🧠 Nvidia just paid $20B to make a problem disappear

Remember Groq? The inference chip company whose demos made everyone ask, “Wait, why is this so much faster than everything else? Nvidia noticed too. It just agreed to a roughly $20 billion package to license Groq’s inference technology, acquire key chip and IP assets, and bring over much of Groq’s leadership team, in what is now Nvidia’s largest deal to date.​

Here’s the thing: Groq’s LPU-style architecture is purpose-built for high-throughput inference — the part of AI that matters when you are actually using models, not just training them. Nvidia’s GPUs still dominate training, but inference at scale is where a huge share of long-term revenue and cost efficiency will be decided.

Translation: Nvidia saw a strategically important rival in inference and wrote a very large check to pull Groq’s tech and talent closer, via a non-exclusive licensing and asset deal rather than a full corporate acquisition.

OpenAI launched "Skills" — reusable instructions you can attach to ChatGPT so it stops forgetting your preferences.

Think: "Always format my code in Python 3.11 style" or "When I ask for emails, make them short and slightly funny." You save it once, apply it whenever.

It's basically custom instructions on steroids. Instead of one mega-prompt trying to capture everything about you, you build a library of specialized behaviors and mix and match. For developers, this is huge. For everyone else, it's the difference between training a new intern for every conversation and having an assistant who actually remembers you hate bullet points.

OpenAI published research on AI agents that can conduct real biological experiments — including pipetting, measuring, and the entire wet lab workflow.

The system combines reasoning models with physical lab automation. You describe an experiment in plain language, and it figures out the protocol, controls the equipment, and executes it.

Early days, but the implications are wild. Biology's biggest bottleneck isn't ideas — it's the time-consuming, repetitive lab work that burns out PhDs. If AI can handle the execution, human researchers can focus on asking better questions. We're officially past "AI writes code" and into "AI does science."

Prompts.chat Vibe Coding Library

A curated collection of prompts specifically for vibe coding projects. Skip the "write me a Python script" basics — these are designed for building actual apps with AI assistance. Bookmark it.

LLM Council — multi-LLM consensus answer tool

LLM Council is an open-source project that lets you ask the same question to multiple language models at once and combine their answers into one. Instead of depending on a single model’s response, it sends your prompt to several LLMs in parallel, has each model review and rank the others’ outputs, and then uses a designated “chairman” model to synthesize a final answer that reflects the best reasoning from the group. Think of it like a panel of experts debating before delivering a consensus reply.

Turn raw data into an interactive presentation. Here’s a quick 3-step setup that transforms static spreadsheets into a dynamic, shareable deck:

1️⃣ Upload your files — Drop your PDF and CSV/Excel into ChatGPT.

2️⃣ Ask for analysis — Tell it what to look for (trends, insights, summaries).

3️⃣ Connect Replit — Have ChatGPT generate an interactive presentation that visualizes the data — complete with charts, filters, and captions.

The result: a live, data-driven presentation you can edit and share instantly — no coding, no manual chart work.