DeepMind has made significant strides in the development of a qualitative AI coding program, the company reveals in an announcement. Currently, the program can even compete with programmers in coding competitions already.
DeepMind is a company that has been specializing in AI since its founding in 2010. Currently, the company claims to be so far along with developments that AI is a formidable competitor to real programmers in specific coding tasks.
The new AI program is named AlphaCode, DeepMind says in the announcement. The program the researchers tested out with coding tasks used in competitions for programmers. The results were very good, as the program’s estimated ranking scored in the top 54 percent.
Small breakthrough
The results sound great, but the program does not currently constitute a suitable replacement for a human coder. Indeed, the coding tasks in the test differ from day-to-day problems programmers face.
Indeed, programmers face challenges every day that require different skills. The challenges of Codeforces, the platform the researchers consulted for the tests, are much simpler.
“I can safely say that AlphaCode’s results exceeded my expectations,” said Codeforces founder Mike Mirzayanov, in DeepMind’s post. “I was skeptical because even simple coding tasks often require not only implementing the algorithm, but also (and this is the hardest part) figuring it out. AlphaCode managed to perform at the level of a promising new competitor.”
Bigger story
For AI coding systems, there have been more breakthroughs in recent years. At the level of a human programmer, no system has been hit yet, but every step counts.
In breakthroughs, it helps if AI models are made open to a wider audience. For that reason, OpenAI made the GPT-3 language model publicly available as recently as late last year. Furthermore, more research on AI models is also simply happening right now.