having a great holiday break. As the year wraps up, I couldn’t help but cover some themes I’m paying attention to in 2024 and predictions for the year around AI, startups, and technology.

An AI-based soothsayer, depicted as a futuristic, robot-like figure, sitting at a table with a crystal ball in front of them. The setting is a blend of traditional and futuristic elements, with glowing digital runes and symbols floating around the crystal ball, indicating its AI capabilities. The soothsayer is gazing into the crystal ball, which is illuminating their face with a soft, mystical light. The background is dimly lit, creating a mystical and enigmatic atmosphere. The composition is in landscape mode to capture a wider view of the scene, size: "1792x1024"

Let’s get into it.

I. Open-Source Models Catch Up to GPT-4

We’ve seen an explosion in open-source models in 2023 with models from Meta (LLama), Mistal (Mixtral) and others. However, OpenAI’s models still remain dominant when it comes to usage in companies given their quality, OpenAI’s brand and ease of use.

Source: A Survey of Large Language Models

Infrastructure for fine-tuning and hosting open-source models has drastically improved in the last six months, offering a viabe alternative for businesses seeking more control or the use of cheaper, lower-latency fine-tuned models.

Open-source models such as LLAMA and Mixtral have still lagged the state of the art closed source model in GPT-4, and are more comparable to GPT 3.5 in performance. But I expect that we will see at least one open-source model approach GPT-4 performance this year, with open-source models gaining a larger share of usage and spending as infrastructure improves and enterprises start to productionize AI workloads, placing more emphasis on cost and control.

II. Moving beyond just chat interfaces in AI

The median use of Generative AI today seems to be a version of ChatGPT on some private company dataset to help answers questions or write drafts of some documents. While very valuable by itself, the chat interface which everyone is rushing to implement in their products today seems to only be scratching the surface of what is possible with these models.

I expect to see more innovation around the actual interfaces with which AI is made available in products to customers, with more action-oriented interfaces and interfaces tied into workflows rather than simple chatbots.

Examples of what this may look like include:

voice-first interfaces
interfaces deeply integrated into workflows of media generation / editing such as tldraw and Adobe’s firefly features in Photoshop
interfaces that allow taking actions
generative interfaces where the interface is generated depending on the question asked and the information being displayed.

Image of bringing Genereative AI to Creative Cloud with Adobe Firefly. — GenAI integrated into Adobe’s editing workflow. Source

III. Explosion of Multimodal AI

I expect 2024 to be the year for multimodal AI. We’ve started to see glimpses of the possibilities of text-to-video and text-to-speech models through Gemini, Pika, Runway, Eleven Labs and others, but 2024 is when I think we’ll start to see what’s truly possible as these models improve and people start building applications with them as they become available via APIs.

I think we’ll see a lot of progress amongst all dimensions including multimodal embedding models, multimodal search, voice-first applications, generative video and video search.

As some of the individual components get better and companies start thinking “multimodally”, we’ll also start to see applications such as live video based AI conversations across use cases of entertainment, sales, etc.

IV. Startup Shutdowns and Acquihires grow

One of the more sobering predictions — In 2023, we saw the number of startups shutting down increase. Unfortunately, I think we’ll see the same again in 2024, as the hangover from 2021/22 continues to fully unwind.

There seems to have been a clear demarcation between AI and non AI startups in terms of attractiveness for investment (as well as just actual demand from customers today), and many in the latter camp had managed to push their runway out to 2024 but are still in search of product market fit.

As the time runs out for these companies, while some will find market pull, others will not, and be faced with the decision of shutting down, or taking an acquihire type offer if one materializes.

The one positive on that front is that I think there will be buyers for strong, talented teams. With big tech companies realising that large M&A is unlikely, and with them having more right-sized employee bases now, more of them may look to acquihire great talent, especially as they push to take advantage of the tailwinds of AI and incorporate it into their products.

V. The IPO Window Creaks Open

In 2023, we saw tech IPOs pretty much at a complete standstill until this past quarter where we saw some signs of life with Instacart, Klaviyo and ARM going public.

US tech IPOs total proceeds in first three quarters

Given their performance has been relatively lukewarm (ARM is the only one above its IPO price) and the current state of the markets, the window isn’t really open right now per se.

But I expect things to be a bit brighter in 2024 with rate cuts on the horizon, the NASDAQ booming, and in general the market for software improving with AI. While 2024 is unlikely to be an “open IPO window” with a flurry of activity, I think we’ll see more companies than this year go public.

I expect it to be driven by a small subset of companies, most of which are 12+ years old, close to profitability, and at a large revenue scale ($750M+), perhaps in the latter part of the year, to set up more companies following and going public in 2025.

As a reminder, there are a lot of $5B+ and $10B+ companies in private markets, and while many have a lot of growing to do to reach their marks (and many smaller ones will never hit their valuations and some will fail completely as we saw with companies such as Convoy this year), many are at the stage where their next financing will likely need to be via an IPO, although they aren’t “IPO ready” today.