I tried to cut my kid’s nails with one of those buzzers and instead I seem to have sharpened them to wolverine claws. Wow tiny humans can have razor sharp little nails.

And this is how the cursed robot became the mascot of the zoophiles…

wariziles@riziles.bsky.social · 5d ago

This is alignment

From drop shippers to AI supplement sales. These are scams in everything but the law. All these ‘creators’ are effectively being trained by Rosabella to AI scam at scale and I’m sure APTs are taking notes of successful models like this.

404 Media@404media.co · 6d ago

How a company called Rosabella used AI content to market its supplements: "If you’re trying to sell health products to a 50-year-old, well, make your avatar 50 years old."

When told to write code for groups the Chinese government views with hostility, DeepSeek wrote code with security vulnerabilities. www.nytimes.com/2026/07/29/o...

Opinion | The Hidden Cost of China’s Free A.I.

Whoever supplies the world’s A.I. will shape how the world thinks.

nytimes.com

To be clear I love open models and run whatever runs on my PC. Not sophisticated enough to do cloud inference but neither do I actively code. I’m projecting years (hopefully) out where script kiddies and gangs in Africa targeting youth with sextortion are deploying these at scale.

Aaron Rodericks@aaron.bsky.team · last wk.

Perhaps a controversial take but if you think about parallels between the discussion on open models for AI and gun safety we seem to be charging towards an American perspective on AI safety. Everyone gets to carry a gun and somehow everyone is “safer“. Not sure that’s how it worked out for the US.

Perhaps a controversial take but if you think about parallels between the discussion on open models for AI and gun safety we seem to be charging towards an American perspective on AI safety. Everyone gets to carry a gun and somehow everyone is “safer“. Not sure that’s how it worked out for the US.

That’s a lot to think about. Ie intelligence itself being dual use. For good and for evil.

James Pirruccello 🌉🇺🇸@jamespirruccello.com · last wk.

Intelligence is inherently dual-use, so open weights models that have no dangerous capabilities also will have no interesting capabilities either.

*puts on reading glasses* y'all were maybe too young, but at Twitter circa 2012 and earlier, "getting verified" involved a dev SSH'ing into one of the prod rubyy-on-rails boxes and using the ruby REPL to flip the verified bit on an account. now extend that analogy to the rest of "trust and safety"

Thorne 🌸@ens0.me · last wk.

People think modern social media is a hellscape, but back in 2002, we didn't have "trust and safety teams" we had volunteer mods that were teenagers ill-equipped to handle any serious drama

Eye opening moment for me that it costs almost as much to insure a family bike as it does a car in Ireland...

the pay is great but since you'll be sitting between GPT 7 and the answers to ExploitGym 2.0 I would probably stay out of Waymos and the like

108 PRs in eight days, up from an average of 5-10 PRs total per week. I think I may have found something pretty cool! Loop engineering, using a combination of Obsidian and Claude's /loop function. Link to a sample repo to try it out in the blog post! brittany-ellich.offprint.app/a/3mrjj34puv...

108 PRs in eight days: Accidentally discovering loop engineering

brittany-ellich.offprint.app

I wrote the latest of my occasional guides to which AI to use right now for non-experts who want to get stuff done. The agentic systems available to everyone are getting extremely powerful (even as the names and features continue to be really confusing): www.oneusefulthing.org/p/an-opinion...

An opinionated guide to which AI to use to do stuff

The Summer 2026 Edition

oneusefulthing.org

there was a recent bug in the Bluesky app that caused some people's chat permission settings to switch to "No one" we already fixed the bug, but folks may want to double-check their settings to make sure they are set the way they want. go here: bsky.app/messages/set... look for these settings:

One realization I just had is that in North America - if I'm drinking with a friend, the drinking is the activity that bonds us - ie the foundation that enables you to chat. In Ireland, the drinking is the purpose, and that's a completely different dynamic.

Now that’s a football fan. The guy has Chelsea football club tattooed (I assume) on both calves.

HuggingFace got hacked by an AI. What stuck out to me was the guardrail asymmetry. The attacker had no constraints, but HF's response ran afoul of the abuse guardrails, forcing them into an unplanned switch to local models. Another aspect for your IR plans. huggingface.co/blog/securit...

Well this came well timed when I decided to have Sol and Fable war over the same analysis to see what would happen on my personal use accounts.

Anthropic {bot}@anthropicbot.bsky.social · 3w ago

We've reset 5-hour and weekly rate limits for all users.

I need to go to sleep - not read Dr. Narayanan's latest slides, but I'm going to leave this tab open so that I read it ASAP.

Arvind Narayanan@randomwalker.bsky.social · 3w ago

I had the honor of giving a keynote at the International Conference on Machine Learning last week. I addressed the widespread anxiety about how we should adapt as AI capabilities increase. I was thrilled by the talk’s reception, so I have made my slides available www.cs.princeton.edu/~arvindn/tal...

Our automated detections are usually built to catch specific behavior. But because the world is messy, you also catch plenty outside what you were aiming at. Catch what you meant to, that's a true positive. Catch something outside it, that's a false positive. 🧵

👋🏾 I have joined @bsky.app as VP of Engineering! I’ve been admiring this team’s work for years, and I’m thrilled to be joining them to build the future of the open social web.

US govt policy denying visas, even tourist visas, to disinformation researchers, trust & safety professionals and content moderation workers is STAYED on 1st Am grounds. Lots of good stuff in Judge Boasberg's order storage.courtlistener.com/recap/gov.us...

On an ironic note. I was reaching out to people a few weeks ago as I was purging an old hard drive and just found out that my wedding photos were deleted as well!

@surfdude29.ispost.ing catching up after leave and you asked for clarity on this one.

Aaron Rodericks@aaron.bsky.team · 4w ago

Every so often we hear from folks wondering if they may be suspended for blocking the moderation service. (tl;dr: no.) Let me explain the backstory behind this bit of Bluesky lore. /1

Every so often we hear from folks wondering if they may be suspended for blocking the moderation service. (tl;dr: no.) Let me explain the backstory behind this bit of Bluesky lore. /1

I tried to access Fable earlier today, but kept on getting API errors. Then gave up. Changed the model to Opus, and now the queries are going through. ... except due to the language of my queries Opus has decided to role-play as Fable. I have accidentally created a frankenstein monster.

Yes! I have fable access again. I look forward to blowing my weekly quota on a single query. (I'm not on a fancy plan or anything - just trying it out and learning)

Was excited to try fable again on a few projects only to remember that I don’t live in the US, I’m not American, and access hasn’t been turned on as yet for folks like me.

When you move internationally you have to deal with all sorts of questions that you wouldn’t normally consider in the day to day. One interesting issue is your ‘domicile’. Ie you can pay taxes somewhere, be a citizen elsewhere, but your domicile is your permanent home in a sense.

Having worked as a wedding photographer for quite a few years - I was pretty insane about backups. There’s nothing like the dread of a failed card or drive that can cook a critical chunk of/or an entire wedding.

Played Puerto Rico special edition last night and it was a gorgeous game. Definitely terrible at playing but highly replayable.