🤖 OpenAI’s New AI-Native Workspace for Scientists

🔬 OpenAI’s New AI-Native Workspace for Scientists

✂️ The $150 Million AI Startup Cutting Compute Costs

📍AI Tier Tracker

🛠Trending Tools

🥪Brief Bites

💰Funding Frontlines

💼Who’s Hiring?

It’s a free LaTeX-native workspace for drafting, revising, and publishing scientific papers. It basically provides AI-assisted support for citations and proofreading, as well as AI-driven automations for editing and formatting.

LaTeX is a widely used markup language and typesetting system for defining the content, structure, and formatting of scientific documents. For example, instead of clicking “bold,” “center,” and “font size,” scientists type “\cite,” “\emph,” and “\section.”

“I think 2026 will be for AI and science what 2025 was for AI and software engineering,” explained Kevin Weil, VP of OpenAI for Science. In other words, it’s not just a random release. ChatGPT already processes over 8.4 million messages every week on advanced hard science topics.

Traditional scientific research is very time-consuming and highly labor-intensive. Scientists spend months, or even years, analyzing empirical datasets, reviewing academic literature, and designing controlled experiments.

OpenAI’s tackling one of the most significant challenges in scientific research: finding effective ways to organize, interpret, and utilize data to drive new scientific discoveries.

LLMs are sophisticated autocomplete machines designed to predict the probability of a sequence of words. For example, when given: “The fat cat sat on the {BLANK}!” LLMs ask themselves: given the words so far, what’s the most likely next word?

They achieve this by leveraging attention mechanisms, which calculate how much each word should pay attention to every other word within a sentence. To do this efficiently, LLMs convert each word into Keys (K) and Values (V), together known as the KV Cache.

It essentially functions as an LLM’s short-term memory, enabling it to retain, recall, and reuse information from earlier parts of a sentence to maintain context and reduce redundant calculations.

As the conversations grow longer, the KV Cache can increase computational costs. The “v” in “vLLM” stands for virtual, as in software that makes it run faster and operate cheaper by applying advanced memory management and optimized batching algorithms.

This engineering innovation can reduce computational costs by roughly 70%, while also ingesting inputs and generating outputs 5x faster, enabling powerful frontier AI models to be economically viable at scale.

It’s a “universal virtual engine” that functions across different semiconductors. This allows AI researchers to run the most powerful frontier AI models on cheaper processors, serving as a primary catalyst in neutralizing NVIDIA’s monopoly on the AI chip market, which is projected to reach $564.87 billion by 2032.

Midship raised a $4.15M Seed Round for autonomous internal audits.

Fiddler landed a $30M Series C to power the control plane for AI Agents.

Compa closed a $35M Series B for AI-based enterprise compensation.

Astranis (San Francisco, CA): Product Quality Intern, Spring 2026

NVIDIA (Austin, TX): ASIC Hardware Design Engineer, Entry-Level

Descript (San Francisco, CA): Product Manager, AI Models, Mid-Level

Anthropic (San Francisco, CA): Public Policy, Events Lead, Senior-Level