A real concern re: Project Glasswing is that all the partners are IT and finance. "We're promising $100M to Amazon, Apple, and JPMorganChase, but the underfunded municipal water utility is on their own." I'd feel a *lot* better about the whole thing if there were some e.g. Honeywells on the list.

It's looking like this is at least partially true. Also: running costs in the range of "more than businesses will routinely spend but a rounding error for a state actor." I do not like this equation.

the funniest thing about the bsky-is-dying discourse is that everyone has already purged mastodon as an alternative from their memory (laudatory)

There's a proposed datacenter project in my area. Science said that it was impossible to find dumber reasons for people to be upset about it. Karen owns property with unmarked graves *two miles* from the proposed site, is upset about it. Science was wrong.

Barber is such a great spectator track. Also, more race tracks need art installations like a giant spider and a lady in the lake.

Good morning, nerds. It's a beautiful day to watch racecars go VROOOOMVROOOM BRINGDINGdingdingding at Barber Motorsports.

A cascade of conditionals: IF Anthropic Mythos represents an actual step change and not marketing hype AND it's in any way related to recent safety commitment walkbacks AND the safety concerns are real AND it gets released anyway BUT gatekept to those deemed worthy That would be a Bad Thing™️

hey, so uh. I know we say nice things about Anthropic around here a lot, but under no circumstances would I recommend people use Anthropic's Dispatch, and I would blanket-ban, fully disable, their computer-use stuff they're off the rails with this stuff and it's irresponsible to have normies use it

Ahmadinejad was killed, but it seems the Poaster's Spirit lives on in Iran.

re: the Cyclorama. the standup show in question was Doug Stanhope at an old abandoned church. we were literally seated in pews like we were receiving a sermon as an actual jug of moonshine got passed around the crowd. that was a weird 24 hours.

apropos of both some conversations and some frameworks I've been looking at today: a key hurdle to meaningful enterprise AI adoption is going to be the fact that agile provides some useful repurposable terminology, but is a structural mismatch for what future workflows are going to look like

So, a really, *really* important thing here. Anthropic's major inference partners: AWS. Google. Microsoft. All defense contractors.

beyond lists being abused in this manner, there's a massive problem with lists that exist purely as a form of block-evasion and a way of getting a few last digs in at someone they're mad at. you'll see people suddenly tagged with a half dozen lists that are literally nothing but insults.

zaslav scamming the ellisons is a possibility I hadn't considered that is, in fact, very funny

1. He's so running lol 2. Other Democrats really need to be paying attention to his messaging. This is someone who's done it in a state with razor sharp margins showing how you reach actually swingable independents.

lol, lmao

the funniest thing about this is he'd clearly been coached that the markets really want certainty, because he kept repeating the word over and over and over during the press conference yesterday. then <24 hours later he's arbitrarily changing things that haven't even gone into effect yet.

....fine. I'll get back to playing minigames instead of automating my production lines.

Scrolling YT live, I get a stream of a Finnish woman blasting metal in the middle of the night while driving around some sort of machine with a mechachainsaw arm cutting down trees. I now understand that AW and Control were just @remedygames.com making loving documentaries about their homeland.

A short story about self-nerdsniping and Claude Code: A conversation with a friend about John Ossoff's age lead to a shower thought, which lead to a Claude Code session, which lead to a couple of hours of data-wrangling, which lead to emailing the US Senate Historian to correct an official record.

short thread because it surprises me that "it's not just generic AI demand" hasn't percolated into the wider consciousness yet

Something I find fascinating about watching Ossoff speak is seeing a Jewish man that has adopted some of the oratory traditions of the black church. (laudatory) It's nice actually having a pair of good senators in the state. Hopefully we can hang on to that.

This is bad because it'd require any companies doing business to cut ties with Anthropic. But the Pentagon is definitely helping Anthropic's court case by telling the press "lol, we're absolutely doing this for arbitrary and capricious reasons."

During the most recent round of moderation discourse, I was assured that the complaints were "silly" and "a waste of time" because the account in question just wasn't blocking hard enough. Who could have foreseen this?

This is probably the most thoughtful and nuanced take I've seen on this so far. Definitely mirrors what I've been experiencing so far.

I didn't realize "prolifically illiterate" could be an accurate descriptor for a person and yet

People are getting the order of operations all wrong. Mass movements are more powerful than fifth columns. Do the powerful thing first, especially when trying to do the weaker thing undermines the ability to do the big thing.

This is an incredibly cool concept. I've wanted something like a generation ship management sim for years and this sounds like this concept zoomed out a few levels.

Probably says something that every time I've seen anyone with domain expertise comment on Annie Jacobsen's nuke work, the reaction has been "WTAF is she talking about?“

This is pretty cool. The fact that it was done only using a relatively basic utilization of Claude Code and the actual execution isn't leveraging any sort of multiagentic feedback is pretty impressive. Also hints at what would be possible with a more robust approach.

It's very funny seeing this thread break containment and a lot of the replies basically screaming "E.G.,"

Even setting aside the morality and legality of it all, the 47 days of training is showing and it's eventually going to lead to a mass casualty event for their agents if they keep repeating these tactics. This is about as dangerous as you could make this situation for yourself.

An aspect of working with LLMs I don't see mentioned often: one way my ADHD tends to manifest is crippling analysis paralysis. Week-months+ not starting something because I'm busy over researching the "right" way to do it. Design/critique workflows get me over that initial just-start hump.

Anthropic/Claude are familiar to people exposed to tech. They're very unknown by the general public. Last I looked into it 6 months ago, they had name recognition in the mid single digits (OpenAI/ChatGPT was ~80%). They're almost entirely enterprise focused; helps them avoid some problematic uses.

Our dog knows he has to lay down before he gets his bowl at dinner time. Sometimes he'll lay down when I have a treat that I don't immediately give him. He just did lay down in front of the locked dog door. It's locked because there's a possum outside. "This is how I get things I want to eat."

Claude continues to prove itself the most sophisticated of the LLMs.

SCOOP: Flock left at least 60 of its AI-powered, people-tracking surveillance cameras exposed and livestreaming to the open internet. We tracked ourselves. The exposure highlights the power of these cameras and types of things they're tracking. Not just cars. www.404media.co/flock-expose...

Really fantastic ongoing thread demonstrating effective use patterns. In particular, the plan-critique-replan cycle is a major key to getting meaningful results. Smart context management is also critical (and happening here but more implicitly).

So, uh, does GPT 5.2 Pro exist solely to get to claim a couple of marginal percentage points higher on benchmarks? Because that's the most "don't use this model" pricing I've ever seen.

The people thriving with it, IME, are the ones who have that unfortunately rare combination of both engineering chops and actual communication skills.

Went to a national wildlife refuge, locked my keys in the car, waited an hour for the SO to bring the spare, and made some new friends.

ND filters on the river are basically cheating.

Finally got around to processing all the photos from Petite Le Mans. A few favorite picks from the day:

I'm really looking forward to having a chance to play with this some. A smaller, cheaper model that appears to be on par with SOTA of six months ago is pretty exciting stuff.

On AI & water usage, it looks like all US data center usage (not just AI) ranges from 628M gallons a day (counting evaporation from dam reservoirs used for hydro-power) to 200-275M with power but not dam evaporation, to 50M for cooling alone. So not nothing, but also a lot less than golf courses.

Hello from the Petite Le Mans at Road Atlanta.

for real, though, i think the ridiculous inflatable costumes are a major flaw in the armor that can be heavily exploited, portland protestors are leaning all the way into our reputation and making heroes out of inflatable frogs and villains out of the security forces

Vendors in the infosec space are absolutely insane. Got a ping that somebody from a random vendor is in the lobby of the corporate office -- that I don't work at -- saying they have an appointment with me. Welcome to permablacklist status. lol

New: Apple banned an app that simply archived videos of ICE abuses. Rather than other apps that record ICE official's real-time location, Eyes Up is to "preserve evidence until it can be used in court." Videos from TikTok etc. Every submission manually reviewed www.404media.co/apple-banned...

My inclination, given past historical performance, is that sama committed to targets he's never going to hit and just got fleeced again.