2023 artificial intelligence development trend TOP5

In 2022, many breakthroughs were made in the AI/ML field. Large -scale technology companies such as Google, Meta and Microsoft have made significant improvements in new innovations from quantum computing to generating artificial intelligence.

For example, some of the biggest breakthroughs include the HypertreeProofsearch (HTPS) of Meta, which solves the problem of international mathematics Olympic problems; the ESMFOLD’s ESMFOLD’s ESMFOLD’s ESMFOLD of the DeepMind is used for protein folding prediction; All Genome Research (GWAS) and so on.

Next, let’s take a look at some predictions of 2023.

ChatGPT swept the Internet with its excellent dialogue. It is based on OpenAI GPT-3. The GPT-3 has 176 billion parameters, which depends on a larger model size. Although there are other LLM parameters of the GPT-3 twice, three times, or even ten times, the number of parameters of some models of DeepMind or Meta (also known as small language model (SLM)) have exceeded GPT-3 in logic reasoning And many tasks such as prediction.

In addition to reducing the size of the model, it is expected that there is a larger model, such as GPT-4, with about 1 million billion parameters. Because the current largest model is Google Switch Transformer model with 1.6 trillion parameters, jumping will be huge.

However, in order to obtain greater delay and predictability, you can see the existing models fine -tuned to serve specific purposes in the next few years. Recently, OPENAI fine-tuned GPT-3 with Da Vinci update.

Trend 1: Genetic AI needs to be explained AI
The generation of text to images is the trend of breaking the rankings in 2022. Dall-E, Stable Diffusion, and Midjourney are among the best in the enthusiasts who want to test AI to generate art. The dialogue quickly shifted from text to image to text to video, then to text to anything, and created multiple models. These models can also generate 3D models.

With the expansion of the language model and the improvement of the communication model, the trend of text to anything is expected to rise higher. Public available datasets make the generation AI model more scalability.

These datasets introduce parts that can explain artificial intelligence, where the properties of each image of these generating models are very important.

Trend 2: Fastsaas Competition begins
The company that caught up with the trend of artificial intelligence has begun to provide it as a cloud service. With the open available use of LLM and GPT-3 and Dall-E, companies are becoming more and more likely to provide them as services, which has spawned FastSaas.

Recently, the Shutterstock plan integrated Dall -E 2 to its platform. Microsoft VS Code added Copilot as extensions. Tiktok announced the launch of the application text to the image AI generator, and Canva launched AI -generating functions on its platform.

Trend II
This is the trend of building supercomputers to depend on generating tasks and provide services to the company. With these increasing data sets and generation models, the demand for supercomputers is rising, and it is expected to rise further. With the competition for Fastsaas, the demand for better and high -performance computing is the next thing.

NVIDIA and Microsoft recently created Quantum-2, a cloud native super computing platform. In October, Tesla announced that its DOJO super computer completely used Tesla’s chips to be constructed from scratch. Soon, it seems to be able to provide access to corporate customers. In addition, CEREBRAS also launched Andromeda, which is an AI super computer with 13.5 million cores, which can provide AI computing capabilities with more than 1 Exaflop. Recently, Jasper cooperated with Cerebras to achieve better performance.

Trend 4: Beyond 3nm chip
As Moore’s Law Foundation predicted, the processing capacity increased with the decrease of chip size. Therefore, if the supercompitious model is running a large model, smaller chips are required, and we have seen that the chip is getting smaller and smaller.

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

Trend 5: The fusion of quantum and traditional computing
As companies such as NVIDIA, Google, and Microsoft provide hardware services to the cloud, more innovation in the field of quantum computing will inevitably happen. This will allow small technology companies to train, test and build the AI/ML model without heavy hardware.

https://oaicon.com/index.php/2023/03/08/2023-artificial-intelligence-development-trend-top5/

The rise of quantum computing in the next few years should definitely be included in developers, because its use will increase in many other fields, such as medical care and financial services.

In the recent announcement, a quantum computer is connected to Europe’s fastest supercomputers to combine traditional computers and quantum computers to solve problems faster. Similarly, NVIDIA also released Qoda-referred to as Quantum-OPTIMISED Device Architecture, which is the first platform for mixed quantum classic computers.

IBM recently announced their quantum hardware and software at its 2022 quantum summit, and outlined the pioneering vision of the super -computing of quantum -centric super -centered quantum processors using 433 quantum positions. At the Global Artificial Intelligence Summit, IBM announced that they will show a 1,000 quantum bit -bits next year, which will become a subverter for further innovation in various fields.