Barry Kelly

Barry Kelly

@barrkel.bsky.social

Software engineer and architect. Zurich, London, Galway. Google, Duco, Delphi.

135 followers140 following1K posts

Yud's buds

Chaddeus Stevens 🇺🇦🇵🇸🌹@inv1nciblesummer.bsky.social

Shouldn’t these nerds love game theory and mechanism design? Both frontier labs are bleating “help, we’re stuck in a Prisoner’s Dilemma!” and Yud and his buds don’t have a 15-year old Lesswrong thread solving this problem they can trot out? I thought they were more imaginative than this

Barry Kelly reposted

It seems like small potatoes but this is reflective of why I don't believe a change in who runs Washington DC will fundamentally change America's position on the world stage

Charlotte Lydia Riley@lottelydia.bsky.social

very very funny to see American progressives not only praising the suspension of the Balogan red card but also attacking “Europeans” who aren’t happy about it, Americans just can’t help Americaning when it comes down to it

Barry Kelly reposted

Replying to @bordenstan.bsky.social

Planners should worry less about the admin that YOLOed into Caracas suddenly deciding to meticulously Desert Storm us, and worry a lot more Garand Thumb or some random chud LARPer livestreaming themselves walking around Coutts in their tactical barbie loadout and getting QTed by Elon.

Barry Kelly reposted

Mona Khalil, Lebanon’s most well-known sea turtle conservationist, died of her wounds today after being injured by an Israeli strike I feel sick

Megaphone News@megaphone.news

Mona Khalil spent over 20 years protecting sea turtles on Mansouri Beach. After she was injured in an Israeli attack, activists continued her initiative at Rmeileh Beach, where they protected the first loggerhead nest of 73 eggs on Thursday. youtu.be/fbBKzK4JNRc

Great thread. Great diagnosis. I don't really buy the cure, and I think the stacks idea is a bit iffy, but a deeper dig into our malaise than just name-dropping McLuhan. Single page here: skywriter.blue/@eliothiggin...

skywriter.bluePage by Eliot Higgins | @eliothiggins.bsky.social🧵 Democracy feels like it's in a rough state at the moment across the globe, and we hear various explanations, like polarisation, extremism, disinformation, and loss of trust. But what if those explan...
Eliot Higgins@eliothiggins.bsky.social

🧵 Democracy feels like it's in a rough state at the moment across the globe, and we hear various explanations, like polarisation, extremism, disinformation, and loss of trust. But what if those explanations are mainly symptoms and we've been trying to treat them rather than the underlying causes?

I like this idea.

antirez@antirez.bsky.social

One of the most powerful automatic coding (autocoding) trick that almost nobody uses: "Create this new project as specified in SPEC.md using as guide for coding style, design sensibility, comments, ..., the code at /foo/bar/". Style transfer is very powerful.

Barry Kelly reposted

If you’re wondering what the big deal is about Jürgen Habermas (who died yesterday), or if you’ve never heard of him until now, you’re not going to learn much from the short newspaper obituaries, which are focusing on more on what he said about current events than his philosophy. 1/

Barry Kelly reposted

INBOX: Washington Post Ukraine Correspondent Lizzie Johnson announces that she has been laid off in the middle of a below-freezing war zone without power, heat, or running water.

These comms from Mandleson are weird. He describes Epstein as his life advisor, yet the information goes the other way: Mandleson filling Epstein in on his proposed scheme. Curious.

Dan Neidle@danneidle.bsky.social

NEW: Previously unreported Epstein emails show Peter Mandelson discussing a Panama-linked tax avoidance structure for a £750k Rio apartment - with Epstein as his "chief life adviser". Exclusive Tax Policy Associates report:

Great thread. The writing looks like it's AI enhanced, but the core ideas and consequences are sound, and it's a concise recitation.

Eliot Higgins@eliothiggins.bsky.social

🧵 @jburnmurdoch.ft.com is spot on about the conditions in his FT piece. Liberal democracy held it together thanks to growth, good demographics, and the promise of a better future. Those days are gone, and that’s the "why" behind the erosion. However...

Barry Kelly reposted

Replying to @pbump.com

If I walk up to you with hammer and say “I could break your knees if you don’t give me $20 but I won’t,” would you feel relieved that I’d ruled out breaking your knees?

Barry Kelly reposted

The incredibly lukewarm Democratic response on Greenland is being noticed in Europe, incidentally. Chuck Schumer not finding a stronger word than “quixotic” for Trump’s plan to seize Greenland has been name checked to me more than once.

Alexander Clarkson@aphclarkson.bsky.social

But Americans who want to save the alliance with Europe must have no illusions about how bad a condition this transatlantic partnership is in.

It seems clear that today's AI agent CLIs aren't the future of software engineering. We need to persist the plans, the dialog, record the decisions, and we need to integrate the flow with code review and link it back to product requirements. They're great for ad hoc, but we need more structure. 1/2

I want a single UI that gives me a task queue and interrupt driven management of a farm of agents asking for feedback / permission / code review, with every flow recorded and associated with task + design + plan. A cacophony of tmux panes is not it, sorry. 2/2

Expanding on this: MoE models, once they route to one path (like tool use for search), architecturally remove the influence of broad knowledge in responses. Models will be trained to trust the results of a search, for grounding and avoiding hallucination. Careful questions are important as ever.

Barry Kelly@barrkel.bsky.social

I'm seeing an interesting problem start to crop up with GPT 5.2. It seems like the model has less general knowledge. It's trained to do a web search instead, and it then synthesizes a response based on the results. But the results are biased by the search query! They may omit important context. 1/3

I'm seeing an interesting problem start to crop up with GPT 5.2. It seems like the model has less general knowledge. It's trained to do a web search instead, and it then synthesizes a response based on the results. But the results are biased by the search query! They may omit important context. 1/3

Just now, I had it with asking about funeral scheduling in Germany. The model decided to do a search and came back with results relating to disposal of bodies (which is legally regulated), so it assumed funeral scheduling was constrained by these regulations. But a funeral isn't a burial! 2/3

It was only after clarifying the notion of urn burial (which was a surprise), and then suggesting it revise its first answer based on the idea of a funeral as a communal event to mourn the deceased person, rather than the logistical disposal of a corpse. The model lacks broad knowledge. 3/3

Barry Kelly reposted

Replying to @jessicaschulberg.bsky.social

the victim, identified as 37 yo Renee Nicole Good, "was one of the kindest people I’ve ever known," her mother told the Star Tribune. She leaves behind a young child whose father died in 2023. “There’s nobody else in his life,” the boy's grandfather said www.startribune.com/she-was-an-a...

startribune.com‘She was an amazing human being’: Mother identifies woman shot, killed by ICE agentRenee Nicole Good, 37, lived in the Twin Cities with her partner.

I too had thought rationals were named after numbers that could be expressed as ratios, or fractions, but it isn't the etymology. Rationals are essentially named because they are the opposite of illogical irrationals, and ratio comes from rational, not the other way around.

Barry Kelly reposted

1. Look, there are basically just two ways to run the global oil market: petro-imperialism and petro-consumerism. Historically USA has embraced both, varying over time. The attack on #Venezuela is a return to naked petro-imperialism. Thread.

Thief II then, because I had played Thief and it changed my perspective permanently and I didn't know about Deus Ex until 2001. Now it would be Deus Ex, because of how revolutionary it was. But if I want to replay either game, it would be Thief II. Really, it's just a better, more focused Thief I.

Julianna 💎@batariangal.eurosky.social

Quote this with your favourite game from 2000 NOW, and what your favourite was THEN (if you had one) Localisations count, remasters/ports don't. Remakes depend. Two-way ties count. For a fuller refresher: backloggd.com/games/lib/po...

I tried Cursor last night, first time in a long while. I found the agent modes close to unusable; it races ahead and make edits to multiple files without confirmation, and it gets stuff wrong right from the first or second edit. Surprisingly poor if you want to retain intellectual control over code.

Barry Kelly reposted

Replying to @michaelpettis.bsky.social

2/6 This is the point I have been making again and again over the years. The global economy is a closed system, and it must balance. This means that domestic imbalances created by countries that control their external accounts must...

There's a problem with LLM problem solving that I'm seeing pretty consistently now. LLMs, when they encounter an error, formulate a hypothesis about what's wrong. Then, rather than validate the hypothesis, they immediately try to implement a solution based on the hypothesis. Check the hypothesis!

This isn't really accurate but understanding why is quite subtle. Mimicking patterns of words relies on approximate knowledge. LLMs have approximate knowledge of lots of things. This means they're usually right on the gist of something but wrong on specifics. 1/

Katie Mack@astrokatie.com

Chatbots — LLMs — do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they’re “right” it’s because correct things are often written down, so those patterns are frequent. That’s all.

View full thread

The more interesting question is if humans know facts. Humans weren't designed to accurately answer factual questions. They're designed to adapt to their environment, build a model of the world that predicts enough to survive and thrive. Are those models factual? Not necessarily. Can be cultural. 3/

This isn't really accurate but understanding why is quite subtle. Mimicking patterns of words relies on approximate knowledge. LLMs have approximate knowledge of lots of things. This means they're usually right on the gist of something but wrong on specifics. 1/

Katie Mack@astrokatie.com

Chatbots — LLMs — do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they’re “right” it’s because correct things are often written down, so those patterns are frequent. That’s all.

In this they're not too different to people about things they're not expert in. Do LLMs encode the fact that the sky is blue? They do, and not as a language pattern. They necessarily encode abstract ideas because the statistical models they build need them to fit it all in. 2/

Barry Kelly reposted

Nathan Gill, Reform's leader in Wales at the last Welsh parliament election, has pled guilty to 8 bribery charges, taking Russian cash while he was one of Nigel Farage’s MEPs to speak in favour of Putin's Russia in that parliament www.theguardian.com/uk-news/2025...

theguardian.comReform UK’s ex-leader in Wales Nathan Gill pleads guilty to bribery chargesGill admits to eight counts of bribery relating to pro-Russia statements made in the European parliament and articles

Barry Kelly reposted

Replying to @aphclarkson.bsky.social

The "Kirk was civil" lady from the Atlantic has been stuck in my head this morning. So much of this is about existing elites focusing on maintaining elite status in the new order. The Zyuganovication of American liberalism.

I'm kind of disappointed in GPT5. I had a tough problem recently. zfsbootmenu handoff to Ubuntu, failure for framebuffer to initialize, making it harder to configure my new rig. 4-5 minute waits for advice that wasn't that useful. Gemini Pro 2.5 was significantly more useful, and faster.

Imagine if we got this 20 years ago, something able to understand and write language, that fits on less than half a CD, but requires monstrous amounts of computation. It would be like alien technology, practically beyond our comprehension.

Simon Willison@simonwillison.net

The new Gemma 3 270M open weights model from Google is really fun - it's absolutely tiny, just a 241MB download I asked it for an SVG of a pelican riding a bicycle and it wrote me a delightful little poem instead simonwillison.net/2025/Aug/14/...

Barry Kelly reposted

Replying to @eliothiggins.bsky.social

Misinformation = false or misleading information spread by people who believe it’s true. Malinformation = true information used in misleading or harmful ways.

Barry Kelly reposted

Replying to @eliothiggins.bsky.social

Media scholar Clare Wardle and colleagues developed a helpful typology: Disinformation = false or misleading information spread deliberately to deceive or manipulate.

This is super interesting stuff. Adds quite a bit of nuance on top of the simplistic Economics 101 model of thinking scarcity means it's fine to let prices rise. It can actually be an opportunity to collude with competitors and raise prices!

Isabella M. Weber@isabellamweber.bsky.social

Our qualitative analysis confirms implicit pricing coordination through cost shocks. Example: "We're very aware of what's going on in the industry, and you're seeing some fairly aggressive moves on pricing. I think there is an ability to price and permission to price." 7/9

There's something I'm starting to get sick of in LLM prose. It's not flattery - it's antithesis. It's starting out not with the point, but instead its opposite. The forced contrast is not effective, it's overused. And I don't think I'm wrong to see it as a tell.

Barry Kelly reposted

Replying to @dot1031.bsky.social

I just had a vision of that Oval Office ambush when Tr*mp accused Zelenskyy of having "no cards". Zelenskyy must have sat there smiling inside, thinking, "You've no fucking idea what cards I've got you dumb moron".

Barry Kelly reposted