TOP5 Artificial Intelligence Development Trends in 2023

2022 saw many groundbreaking breakthroughs in the field of AI/ML. Big tech companies like Google, Meta, and Microsoft are making major strides in new innovations from quantum computing to generative artificial intelligence.

For example, some of the biggest breakthroughs include Meta’s HyperTreeProofSearch (HTPS) for solving International Mathematical Olympiad problems; DeepMind’s Alpha Fold and Meta AI’s ESMFold for protein folding prediction; Google’s DeepNull to simulate the correlation between phenotypes The relationship between variable effects and improving genome-wide association studies (GWAS), etc.

Next, let’s look at some predictions for 2023.

ChatGPT is popular on the Internet with its excellent conversation capabilities. It is built on OpenAI’s GPT-3, which has 176 billion parameters and relies on larger model sizes. Although there are other LLMs with two, three or even ten times the parameters of GPT-3, some models of DeepMind or Meta (also known as small language models (SLM)) have more parameters than GPT-3 in logical reasoning. and prediction on multiple tasks.

In addition to reducing the size of the model, it is expected that a larger model, such as GPT-4, will have approximately 100 trillion parameters. Since the largest model currently is the Google Switch Transformer model with 1.6 trillion parameters, the jump will be huge.

However, to achieve greater latency and predictability, the next few years could see existing models being fine-tuned to serve specific purposes. Recently, OpenAI fine-tuned GPT-3 using the DaVinci update.

Trend 1: Generative AI requires explainable AI

Text-to-image generation is the trend that will break the charts in 2022. Models like DALL-E, Stable Diffusion, and Midjourney top the list among enthusiasts who want to experiment with AI-generated art. Conversations quickly moved from text to images to text to video to text to anything, and multiple models were created that could also generate 3D models.

As language models expand and propagation models improve, the text-to-anything trend is expected to rise even higher. Publicly available datasets make generative AI models more scalable.

These datasets introduce a section on explainable artificial intelligence, where the properties of each image used to train these generative models become critical.

Trend 2: The FastSaaS race begins

Companies that have caught up with the trend of generating artificial intelligence have begun to provide it as a cloud service. As LLM and generative models such as GPT-3 and DALL-E became publicly available, it became increasingly easier for enterprises to offer them as a service, which gave rise to FastSaaS.

Recently, Shutterstock plans to integrate DALL-E 2 into its platform, Microsoft VS Code added Copilot as an extension, TikTok announced an in-app text-to-image AI generator, and Canva launched AI-on-its-platform Generate function.

Trend Three: Reliance on Supercomputers

This is the trend of building supercomputers to rely on generating tasks and providing services to companies. With these ever-increasing data sets and generative models, the demand for supercomputers is rising and is expected to rise further. With the competition for FastSaaS, the need for better and high-performance computing is the next thing.

NVIDIA and Microsoft recently collaborated to create Quantum-2, a cloud-native supercomputing platform. In October, Tesla announced that its Dojo supercomputer was built entirely from scratch using chips developed by Tesla. Soon, it looks like it could provide access to enterprise customers. Additionally, Cerebras launched Andromeda, a 13.5 million-core AI supercomputer that delivers over 1 exaflop of AI computing power. Recently, Jasper partnered with Cerebras to achieve better performance.

Trend Four: Beyond 3nm Chips

As predicted by Moore’s Law, processing power increases as chip size decreases. So for supercomputers to run large models, they need smaller chips, and we're already seeing chips getting smaller.

In recent years, the chip industry has been pushing for miniaturization, with manufacturers constantly looking for ways to make chips smaller and more compact. For example, for the M2 chip and A16, Apple uses 5nm and 4nm chips respectively. It is expected that TSMC will develop 3nm chips in 2023, which will improve the efficiency and performance of AI/ML algorithm development.

Trend Five: Integration of Quantum and Traditional Computing

As companies such as NVIDIA, Google, and Microsoft provide hardware services to the cloud, more innovations in the field of quantum computing are bound to occur. This will allow small tech companies to train, test and build AI/ML models without the need for heavy hardware.

The rise of quantum computing in the coming years should definitely be taken into account by developers as its use will increase in many other areas such as healthcare, financial services, etc.

In a recent announcement, a quantum computer was connected to Europe's fastest supercomputer to combine classical and quantum computers to solve problems faster. Similarly, Nvidia has also released QODA - Quantum-Optimised Device Architecture for short, which is the first platform for hybrid quantum classical computers.

IBM recently announced their quantum hardware and software at its annual Quantum Summit 2022, outlining a groundbreaking vision for quantum-centric supercomputing using a 433-qubit (qubit) processor. At the Global Artificial Intelligence Summit, IBM announced that next year they will demonstrate a 1,000-qubit system that will become a disruptor for further innovation in various fields.

The above is the detailed content of TOP5 Artificial Intelligence Development Trends in 2023. For more information, please follow other related articles on the PHP Chinese website!