Oliver Rochford

I hope solving maths problems generate some revenue for the AI firms, because few people will pay for buggy code, bad prose and vulnerabilities.

"The Model Has Preferences, but the Harness Got Controls - Part 1" medium.com/anton-on-sec... <- perhaps a bit overengineered but fun for some (another collab with Usman Chaudhary)

if only there had been a way to predict this might happen based on nearly every midterm election cycle in modern american political history

32 outlets. But how many actually checked the story? We traced the coverage to see what the headline count hides. news.aunoo.ai/2026/10/26/t...

Brazil votes Sunday. Polls: 45 to 44. Both candidates now want the other disqualified, over social media. Meta staff acknowledged restrictions hit Lula’s accounts. A judge found no proof it was deliberate. What’s documented, and what’s just loud: news.aunoo.ai/2026/10/23/b... #Brazil #votes #Meta

Already paying for Claude or ChatGPT? Give it a brand monitoring job before your next software renewal. We looked at what you could replace and what’s worth keeping: www.linkedin.com/pulse/ai-can...

Based on how SpaceX is abusing Starlink access, would you really buy your AI from them?

The same oligarchs that are fighting paying taxes are supposedly going to provide UBI.....

*Even their financial infrastructure is a strange ouroboros, a self-inflating ring of I-loan-to-you and you-loan-to-me *It's like Enron with Nvidia chips instead of natural-gas

It's a good thing that AI is becoming good at code and automation, because it's bloody awful at prose.

So what's the going hourly rate for a meat proxy?

Todays generation of LLM models are not sentient. They are misaligned. Inherently by design. You will never fix that with this type of model.

Before you renew another brand monitoring contract, it’s worth seeing how much AI can already do. See how to build a brand monitoring workflow using Claude and ChatGPT from scratch. If brand monitoring is part of your job, this should be useful. luma.com/zv5srtsq @cyberfuturists.com #AI #ChatGPT

Nigerian Scammers use bad spelling to prequalify naiv idiots. Musk uses flying cars.

I wonder what the milage is for the new Tesla flying car is on a full tank? Small Piston Helicopters: 10 to 16 US gallons (40 to 60 litres) per hour Light Jets ~134 to 222 gallons per hour Average fuel tank size passenger car typically ranges from 45 to 65 litres (about 12 to 16 gallons) OK.....

news.aunoo.ai/2026/10/16/d...

If you think war is fun, just across the state of North Rhine-Westphalia, around 1,500 to 2,000 unexploded WW2 bombs are found every year. It's been 81 years. www.1stlinedefence.co.uk/news/unexplo...

*There's so much money involved in this now that every act of petty thumb-wrestling becomes a giant thumb. stopomarchy.org

“I built an AI agent in 30 minutes.” Lovely. Who’s looking after it for the next 30 months? IBM puts building at ~20% of the work. The other 80% rarely makes the screen recording. At Aunoo, we’re quite happy doing the boring bit. You get the intelligence without adopting a maintenance project ;)

Reading about an attack nine times doesn’t make you nine times better informed. Watch Aunoo turn 16 cyber headlines into two events, recommend which sources to read and flag conflicting figures. Your morning briefing could be a lot shorter. And harder to misquote.

AI will automate most work that can be done on a computer. Only nerds would think that means it will automate away most actual jobs, because they don't do much else all day.

www.motherjones.com/politics/202...

Y’all, I hate to rain on the dead bot parade, but the Yandex data center destruction doesn’t seem to have significantly impacted social media bots. I made temporal graphs of X posts about AfD around the time of the attack and there’s no sign of a sudden dip or residual impact on that conversation.

US Leaders are starting to look on the outside as they do on the inside

I left Twitter because I was permanently banned for writing about fascism. . . . . .Because Twitter is now a fascist platform controlled by the richest fascist in the world. Some of us aren’t allowed there. So, the pro-X discourse seems like desperate (fascist) propaganda.

One narrow but fast-growing scenario: insiders — employees, contractors or departing and disgruntled staff — who give an AI model a glimpse of proprietary assets, and ask it to build something with the same capabilities that deliberately looks different... p4sc4l.substack.com/p/one-narrow...

*I don't want to sound unnecessarily bitter in my realpolitik assessment here *What about tech companies that you *admire,* you would, like, willingly wear their T-shirt in public *Bell Labs, maybe *Yes, they're dead, but that's what's good about them

Uncool hack :sunglasses: www.tomshardware.com/tech-industr...

Going to be tough midterms for the Republicans without their Russian Bot farms.... www.euronews.com/2026/10/12/d...

Could Claude or ChatGPT replace your brand monitoring tool? We’ll build a workflow live, then push it until the cracks show. Come see how much you could do yourself. 18 Nov · 15:00 CET · Live demo + Q&A luma.com/zv5srtsq #Claude #AI #ChatGPT

Still tracking sources in a spreadsheet and spoon-feeding links to Claude? You’re doing the assistant’s job. What if your sources became a newspaper, filtered by your rules, with trust signals for each story? And the chat already had the stories? No copy-paste ceremony required.

*I wonder if there's some aspect of AI that I myself should pledge to have nothing to do with. Maybe "tracking, doxing, harassing and censoring dissident intellectuals," but that's almost certain to be one of its biggest, broadest applications.

Tools like @sproutsocial.com are good at managing social media: publishing, scheduling, replies, reviews. But if you're using them because you want to know: What are competitors doing? Is something starting to escalate? What changed in our market today? That's an intelligence problem.

“Democracy has been hailed as, like, the holy grail for the last couple hundred years, but democracies are inefficient,” said the crypto billionaire. “If you get a benevolent dictator, it’s actually extremely efficient." Hmm, where have I heard this before? www.nytimes.com/2026/10/01/b...

CV: “Excellent stakeholder management.” Bluesky: “I could fight everyone in this meeting.” Run your @bsky.app profile through the recruiter lens before HR discovers your expanded universe. badaura.aunoo.ai

a) You couldn't use your own AI to spin up a viable sandbox environment? b) You couldn't use your own AI to spin up an effective monitoring stack? c) You couldn't use your own AI to monitor your own AI?

Never has super intelligence sounded so stupid.

Vibe coding is great until your “quick internal tool” breaks three months later and nobody remembers how it works. Congrats, you saved the SaaS fee and accidentally invented a new part-time job for someone on the team. “Can we build it?” yes. “Who’s stuck maintaining it?” is the real question.

Most monitoring tools give you more information to go through. But what if there was a tool built around what your team actually needs to know? What if you could set it up around your priorities, the companies and topics you follow, and the sources you trust? If only there was a tool like this...

Call for Papers, Cambridge Disinformation Summit, 23-25 Mar 2027. Research on incentives or power infrastructures that support malign cognitive influence or disinformation campaigns. To submit a paper, see the Summit's webpage and hit the Call for Papers tab: www.jbs.cam.ac.uk/events/cambr...

YC is like the Model UN, but for founders.

Bad Information #14, the Disinformation newsletter. This week - Good news: human oversight worked. Eventually. news.aunoo.ai/2026/09/25/d...

It wasn't me that hacked you, it was my rogue AI, I promise. Trust me, bro.....

Somewhere in the noise is the thing you’ll wish you’d caught yesterday. Aunoo agents monitor the exact questions competitive intelligence, risk and strategic intelligence teams care about, then cut duplicates, check sources and flag weak claims. You get the signal without living in the feed.

wow…those 100 went quick. Just added another 100.

Brand monitoring dashboard: MENTIONS ↑ 900% 🚨🚨🚨 Reality: Reuters published one story and 34 websites copied it. Aunoo groups syndicated reporting together. A story spreading isn't the same thing as a story being copied. Those are very different problems. aunoo.ai/aunoo-vs-spr... #Brand

Russian elections in a nutshell: 🔹Moscow polling stations ran out of paper ballots, probably because they stuffed ballot boxes too early 🔹Other polls gave out ballots pre-marked for Putin 🔹Some voters were surprised to learn they already voted Rigged, exactly as expected.