But agents don't _know_ why they gave an answer. They can only give "reasoning" that links to something in their context, and even then, you would have to parse out their response with some heuristics to try to match against something upstream of that turn.
I could see it being done, and if you're fond of the "models all the way down" mode of thinking, you could use a smaller model to identify it, but it could just as well be a "load-bearing seam" (ha) for something else in the conversation.
I’ve struggled with this line of thought. In a strict sense, it’s true; an AI model doesn’t actually know why it fired the way it did. But I don’t know how practically useful that observation is because, using that strict definition of “know”, humans also do not know why they do what they do.
If I asked you why you posted this comment, I suspect you’d be able to articulate a good reason why from your memories. But the firing of your subconscious and the entirety of your lived experience that molded the state in your brain that lead to you pressing that reply button and writing those exact words are inaccessible to you.
It reminds me of the timeline exercise in SRE work. I always start with a timeline and plunk pagers, metrics, slack messages, etc onto that timeline first. Then I go to individual people to get their understanding of the incident and get that into the timeline. People’s understanding of the incident and why they did what they did rarely matches the timeline.
When I ask an LLM to review its reasoning over prior conversation turns, it has access to its tool call results, web searches, and reasoning traces that aren't visible to me in the UI.
There's definitely value in the "meta-cognition" that an LLM can do even if it is blind to the underlying weights and layers of the model itself.
I think we need to distinguish between "what the model received" and "why the model generated this answer." ThoughtDAG currently focuses on the former: accurately displaying the context of the incoming request and allowing users to modify it.
Sometimes it’s just an awkward turn of phrase on my part that creates a wrinkle in the conversation. Sometimes agents identify that, and we can work together and direct that, but correction itself eventually loses competition to the original error.
Most LLM tools (Claude Web, OpenAI, and their harness) offer re-editable questions. That is how I avoid such problems by myself.
In ThoughtDAG, you can re-edit questions by double-clicking the question. Or edit the answer by clicking the edit icon at the end of each answer text. Or.. you can just remove the connection or delete the node.
Incentives to optimize cache-usage are only aligned between anthropic and you, dear user, when there is not enough compute available to serve the tokens fresh. Apparently enough compute is now available that frugal usage is no longer a requirement and user cost/profit maximization is now the name of the game. I would expect to see more and more "cache-busting" strategies implemented in order to both eek out marginal performance increases and to heavily increase profit.
Listen to the video again, he clarifies early on that this is supporting multiple racks. A megawatt is 1000kW. If you took 5 B300 systems (10 rack units each), you'd still only be at 75kW, which if you're doing redundancy properly, means you have wired 150kW to the rack, but you're not generating a megawatt of heat.
NVidia has been talking about 800VDC 1MW racks for about a year now.[1] They're trying to get others to solve the power and cooling problems. The Schneider Electric view: [2]
The view from Flex.[3]
This was being talked about more in 2025 than it is now. Nobody seems to have deployed a 1MW rack yet. There are major problems to be solved for that level of power density. The fire protection people want racks in fireproof enclosures with a built in inert gas extinguishing system and smoke detectors that pull in air from many points for very fast detection. The cooling people talk about 400 gallons per minute of chilled glycol. The power people talk about arc flash protection and worker safety.
This is a scary level of energy density. Not hopeless; the switchgear and semiconductor inside locomotives are at comparable levels.
> The power people talk about arc flash protection and worker safety.
800VDC is insanely scary stuff.
> At present, no consensus standard provides a validated methodology for calculating direct current (dc) arc-flash incident energy.
This means that establishing proper PPE for DC arc flashes is not possible right now. If I was asked to provide electricians to work on these racks, I would only allow my electricians to approach them, let alone do any work, if they were de-energized, full stop. Even inspecting anything while they’re energized (without doing any work) would be insane, in my opinion. Nobody’s life is worth losing over a rack of servers, and the incident energy for a 480V AC circuit that can power a 1 MW rack (1503 amps) can exceed 40 cal per cm^2 depending on the electrical system already, and there is no PPE that protects you above that incident energy level.
Subway and trolley traction power is at those levels. SF Muni overhead is 600VDC. BART is 1000VDC. Caltrain is 25,000VAC, because it was wired for high speed rail.
I think the sentiment is usually paired with discussion about those products as long-lasting, revenue-generating things. Many of those ended up feeding back into Search and Ads. As an exercise, out of the list you described, how many of those are meaningfully-revenue-generating, without ads?
A phrasing I've heard is "Google regularly kills billion-dollar businesses because that doesn't move the needle compared to an extra 1% of revenue on ads."
And, to be super pedantic about it, Android and YouTube were not products that Google built but acquired.
They bought YouTube but you have to give Google a hell of a lot of credit for turning it into what it is today. Taking ownership of YouTube at the time was seen by many as taking ownership of an endless string of copyright lawsuits, suing them into oblivion.
Youtube maintains an independent campus from the google/alphabet mothership, I'm curious how much direction they get, as (outwardly, at least) appear to run semi-autonomously.
Before Google touched Android it was a cool concept but not what we think of today. Apparently it didn't even run on Linux. That concept came after the acquisition.
I think I can see some of where this could be utilized, but I think I'm still missing a step and I'm hopeful someone can fill me in.
There's a comparison against Argo Workflows, but with the description here and in other comments, Koreo seems to be aiming more for what I would use Argo CD for - managing the entire state of the cluster, the controllers, configuration, etc. Because of it tying into repos, you can then define the entire state of your cluster in code, and Argo CD has tools for doing some of the interpolation of variables into your YAML.
The project looks cool, and I don't think that the world suffers from having multiple ways of doing something, I just want to understand it better.
My understanding of Argo CD is limited so forgive me but I believe Koreo is a bit more similar to Argo Workflows. Where the scripts and logic is baked into the Argo Workflow image, in Koreo it leverages other K8s controllers to do the heavy lifting of interacting with the APIs. If I understand Argo CD correctly, it is an implementation of Argo Workflows. In this way, you are able to build your own CI/CD workflow via Koreo if you so wanted.
Thanks for sharing! I was thinking how this is a good example of the kinds of things that you can do with a game engine that already exists, as opposed to having to write everything yourself.
Also a nice mention of octrees' use outside of voxels.
This is beautiful, genuinely (the SVGs are lovely bits of artwork) and it's a fun puzzle to read, and I've shared with friends that like esolangs, including your book (good luck on publishing!).
In the "Data Strands" -> "Value Strands" section, you describe it as "Value strands (and other data strands), begin with a hook that points up (as in the third strand below) or to the left (as in the first two)" for the following example:
1 ╵╰──╮╭──╯╶╮
2 ─┘└─ └─╮
3
5 ╷
but the way I was parsing it was that the hook is '╰' or '╯', in which case both of those are hooks pointing up? It looks like a fairly innocuous typo, but I'm never sure with an esolang so wanted to ask.
Yes, that's a typo: the first two strands have hooks pointing up, the third to the left.
I forgot to mention that you can run the interpreter with -p to convert the program to pseudo-code. This makes it much easier to tweak the examples and experiment. I'll add that to the readme, along with more pseudo-code for the example programs.
And thanks so much!! Very excited to publish (Sept 2025).
Thanks, as I said, easy to fool yourself into thinking you're wrong when esolangs are involved.
The CLI tool is fantastic. I never thought that I'd see an esolang where "ergonomics of using it" was something the author would work on. I'll play around with that some more later, thanks again!
I appreciate that the article is trying to describe a complex topic, and many of the issues described are shared by us here in the US. It does cite different statistics of other nations as comparisons, and calls out that access to jobs, ease of movement, and housing are all things that help make countries wealthy.
Now for the downsides.
I feel like this article maintains a tone throughout that I might describe as "dismissive nationalism." The comparison against France is preceded by a paragraph that, due to the framing, is presented as ludicrous, proceeds to describe all of the things France does better, and then somehow still makes the claim that Britain has "many advantages." Perhaps that's intended in a vacuum, but with how hard it's pushing for private investment (to the extent that it touts all that earlier infrastructure in Britain as "built by a total of 1,116 private companies" and that it describes the State's role as functionally just exercising eminent domain), it's hard to overlook.
And while the underlying paper is discussing foundational, long-term effects, that there isn't a single mention of Brexit in the article feels stark in contrast to the conclusory statement by the paper's authors of "it is vital that Britain should once more have the strength and self-belief of 1939 with which to play its part in the leadership of the free world."
---
I'm not an expert in economics (as my critique above might have clued you in), but I do think it's important to be careful in which experts you choose to listen to. I also don't know the biases that this particular publication or the underlying paper subscribe to.
I may try read the Foundations paper (https://ukfoundations.co/ - it's not that long), but if this article is summarizing well and is all, then it feels "too shallow." What's the underlying mechanism that got them to that state? If the proposed solution is "allowing more private investment," then what about all of the social shifts that happened in the meantime? NIMBYism is brought up as "logical" given the incentives at one point, but I'd argue it's fundamentally a social issue, related to a perceived quality of life and a lack of perception of a larger group as being part of "your community."
I digress.
Thanks for sharing the link, and I hope others with more of an economic / politics background might have more fruitful critique.
> What's the underlying mechanism that got them to that state?
It's the people behind the ukfoundations.co site who got Britain to its current state. The site is literally the work of Adam Smith Institute members whose Thatcherite neoliberal policies have dominated the UK since the 70s.
Brexit isn't mentioned because it's not relevant - the perception that it was important economically is a hangover from the times when the establishment was telling everyone it would be. In fact there was no economic impact. Britain continued in about the same place it was before rankings-wise, there was no Brexit recession and even trade with the EU continued on its prior trend line! An outcome so embarrassing to the Remainer classes that they just ignore the real outcomes and talk as if the economist's models were accurate.
> What's the underlying mechanism that got them to that state?
The Foundations essay discusses that. It's a combination of a big overhang of very socialist regulation from the 1940s, especially related to construction, combined with the lack of a big enough libertarian faction in the right wing party that would fix it. America has a strong Republican movement that pushes for pro-capitalist, pro-freedom outcomes (or did pre Trump), and inter-state competition makes it easier to compare outcomes. The UK had such a government for a while in the 1980s but British society didn't produce enough Thatchers to keep it up and the reforms were never completed. Blair accepted the 80s changes were necessary but didn't continue, and after Blair stepped down Labour shifted to the left (reverting to its historical mean). Cameron took advantage by steering the Conservative Party towards being continuity New Labour i.e. a center left party. They found that maintaining and doubling down on super strict building regs satisfied both the left wing greens and the right wing pensioners/house flippers, making it into a winning electoral strategy. They didn't have enough people making the long term arguments about the economic damage of that strategy, and indeed the Conservatives set things up to suppress their own right wing in the belief that this would yield bigger victories. Everything continued to drift ever more listlessly to the left until their activists, funders and ultimately voters basically gave up on them. Labour won more or less by default (not because they became more popular) and is now acting much like they did in the 1970s, so it looks like history travels in a loop and things will get worse for Britain before they get better, unfortunately.
> NIMBYism is brought up as "logical" given the incentives at one point, but I'd argue it's fundamentally a social issue
It's mostly not a social issue but financial. The USA has unusually sophisticated investing culture, it's common for people to understand and engage in stock trading there. In the UK it's normal for people's only investment to be their houses, and the social expectation that house prices always rise is abnormally strong. Therefore, anything that could lower house prices became political death -> NIMBYism.
For anyone else coming along reading this thread, in searching for Brexit's impact, I see mention that while a sharp recession and many of the doom-and-gloom style predictions didn't happen, it _absolutely_ has had a negative impact on the economy, and that it's expected that the effects are long-term and will continue.
I do think it's potentially fair to say that some of those effects aren't relevant for this particular article or discussion, due to the recency. The movement around it didn't arise out of nothing, after all.
Can you show me the data showing negative effects post-implementation?
The expectation of long term effects comes, as far as I know, only from the same people who were wrong last time and are trying to save face by making new predictions with effectively unfalsifiable timelines.
I realize it may not seem that way, but even under Trump the Republicans are still quite a bit stronger on the pro-business pro-liberty agenda than political parties in other countries. Also Trump only really matters at the federal level, whereas a lot of the decisions that matter for this UK vs USA discussion take place at the local level. It's more about the councillors and governors than the President or the Prime Minister. That's where permitting decisions are taking place for instance.
Echoing a similar sentiment, so supporting legacy retro video game sprites are very important to ensure are in and never change, but including flags is not okay? I re-read the justification for it recently and it still doesn't hold water, because it felt like it boiled down to "It's hard."
I was thinking about how Minecraft has a system of components and layers that let you compose various flags on their banners. Obviously that's far, far simpler than country (and autonomous region, and county, and province, and and and) flags that can include text, symbols, and practically entire images. But I did wonder if there was some way that could be represented. Unfortunately, I'm not nearly well-versed enough in code points and their ilk to propose anything useful.
But, I am torn. Archival projects are important, too, and language evolves. These decisions will live for potentially hundreds or thousands (Linear A) of years, and interoperability in computing is important.
Flags are in Unicode, they are just encoded like "the flag of United States" instead of "seven horizontal stripes and a specific arrangement of 50 white stars in a top-left blue background". (And the flag of US changes, although it hadn't been for many decades.)
In that case, the post clearly identifies the problem:
> Identities are fluid and unstoppable which makes mapping them to a formal unchanging universal character set incompatible.
If you really want to have identity flags encoded in spite of that, you don't really need Unicode's blessing. The pride flag is already not a single character anyway, it's U+1F3F3 U+FE0F U+200D U+1F308 (white flag -- emoji force -- ZWJ -- rainbow) and you can always create new ZWJ sequences with your own font. Or you can make a font that automatically synthesizes flags from some ZWJ sequence pattern, which is no longer semantically valid but should be much more flexible. Once they got sufficiently popular, there is no other reason that your new ZWJ sequence(s) shouldn't go into Unicode per se.
Unicode's decision to not process non-country flag emoji proposals is because they are closely tied with (minority) groups and Unicode wasn't expected to do any resulting conflict resolution. If you can somehow resolve that problem in advance, then you should probably do that first and propose what you've done.
Cool project! Graphics programming is _hard_ and anything to make it easier is welcome.
Maybe a dumb question, but why not Dear ImGui (https://github.com/ocornut/imgui). "It's way too big and complex" is a completely reasonable answer, but I found it fantastic for debug menus, and there are a few applications that have used it as their _main_ GUI (Ship of Harkinian as an example).
I originally wanted to port dear imgui to fenster but that looked difficult and microui already had a very straightforward little bytecode to implement
I could see it being done, and if you're fond of the "models all the way down" mode of thinking, you could use a smaller model to identify it, but it could just as well be a "load-bearing seam" (ha) for something else in the conversation.