All articles tagged with #code generation
Companies increasingly deploy AI agents to generate code and automate tasks, but token-based costs can dwarf salaries, with teams racking up six-figure monthly bills and at times consuming entire AI budgets; while leaders tout incentives and pricing strategies, questions linger on ROI and whether forcing workers to rely on AI actually boosts productivity.
As firms rush to leverage AI, outages and flawed outputs—like Amazon's AI-driven coding mishap—underscore the dangers of speed without discipline. Companies are imposing guardrails and audits to balance rapid experimentation with risk, while many workers rely on AI outputs without thorough checks. Experts advise pairing AI with human reviews and defining risk tolerances to turn missteps into learning opportunities and strengthen controls.
Anthropic's Claude Code, launched in May, is gaining traction by letting people with no coding experience generate software from prompts. The tool is part of a broader vibecoding trend alongside rivals like Base44 and Cursor, with subscriptions ranging from $20 to $200 per month as more users discover its capabilities.
Apple released an open-source AI model called DiffuCode-7B-cpGRPO that uses a diffusion-based approach for code generation, allowing faster and more globally coherent code output by generating out of order and refining multiple chunks simultaneously, building on Alibaba's foundation model and recent research in diffusion architectures.
StarCoder 2, an open source code generator developed by Hugging Face in collaboration with ServiceNow and Nvidia, has been released in three variants, offering improved performance and the ability to run on most modern consumer GPUs. The AI model, trained on a larger and more diverse dataset, aims to make coding more efficient for developers. However, concerns about security vulnerabilities and licensing restrictions have been raised, despite claims of improved efficiency and ethical training practices.
Meta AI has released Code Llama 70B, an open-source AI model for code generation that can write code in various programming languages from natural language prompts or existing code snippets. With a larger context window and fine-tuned variants for understanding natural language instructions and generating Python code, Code Llama 70B sets a new standard for accuracy and adaptability in code generation. Available for free download, it is expected to have a significant impact on the software development industry and lower the barrier to entry for learning coding.
Stability AI has launched StableCode, an open large language model (LLM) designed to generate programming language code. StableCode is available at three different levels and initially supports Python, Go, Java, JavaScript, C, markdown, and C++. The model is built on the BigCode project and utilizes a longer context window of 16,000 tokens, allowing for more specialized and complex code generation prompts. StableCode also employs rotary position embedding (RoPE) instead of the ALiBi approach to position outputs in a transformer model. Stability AI aims to work with the developer community to explore the potential of this new generative AI model for coding.
Google's Bard AI has been updated with better advanced reasoning and math abilities, allowing it to generate background code to improve correctness. The update boosted computation-based word and math problems’ accuracy by 30 percent. Bard can provide information such as the prime factors of numbers in the millions and the growth rate of savings. Google has also announced a function that allows data produced by Bard to be editable and manipulated in Sheets.