Dec 07, 2023 15:00:00

AI 'AlphaCode 2' specialized in programming based on Google's multimodal AI 'Gemini' has arrived, with performance that ranks in the top 15% of competitive programming

Google released the multimodal AI '

Gemini ' on December 6, 2023 local time. Gemini has improved programming ability compared to the large-scale language model ' PaLM 2 ', and an AI ' AlphaCode 2 ' specialized for competitive programming based on Gemini has also been developed.

Introducing Gemini: Google's most capable AI model yet
https://blog.google/technology/ai/google-gemini-ai/#scalable-efficient

Gemini: Excelling at competitive programming - YouTube


Gemini is said to be able to handle Python, Java, C++, and Go.

In a test that asked students to write about 200 types of Python code, PaLM 2 recorded a 45% correct answer rate, but Gemini's correct answer rate improved to 75%.

Furthermore, when Gemini is instructed to check and modify the code, the correct answer rate jumps to 90%.

Users can instruct Gemini to code via chat format. A demo video released by Google shows how a draft of the code is created simply by giving instructions such as ``Help me create a web application that displays the current location of a running train on Google Maps.'' .

Furthermore, 'AlphaCode 2', an AI based on Gemini and specialized for competitive programming, was also announced.

'

AlphaCode ' announced in 2022 had programming ability that ranked in the top 50% of competitive programming, but AlphaCode 2 has improved its ability to be in the top 15%.

Competitive programming tasks require not only simple coding ability, but also knowledge of mathematics and computer science, and the ability to understand problems. AlphaCode 2 is equipped with these knowledge and abilities, and is said to be able to solve difficult questions with a correct answer rate of 0.2%.

AlphaCode 2 is an AI specialized for competitive programming, but Google is already working on incorporating some of AlphaCode 2's functions into Gemini.

The AlphaCode 2 technical report can be viewed from the link below.

AlphaCode2_Tech_Report.pdf
(PDF file) https://storage.googleapis.com/deepmind-media/AlphaCode2/AlphaCode2_Tech_Report.pdf

◆Forum now open
A forum related to this article has been set up on the GIGAZINE official Discord server . Anyone can write freely, so please feel free to comment! If you do not have a Discord account, please create one by referring to the article explaining how to create an account!

• Discord | 'Do you use AI for coding? What kind of work is it useful for?' | GIGAZINE
https://discord.com/channels/1037961069903216680/1182242848054202448