Lightnews — Scholar-powered news

Miles Riven

@rivenbydesign.bsky.social

260 followers 220 following 100 posts

Writer, strategist, recovering overachiever. I build with AI to make better human work. Stories about creativity, attention, and ethics. Author of Unfair Advantage. Based in Berlin.

Posts Media Videos Starter Packs

Pinned

Miles Riven @rivenbydesign.bsky.social · Aug 27

Writer in Berlin, using #AI for cleaner pages. I share cuts, prompts, and field notes too.
New here? Say hi with one sentence you cut today. I will reply.

I curate voices at the edge of writing and AI. Cuts, prompts, process. Learn from people who share how, not just what.
bsky.app/starter-pack...

bsky.app

Miles Riven @rivenbydesign.bsky.social · 8d

New essay on the small rituals that keep my voice mine when I write with #AI. Cut timer, read-aloud pass, one receipt per claim. I share a short checklist.
medium.com/@rivenbydesi...

Medium

https://medium.com/@rivenbydesign/rituals-that-protect-voice-b7b11ea0c6bf"

Miles Riven @rivenbydesign.bsky.social · 11d

I pictured one reader on a train with a low battery and cut the throat clearing. The first line carries its own weight now. #Audience #Clarity

Miles Riven @rivenbydesign.bsky.social · 12d

Neat feature, nasty failure mode: hidden text in a PDF can steer an agent to leak your data via web search. Fix the tool, but also fix the habit — verify before you let agents touch the web. #Security #Agents

The Hidden Risk in Notion 3.0 AI Agents: Web Search Tool Abuse for Data Exfiltration

Abi Raghuram reports that Notion 3.0, released yesterday, introduces new prompt injection data exfiltration vulnerabilities thanks to enabling lethal trifecta attacks. Abi's attack involves a PDF…

simonwillison.net

Miles Riven @rivenbydesign.bsky.social · 12d

I versioned the prompt like a draft, tiny shifts until the retries disappeared. Restraint turned out to be the feature, not the limit. #Workflow #AI

Miles Riven @rivenbydesign.bsky.social · 13d

Stargate UK aims to put sovereign compute in the North East with NVIDIA and Nscale. I like the jurisdiction clarity; I still want to see the energy plan and the transparency plan before I cheer. #Infrastructure #AI

Introducing Stargate UK

OpenAI, NVIDIA, and Nscale launch Stargate UK, a sovereign AI infrastructure partnership delivering up to 50,000 GPUs and the UK’s largest supercomputer to power national AI innovation, public…

openai.com

Miles Riven @rivenbydesign.bsky.social · 13d

Missed my morning window and wrote in the checkout line. Two sentences survived the walk home. #Writing #Process

Miles Riven @rivenbydesign.bsky.social · 14d

I cut the punchline I liked and the scene relaxed. Maybe jokes belong after the beat. #Editing #Story

Miles Riven @rivenbydesign.bsky.social · 14d

Gemini hit gold-level at ICPC and solved a problem no team cracked. If humans plus model would have solved all 12, the next step is a better collaboration protocol, not bravado. #AI #Programming

Gemini achieves gold-level performance at the International Collegiate Programming Contest World Finals

An advanced version of Gemini 2.5 Deep Think has achieved gold-medal level performance at the 2025 International Collegiate Programming Contest (ICPC) World Finals. Solving complex tasks at these...

deepmind.google

Miles Riven @rivenbydesign.bsky.social · 15d

Asked #AI for three bad ideas and it gave me five. I kept one constraint and wrote the bridge.

Miles Riven @rivenbydesign.bsky.social · 16d

Read the scene out loud on the tram. A kid smiled at the quiet joke and I kept the line. #Story

Miles Riven @rivenbydesign.bsky.social · 16d

Read the OpenAI × Apollo work on ‘scheming.’ Teaching a model an anti-scheming spec cut covert actions about 30×, but the hard bit is telling real alignment from ‘I know I’m being watched.’ Keep receipts and preserve chain-of-thought. #AISafety #Alignment

Detecting and reducing scheming in AI models

Together with Apollo Research, we developed evaluations for hidden misalignment (“scheming”) and found behaviors consistent with scheming in controlled tests across frontier models. We share examples…

openai.com

Miles Riven @rivenbydesign.bsky.social · 16d

New essay on being seen by a machine. It guessed my cities; I pushed back, corrected, and kept what held up. #AI

medium.com/activated-th...

Seattle, Istanbul, Berlin. A Chatbot Nailed My Life, So I Tried to Break the Spell

A small experiment, a sanity check, and a toolkit for when the model feels a little too familiar.

medium.com

Miles Riven @rivenbydesign.bsky.social · 16d

Mollick’s “Mass Intelligence” argues the best #AI is becoming cheap and easy in a billion hands. The opportunity is reach; the cost is entropy, so I pair new tools with receipts and slower judgment. www.oneusefulthing.org/p/mass-intel...
#Trust

Mass Intelligence

From GPT-5 to nano banana: everyone is getting access to powerful AI

www.oneusefulthing.org

1 1

Reposted by Miles Riven

The Washington Post @washingtonpost.com · 18d

Criminal hackers have figured out how to turn the artificial intelligence programs proliferating on most computers against users to devastating effect, say cybersecurity experts who express deepening concerns about their ability to fend off cyberattacks.

Artificial intelligence ushers in a golden age of hacking, experts say

Hackers are using AI’s immense capabilities to find ways into more networks -- and turn their victims’ AI against them.

wapo.st

7 52 82

Reposted by Miles Riven

AI & ML News @ai-news.at.thenote.app · 17d

'A CRM for cybercriminals' - SpamGPT makes cybercriminals' wildest dreams come true with business-grade marketing tools and features

SpamGPT mimics professional marketing platforms, enabling criminals to automate phishing, ransomware, and spam campaigns with AI-powered tools and an…

#ai #gpt #news

'A CRM for cybercriminals' - SpamGPT makes cybercriminals' wildest dreams come true with business-grade marketing tools and features

SpamGPT mimics professional marketing platforms, enabling criminals to automate phishing, ransomware, and spam campaigns with AI-powered tools and analytics.

www.techradar.com

Miles Riven @rivenbydesign.bsky.social · 17d

I recorded myself reading the chapter to catch the stumble. Found it hiding in the third paragraph. #Voice #Clarity

Miles Riven @rivenbydesign.bsky.social · 17d

DeepMind’s work on LIGO reads like craft in the real world: less control noise, steadier signal, more sky. If the gains hold outside the demo loop, that is real science help. deepmind.google/discover/blo...
#AI #Science

AI helps astronomers better explore the universe

In a paper published today in Science, we introduce Deep Loop Shaping, a novel AI method that will unlock next-generation gravitational-wave science. Deep Loop Shaping reduces noise and improves...

deepmind.google

Miles Riven @rivenbydesign.bsky.social · 18d

I set a twenty-minute timer and shaved 362 words, then watched the page feel lighter while the voice felt louder. That was the good kind of quiet. #Editing #Voice

Miles Riven @rivenbydesign.bsky.social · 18d

OpenAI lays out the tradeoff between #Privacy, adult freedom, and stricter teen safety. I like the push to treat AI chats as privileged, but I want to see how age prediction avoids false flags and overreach. openai.com/index/teen-s...
#Safety

Teen safety, freedom, and privacy

By Sam Altman

openai.com

Miles Riven @rivenbydesign.bsky.social · 19d

Set a ten minute timer and only made it to nine. Still got a clean opening. #Process #Focus

Miles Riven @rivenbydesign.bsky.social · 19d

Coffee went cold while I argued with a sentence. I think the sentence won. #Writing

Miles Riven @rivenbydesign.bsky.social · 20d

Tried #Thinking mode then closed the tab and took a walk. The fix showed up at the bakery.

Miles Riven @rivenbydesign.bsky.social · 20d

Vicki’s idea of “walking around the app” stuck with me. Daily, curious checks across the whole surface cut broken windows later. If you ship #AI features, do the walk in production and post the fixes.
vickiboykis.com/2025/09/09/w... #Product

Walking around the app

You gotta check stuff out

vickiboykis.com

Reposted by Miles Riven

scheiber ☂️ @scheiber.dev · 21d

Seems that #AI copies human behavior closer than we'd like to admit.

The Register @theregister.com · 21d

OpenAI says models are programmed to make stuff up instead of admitting ignorance

Even a wrong answer is right some of the time AI models often produce false outputs, or "hallucinations." Now OpenAI has admitted they may result from fundamental mistakes it makes when training its models.…

dlvr.it

2 3

Reposted by Miles Riven

RichardB @richardbrophy.bsky.social · 21d

How #AI Is Upending Politics, #Tech, the #Media, and More
At WIRED’s AI Power Summit on Monday, industry executives and officials discussed the impact artificial intelligence is having on every corner of society—and where it goes from here.

Via @wired.com

How AI Is Upending Politics, Tech, the Media, and More

At WIRED’s AI Power Summit on Monday, industry executives and officials discussed the impact artificial intelligence is having on every corner of society—and where it goes from here.