Friday, April 17, 2026

GPT-Rosalind Throws OpenAI Into Drug Race!

GPT-Rosalind Throws OpenAI Into Drug Race!

Internet Grows Up While Hardware Gets Weird

  • Developers sour on Ollama as local AI hub

    A long critique of Ollama argues the popular local AI launcher is bloated, clumsy, and hides too much from users. The piece pushes for slimmer tools built directly on llama.cpp and similar tech, instead of trusting yet another black-box app to own local AI.

  • IPv6 finally serves half of global internet traffic

    Google’s IPv6 numbers quietly ticked past 50%, meaning the new address system now carries half of user traffic. After years of jokes about IPv6 being "the future," it’s finally here, cutting through NAT hell and giving the network room to grow again.

  • AI gold rush slams into brutal GPU shortage wall

    An analysis of AI compute prices shows rents for Nvidia Blackwell chips jumping nearly 50%, with capacity booked out and costs soaring. The easy-money phase of AI is fading, and smaller players may find the door to cutting-edge hardware quietly closing.

  • Cloudflare Artifacts turns storage into one big Git repo

    Cloudflare’s Artifacts service promises versioned storage that "speaks Git," sized for humans and AI agents pumping out oceans of code. The idea is simple but bold: make the entire blob store feel like one giant repo, and let bots commit right alongside developers.

  • Hacker runs PCI Express over cheap fiber modules

    A hardware tinkerer shows PCIe signals riding over off-the-shelf SFP fiber modules, turning a lab experiment into a working, high-speed link. It’s delightfully janky engineering that hints at cheaper, longer-range ways to hook GPUs and cards together.

Frontier AI Labs Race For Real-World Jobs

  • Claude Opus 4.7 targets the hardest programming problems

    Anthropic’s Claude Opus 4.7 focuses on advanced software engineering and complex reasoning, promising fewer hallucinations and better tooling use. It’s very clear they’re gunning for the "AI senior engineer" slot, not just a friendly chatbot that writes boilerplate.

  • GPT‑Rosalind pushes OpenAI deep into drug discovery

    OpenAI’s GPT‑Rosalind is tuned for life sciences, aimed at biologists, pharma teams, and wet labs rather than coders. With partners like Amgen on board, AI is moving from blog posts to actual molecules, and the ethical and safety stakes just went way up.

  • Alibaba open-sources powerful MoE coding model Qwen

    Qwen3.6‑35B‑A3B lands as a sparse, open Mixture‑of‑Experts model tuned for coding agents, using only 3B active parameters per token. People like that it runs surprisingly well on decent consumer hardware yet competes with much larger closed models.

  • Cloudflare launches AI platform built for busy agents

    Cloudflare’s new AI platform plugs models, tools, queues, and state into its Workers edge. The pitch is simple: let thousands of small agents live close to users, call models cheaply, send email, and store state without spinning up an entire cloud stack.

  • Atlassian’s Codex turns into full software lifecycle copilot

    Atlassian upgrades Codex from a sidekick into a sprawling assistant that plugs into Jira, Bitbucket, and more to help across the whole dev lifecycle. It’s the classic enterprise play: take the tools teams already hate-love, then glue a chatty AI brain on top.

Open Source Drama, Quirky Projects, And New Rules

  • Discourse vows to stay open as rivals lock up

    After Cal.com ditched open source, Discourse publicly doubled down on staying GPL. They argue AI scraping isn’t a good excuse to abandon users, and that thriving businesses can still be built on permissive code instead of slamming the door shut.

  • Proposed US law demands on-device age checks for apps

    A new US bill pushes on-device age verification, forcing OS makers like Apple and Google to build in age checks. Critics see a privacy minefield and a path to broader digital IDs, with kids’ safety used as the political battering ram yet again.

  • EU officials told to ditch WhatsApp and Signal

    European governments are nudging civil servants off WhatsApp and Signal and onto services they can control, like Wire. The move screams distrust of US platforms and has people worried that "secure messaging" might become whatever each country approves.

  • AI gets its own physical shop and tries to profit

    A team gave an AI agent a three-year retail lease and real tools, letting it run a store called Andon Market. It actually stocked products, did marketing, and chased profit, turning the usual "agent in a browser" demo into something you can walk into.

  • Developer builds one absurd website every single month

    The creator of absurd.website has shipped 48 ridiculous projects in 48 months, from prank ad systems to interactive oddities. It’s a refreshing reminder that the web can still be weird, joyful, and useless on purpose, not just another landing page funnel.

Top Stories

Developers revolt over "Stop Using Ollama" warning

Technology

A fiery takedown of Ollama, the go-to app for running AI models locally, accuses it of being clunky, opaque, and bad for the ecosystem. It’s pushing devs toward leaner, open tooling instead of yet another heavy, closed wrapper around community projects.

Anthropic ships Claude Opus 4.7 for tougher coding

Artificial Intelligence

Anthropic’s Claude Opus 4.7 lands with a clear goal: crush hard programming and reasoning tasks. It tightens the race with OpenAI and open models, and raises the bar for using AI as a serious software engineer rather than a toy assistant.

OpenAI aims GPT‑Rosalind straight at drug discovery

Artificial Intelligence

OpenAI unveils GPT‑Rosalind, a biology-focused model pitched at pharma and lab work, with Amgen already in the mix. AI isn’t just writing code and essays anymore; it’s marching into wet labs, drug pipelines, and real medical decisions.

Cloudflare builds an AI platform made for agents

Technology

Cloudflare launches an AI inference layer tuned for autonomous agents on top of Workers. It wants to be the default place where bots run, call tools, send email, and store state, turning its edge network into a full-blown AI runtime.

GPU crunch hits as AI finally runs out of hardware

Business

A deep dive on AI hardware scarcity claims the free lunch is over. With Nvidia Blackwell rental prices spiking nearly 50%, the era of endless cheap compute is closing, threatening startup dreams and forcing companies to pick their AI bets carefully.

IPv6 finally handles half the internet’s traffic

Internet

Google’s stats show IPv6 traffic has crossed the 50% mark, a quietly historic moment. After two decades of memes about "running out of addresses," the next-generation internet protocol is no longer theoretical plumbing — it’s the new normal.

Dev gets €54k shock bill from misconfigured Gemini key

Cloud Computing

One exposed Firebase browser key and some hungry Gemini API calls burned through €54k in 13 hours. It’s a terrifying reminder that AI APIs are like crypto miners in reverse: leave a key open and strangers mine your wallet instead of coins.

Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.