The Robo-Coders: These artificial intelligences convert text into program logic

Reading texts and have been able to write AI models for a long time. Two new AIs have now been launched that can even convert texts into code.

The highly praised GPT-3 AI model from OpenAI is facing competition. Some of it comes from our own company. With Codex, OpenAI brings an improved version of its own GPT-3 model. Codex has been in a private beta phase since Tuesday and is available for testing to a limited number of developers. Microsoft had previously integrated GPT-3 into its low-code platform Power Apps. Incidentally, an early version of Codex powers the Github copilot.

As OpenAI says, Codex was both fed huge amounts of text and trained on billions of lines of publicly available computer code. This link is intended to enable Codex to write program-logical processes from natural-language instructions in English.

Greg Brockman, co-founder and OpenAI’s chief technology officer, wants to use Codex to remove the barriers to entry for normal people in programming and to create a future “in which the entire world can be programmed”. Brockman is not only concerned with creating code by voice, but much more generally about the fact that people use voice to trigger actions that previously made other tools necessary. Brockman dreams of a future in which you can “talk to your computer and get it to do what you ask it to do, in a competent and reliable manner”.

The second new AI on the market comes from the Israeli startup AI21 Labs. On Wednesday they released a whole set of speech-generating AI models and named it “Jurassic”. The largest in the “Jurassic Jumbo-1” series is, with 178 billion parameters, i.e. the values ​​that a neural network tries to optimize during training, the largest in the world and thus even larger than OpenAIs GPT-3.

The AI ​​models are to be offered together under the name AI21 Studio. This will be a tool kit that developers can use to do natural language programming. AI21 Labs even gives out the test access free of charge. Only the commercial use is then billed for a fee.

Platform boss Dan Padnos wants to gain market share quickly with this. Coders should register now “to quickly create all text-based applications they can imagine,” said Padnos when introducing the product.

Who now after asks the differences, finds rather marginal ones. Just like OpenAI, AI21 Labs has also unleashed its algorithms on huge amounts of text in order to learn to read and write texts. However, AI21 Labs is putting a stronger focus on providing a customizable, easy-to-use interface that highlights features like writing summaries. The AI21 developers even have an app to dejargonize language.

Natural Language Processing (NLP) is considered one the most exciting areas of AI research. It is seen as having a major impact on how we will communicate and work in the years to come. That can also be expressed in numbers. It is estimated that the global market for NLP will grow from around $ 21 billion in 2021 to over $ 127 billion in 2028. Some even consider this estimate to be too conservative.

As with all AI models, there are complex problems in NLP research . Most of them are of a technical nature, but ethical difficulties also need to be addressed openly. Tools like GPT-3 could be used for large-scale disinformation campaigns.

Any kind of influence – especially via social media – could benefit from an AI that tirelessly produces texts. Both OpenAI and AI21 want to put a stop to this by promising that they will prevent unauthorized access to their systems.

You might also be interested in