At the latest TC39 meeting in February, three JavaScript proposals reached Stage 4, signaling their readiness for inclusion in the ECMAScript standard—the official foundation of JavaScript.
Sarah Gooding of Socket highlighted the developments in a blog post, breaking down the proposals that are now set to become part of the language spec:
- Float16Array introduces a new typed array for 16-bit floating-point numbers. It offers a memory-efficient option alongside existing types like
Float32Array
andFloat64Array
, making it ideal for applications that don’t require full-precision floats. - Redeclarable Global
eval
Variables aims to smooth out inconsistencies when usingeval
at the global scope. As it stands, usingvar
in globaleval
creates configurable properties, but re-declaring them withlet
orconst
throws an error. The proposal would allow such re-declarations, simplifying the language’s rules. - RegExp Escaping brings a long-awaited
RegExp.escape
utility to JavaScript. This method will help developers safely insert strings into regular expressions by escaping special characters, reducing chances of bugs and incorrect pattern matching.
JetBrains Tests AI’s Kotlin Skills
JetBrains recently explored how well large language models handle Kotlin-related tasks—and the results are a mixed bag.
Developer Vera Kudrevskaia explained on the Kotlin blog that while today’s models can discuss Kotlin competently, their answers can be outdated or incorrect. Common issues like losing track of context or miscounting elements still persist.
JetBrains Research evaluated several models, including DeepSeek-R1, OpenAI 01, and OpenAI 03-mini, using a custom Kotlin benchmark. They ranked the models and studied how each performed on a range of Kotlin-specific tasks.
According to Kudrevskaia, DeepSeek-R1 stood out in open-ended questions and reasoning, while OpenAI’s latest models also showed solid Kotlin skills. The evaluation included detailed code-based tests to identify where the models shine—and where they fall short.
JetBrains encourages developers to share unusual or incorrect AI-generated Kotlin answers in their public Slack or blog comments for further discussion.
OpenAI Introduces GPT-4.5 in Research Preview
OpenAI has unveiled a research version of GPT-4.5, describing it as their most advanced conversational model to date.
According to OpenAI, GPT-4.5 takes significant strides in both pre-training and post-training capabilities. The model is better at pattern recognition, forming connections between ideas, and generating creative responses without needing explicit logical reasoning.
Key improvements include a broader knowledge base, sharper intent recognition, and more emotionally intelligent interactions. OpenAI also claims that GPT-4.5 will produce smoother, more human-like dialogue and could be less prone to hallucination—although they admit more real-world feedback is needed.
The team released a blog post and a short video (around 13 minutes) introducing the model, inviting users to experiment and provide insights into unexpected or novel use cases.
Next.js 15.2 Rolls Out Performance and Debugging Enhancements
Next.js version 15.2 is now available, bringing with it notable updates in debugging tools, page metadata handling, and compile performance.
One major improvement is a redesigned error interface with clearer stack traces, making it easier for developers to identify and resolve issues during development.
Streaming metadata has also been added, enabling async metadata to load without delaying page rendering or navigation—leading to snappier transitions between pages.
TurboPack, the build system behind Next.js, has received optimizations as well. Developers have reported significantly faster compilation (up to 57.6% improvement in some routes) and lower memory usage—Vercel saw a 30% drop during local development.
The update also includes experimental features like support for the new View Transitions API from React and the use of Node.js runtime in middleware.
The Next.js team emphasized that TurboPack should now outperform Webpack in nearly all scenarios, and they’re inviting feedback from developers who find otherwise.