10 AI Predictions For 2023

1. A major release is coming soon - GPT-4

GPT-4 is set to be released soon and will surely be a game-changer. According to DeepMind researchers, today's language models are larger than they need to be and should have fewer parameters but be trained on larger datasets. GPT-4 is likely to be trained on an unprecedented dataset of up to 10 trillion tokens while still being smaller than its predecessor GPT-3. It may also be a multimodal model, able to work with images, videos, and other data modalities in addition to text.

If GPT-4 is only a text-based model, its performance on language tasks will surely be impressive. It is predicted to have a strong memory, recall information from previous conversations, superior summarization capabilities, and distill large bodies of text into their essential elements. No matter what, GPT-4 is sure to be a major step forward in the development of language models and will be a significant milestone in AI.

2. Data will run out soon to train large language models.

AI researchers are running into a potential roadblock: a shortage of quality language data. What was once a metaphor - data being the "new oil" - is proving true in the language AI world. The most effective way to make large language models (LLMs) more powerful is to train them on more data. But how much more usable data is available? Estimates suggest the world's high-quality text data stock could be as low as 4.6 trillion and as high as 17.2 trillion tokens. DeepMind's Chinchilla model has already used 1.4 trillion tokens. The world's supply of language training data may be exhausted within one order of magnitude.

This looming data shortage has sparked concern among AI researchers and entrepreneurs. Solutions like synthetic data and transcribing spoken content are being considered to tackle the challenge. OpenAI's GPT-4 research will be closely watched to see how they deal with the data shortage. LLM researchers are actively seeking ways to address this issue, which will likely be a focus of attention in 2020.

3. In the near future, the general public will begin using driverless cars on a daily basis.

Autonomous vehicles are no longer a novelty; they are becoming a reality. Cruise and Waymo are offering fully driverless rides to the public in San Francisco, and the services are expected to become available 24/7 very soon. By 2023, robotaxi services will become a popular and convenient way to get around cities. Meanwhile, the rollout of autonomous vehicles will happen on a city-by-city basis. Phoenix, Austin, Las Vegas, and Miami are expected to be the next cities to join San Francisco in offering this service to the public.

4. Venture capital funding will be raised by Midjourney.

Midjourney , the leading text-to-image AI platform, has achieved remarkable success without relying on venture capital funding. With nearly 6 million users and substantial revenue, the company has grown significantly since its launch. Founder/CEO David Holz has spurned all outside funding despite the many VC suitors who have sought to invest in the company.

However, the demands of blistering growth, intensifying competition, and a massive market opportunity could lead Holz to pursue a large funding round in 2023. Holz's experience with his VC investors during his time at Leap Motion, a virtual reality startup that raised nearly $100 million in venture capital, has likely influenced his decision to remain self-funded. With OpenAI raising $1 billion from Microsoft and Stability AI raising $100 million, Midjourney could risk being left behind if it does not pursue outside funding soon.

5. 2023 will be the year when search changes more than ever.

Search is at the core of the modern internet experience, and over the past two decades, it has remained largely unchanged—until now. Thanks to advances in language models, we are on the cusp of a new golden age of search. In 2023, the way we search will look very different.

Conversational search, enabled by natural language processing (NLP) models, will allow users to have a dynamic conversation with an AI agent to find what they are looking for. Hebbia and Glean are two startups leading the charge to transform enterprise search using large language models, making it more powerful and productive. Meanwhile, startups like Twelve Labs are pushing the boundaries of multimodal search, allowing users to query and retrieve information across data modalities—especially video.

The accuracy of the conversational search will be a major challenge in 2023. Consumers won't accept a search application that is accurate only 95% or 99% of the time. But with the continued development of large language models and the emergence of exciting new startups in the search space, 2023 is a year of major changes in how we search.

6. Humanoid robots will attract considerable funding, talent, and attention.

The humanoid Robot has long been a symbol of Hollywood's sci-fi visions of artificial intelligence, from Ex Machina to I, Robot. Now, futuristic technology is becoming a reality.

One of the main reasons for creating humanoid robots is to automate complex tasks in the world - from factories to offices - that humans would otherwise do. Tesla has recently made strides in this field with the launch of their Optimus Robot at their AI Day in September. They plan for Optimus to be even more valuable to Tesla than their car business.

Startup companies are also joining the race to build humanoid robots, such as Agility Robotics, Halodi Robotics, Sanctuary AI, and Collaborative Robotics.

In 2023, we'll see a surge of talent and investment as more companies and startups recognize the potential of humanoid robots. It will be similar to the rise of autonomous vehicles in 2016, and the competition to create the most advanced humanoid robots will be fierce.

7. The idea of “LLMOps” is projected to be a fashionable upgraded version of “MLOps.”

As a new technology platform emerges, so does the need for the tools and infrastructure to support it. MLOps has been one of the most popular categories for startups over the last few years, with companies like Weights & Biases, Tecton, Snorkel, and OctoML raising large sums of capital at high valuations. The latest wave of AI technology is Large Language Models (LLMs), which come with unique workflows, skillsets, and possibilities.?

This has created an opportunity for a new suite of tools and infrastructure to emerging, and the term “LLMOps” is set to become a popular way to refer to the “picks and shovels” of this new technology platform. Examples of tools and offerings could include fine-tuning foundation models, no-code LLM deployment, GPU access and optimization, prompt experimentation, prompt chaining, and data synthesis and augmentation.

8. The number of research projects that develop upon or reference AlphaFold will increase exponentially.

DeepMind's AlphaFold platform was a game-changer for the scientific world when it was first unveiled late in 2020, allowing researchers to accurately predict the three-dimensional shape of a protein based on its one-dimensional amino acid sequence. This feat had eluded humans for decades. This discovery has the potential to unlock countless new possibilities in biology and human health, from developing life-saving treatments to improving agriculture and fighting disease.

In July 2021, DeepMind opened AlphaFold to the public, releasing a database of 350,000 three-dimensional protein structures and a further 200 million proteins a few months later. Over 500,000 researchers in 190 countries have accessed the platform in the short time since.

By 2023, it is expected that the number of research projects built on or citing AlphaFold will rise sharply. Scientists will utilize this new and vast repository of foundational biological knowledge to develop groundbreaking applications and medical solutions across a range of disciplines.

9. DeepMind, Google Brain, and/or OpenAI will endeavor to construct a basic model for robotics.

AI research organizations like DeepMind, Google Brain, and OpenAI are set to explore building a foundation model for robotics in 2023. First introduced by Stanford researchers, this concept is a massive AI model trained on data that can be used across multiple activities. These models have already shown great potential but have only been used in the digital realm.

Robotics research has yet to be impacted by this new model paradigm. But that could all change in 2023. The leading AI research organizations are expected to lead the charge in developing a foundation model for robotics. This model is trained on data from different sensor modalities and can be used to develop a general understanding of physics and real-world objects. This model would also be fine-tuned for specific hardware and activities, making it more applicable in the real world.

10. Billions of dollars will be invested in chip manufacturing in the U.S. to prepare for Taiwan.

The U.S. government is making contingency plans to reduce its reliance on Taiwan for advanced semiconductors essential for powering Artificial Intelligence. Tensions between China and Taiwan have escalated, raising the risk of China invading and reabsorbing Taiwan. To mitigate this, the government has passed the CHIPS and Science Act, providing legislative and financial resources to incentivize and subsidize the construction of advanced chip manufacturing facilities in the U.S.

Two weeks ago, TSMC announced it would invest $40 billion to build two new chip manufacturing plants in Arizona, with production beginning by 2026. Expect more commitments in 2023 as the U.S. works to ensure a secure global supply of critical AI hardware.