Meta Platforms Goes To AI War: Launches New AI Coding Software To Compete With OpenAI

Meta Platforms Inc META, looking to gain prominence in the competitive arena of generative AI, has embarked on a spree of open-sourcing its tools, intensifying rivalry with Microsoft Corp MSFT-backed OpenAI and Alphabet Inc’s GOOG GOOGL Google.

With their latest endeavors in AI models that can generate text, conduct translations and craft audio, the tech titan recently announced the open-sourcing of Code Llama, a cutting-edge machine learning system proficient in developing and elucidating code in plain English, TechCrunch reported.

Code Llama's Capabilities: Having parallels with tools such as GitHub Copilot and Amazon.com, Inc. AMZN CodeWhisperer, as well as with other AI-driven code generators such as StarCoder, StableCode and PolyCoder, Code Llama's prowess lies in completing code and troubleshooting in various languages, including but not limited to Python, C++ and Java.

Meta's Stance on Open Source: In a shared communication with TechCrunch, Meta said, "At Meta, we believe that AI models, especially large ones tailored for coding, thrive with an open approach, both for innovation and safety."

Code Llama's Foundation: Built on the backbone of the previously open-sourced Llama 2 model by Meta, Code Llama comes in various versions, some of which are tailor-made for Python or fine-tuned to interpret instructions like "Write me a function to present the Fibonacci sequence."

The initial Llama 2 was able to produce code, but its quality didn't match dedicated models such as Copilot.

Training Insights: For Code Llama's training, Meta employed a similar dataset as they did with Llama 2, comprising diverse public web sources. The company granted Code Llama more extensive learning on the connection between code and language than its predecessor, Llama 2.

The Popularity and Risks of Generative AI: Code-generating utilities are witnessing soaring demand from both programmers and non-developers. Yet, alongside the benefits, generative AI has its pitfalls. Concerns range from the inadvertent introduction of security vulnerabilities to intellectual property dilemmas and malicious entities' potential misuse of open-source code-generating tools.

Code Llama's Safety Measures: Meta conducted internal testing on Code Llama with a team of 25. Despite not being subjected to a comprehensive third-party review, the model exhibited errors that could raise eyebrows. 

META Price Action: Meta Platforms shares are trading down 2.25% at $287.73 on the last check Thursday.

Read Next: Apple Losing Its Bite With Big Whales? Why Institutional Investors Are Picking Different Tech Orchards

Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.