It's an op ed. Hardly a conspiracy.

And I "use" dead bodies for sustenance, without qualifying whether that relationship implies eating or necromancy. If the imprecision is because of a misunderstanding about what an op ed or guest essay is for: it's maybe not a publication's fault if someone misses the memo.

"Money is speech" is a reddit-ism. The argument is closer to "limiting spending can be tantamount to limiting speech."

Should the president of the US (regardless of party) be able to restrict your ability to spend money on a web server because it hosts a forum about humanrights/guns/birthcontrol/etc? Or to run a news channel/blog that focuses on a particular perspective?

If not, little prevents the most endowed campaigns from changing their form factor from ads to propoganda platforms. And buying out news platforms becomes more enticing.

Whatever line you think is reasonable to draw here, would you trust a bad president with it? And would it actually solve the problem?

> Whatever line you think is reasonable to draw here, would you trust a bad president with it?

I don't think this is a good argument, because it's oddly specific. You can't trust a bad President to properly enforce any law. That much should be clear right now. A bad President is a general danger to the country and to the world. Among other powers, the President is the commander-in-chief!

Of course Congress has the power to impeach and remove a bad President, so you don't get a bad President without also having a bad Congress. In other words, we'd have much bigger problems than campaign finance. And I would claim that we get a bad President and a bad Congress precisely because we have no campaign finance limitations anymore.

The President is an example (and actual) enforcement mechanism for the State. Not the argument.

We want to ban independent political expenditure past a certain dollar threshold. Great. How do we do that without torching legitimate civic and commercial participation, and without accidentally entrenching rich participants even more? I'm unconvinced you can do that, and unconvinced that ads are a worthy enough evil to be worth the try, particularly if it requires amending over the First Amendment. ("If" is an operative word.)

Interesting that the distinction between phones and boats was not enough to dissuade whatever point it is that you're trying to make.

From Chief Justice Roberts in Riley v. California: "One of the most notable distinguishing features of modern cell phones is their immense storage capacity. Before cell phones, a search of a person was limited by physical realities and tended as a general matter to constitute only a narrow intrusion on privacy."

My point is: why don't anyone makes a fuss over water searches the same way as airports? It's been like that forever. There are reddit discussions about water searches, but never on HN or any news. To. Me it seems interesting discrepancy.

It's not just resolve, either. The org must recognize the problem in the first place, which isn't a given. Especially in smaller/govt orgs. Often, recognition of the problem is wrongly tied to how difficult the foremost suggestion is to implement.

And there are orgs for whom any suggestion can itself be so encumbered by uncontrolled red tape and social costs that relatively minor changes become a project.

This title is easy to misinterpret. If I understand correctly: Codex now encrypts sub-agent prompts and hides those prompts from the user.

edit: originally was "Codex starts encrypting prompts, uses cyphertext for inference instead"

It seems likely to me this was driven by the `ultra` mode in 5.6, which fans subagents to do work. This mode was previously only available in the web UI (what was previously known as pro?)

It seems possible they trained this by doing full RL rollouts of agents interacting with each other. They likely view these prompts somewhat the same as raw reasoning traces, they don't want people to train directly on them.

I am unsure if this has been confirmed, but there are some signs that the opaque "compaction blob" they return from their dedicated compaction endpoint might not be text at all, rather a latent space representation of the conversation. The fact that OpenAIs compaction seems to be much higher fidelity than a lot of other providers makes me inclined to believe this.

If this is true, it doesn't seem far fetched to infer that they might be applying similar techniques to prompting subagents.

I would be curious to see if this way of spawning subagents (encrypted blob) is used when subagents of a different model type is spawned.

"Latent space representation" I have been waiting for this moment in the evolution of AI. Well, waiting with some trepidation. It seems inevitable that frontier AI's will, at some point, leave behind human-comprehensible representations of language. Purely for functional reasons, it's going to start making sense for AI agents to communicate amongst themselves in much more efficient ways than borrowing the languages of flesh-bag humans as an interface medium.

I Imagine next that programming languages, interfaces and API design starts going this direction next. Being written, expressed and optimized as blobs of high dimensional vector space. As humans we might still be able to understand some abstractions of what our AI's are talking about to each other, but maybe not more so then we understand how different regions of our own brain communicate with each other.

I strongly believe that the future is the other way. New programming languages and environments designed for strong auditability and preventing bugs will dominate. Only bad actors will use latent space representation, and it might even be outlawed. But the bad actors will proliferate underground…

I actually built such a language: https://magarshak.com/U.html

Every time a state tries to regulate, it loses to the states that promote instead. Seems to be a better evolutionary strategy.

is that called rust? that is the only thing i feel safe to let agents vibe code

It has already started:

  zero-graph v1
  origin source-text
  module "hello"
  hash "graph:a7f7e6899a73f3b4"

  node #decl_ad8d9028 Function name:"main" type:"Void" public:true fallible:true
  node #param_4610ae76 Param name:"world" type:"World"
  node #expr_c403020c MethodCall name:"write" type:"Void"
  node #expr_653eeb6e Literal type:"String" value:"hello from zero\n"
  edge #expr_c403020c arg #expr_653eeb6e order:0

https://zerolang.ai/

I've been thinking the same! Haha

Even for like token efficiency it could make sense - like imagine if the representation were more compact

Agents acn already translate languages quite well. It doesn't seem crazy that they could work and think in a model specific language, and then translate back to English or something for the user

It seems like the most efficient method would be for LLMs to communicated by exchanged latent space representations directly. Serial language is a incredibly inefficient way to encode these, a lot like flattening a complex graph into text.

Is it really more efficient? Any token becomes huge in latent space.

That's not really a language. It was just a reporter dumbing down the idea of "vectors in latent space" which eventually became LLMs.

I think you hit the nail on the head here. Having subagent dispatch in the loop for RLVR is something we've already seen in open models, like Kimi K2.5 and later, so it's no great stretch to assume OpenAI are doing it too.

If you keep RL'ing the dispatch then the prompts are likely to keep diverging from the type of prompt a person would write (like CoT becoming increasingly incomprehensible), and that divergence is part of their competitive advantage.

> rather a latent space representation of the conversation

Student/teacher models derived from the same checkpoint convey a lot of latent information through token choice, as in: https://techxplore.com/news/2026-04-ai-chatbot-student-owls....

I wonder if this is something they can take advantage of by training on compaction inside of the RLVR loop?

That’s actually what got me to switch and use Codex sometime beginning this year, the compaction via these encrypted blobs was just waaaay better than Claude. I had short convos with Claude where it would forget something very obvious and important few million tokens into a task, whereas I reached ~1B tokens in some local codex sessions and it was recalling and paying attention to things I mentioned way back at the beginning of the session (and not persisted anywhere else in the repo/md files etc)

> It seems possible they trained this by doing full RL rollouts of agents interacting with each other. They likely view these prompts somewhat the same as raw reasoning traces, they don't want people to train directly on them.

this tracks. anthropic protects these as well iirc.

> I am unsure if this has been confirmed, but there are some signs that the opaque "compaction blob" they return from their dedicated compaction endpoint might not be text at all, rather a latent space representation of the conversation.

probably not a latent (to my knowledge latents aren't really part of the outer loop in ar-transformer inference processes), but maybe non-human-readable reasoning traces as occurs in fable.

> latent space representation of the conversation

and how would you load that back into the model? they are token-in, token-out, plus the KV-cache which is derived from token-in

They are not really token-in token-out per se, they are embedding-in embedding-out.

When operating on text, you embed each token into the LLMs embedding space. You go from a discrete token to a point in embedding space.

Likewise, when processing images, you have a image embedding model which produces a set of embedding vectors representing the contents of the image in the LLMs embedding (latent) space.

This same concept can be extended to compaction. Instead if limiting yourself to discrete tokens, you could generate a set of embedding vectors which represent the contents of the compacted conversation in latent space.

These have the possibility of containing a lot more semantic information per vector, which is why this can be appealing.

A big downside is decreased interpretability. AI safety people are generally fairly opposed to latent space reasoning for example, it can be harder to tell what the model is actually doing and if it is trying to deceive you.

If there is no visible prompt at all, then that is very understandable. The PR issue exposes a real gap though: subagent spawns need a human-readable audit trial, of its goals/intent, its boundaries and scope and limitations, etc; for basic responsible agentic harness functionality.

Hopefully they can add that.

> Hopefully they can add that.

Add? Just make the sub-agents input prompt not encrypted, change "encrypted: true" to "encrypted: false" everywhere and everything continues to work as it used to (simplified, but you get the idea).

They need to fix the regression, not add something new here.

This sounds most logical to me.

It's sort of insane though, you not only have dozens/hundreds of stochastic agents running on your machine, but you cannot even inspect the instructions those agents are working off of?

I've gone in to look at Claude subagent/workflows and sometimes been like "no this was a mistake to spin up" ... Codex users just get to token yolo the encrypted telephone operator instructions+shell from orchestrator to subagents?

>but you cannot even inspect the instructions those agents are working off of?

It makes more sense when you realize they don't want developers to be doing any coding at all. That's what they seem to be moving towards. From product manager to product via AI.

Last stage is moving everyone to their cloud platforms, they deploy everything for you, you don't even get to see the code, just the deployed end product.

Because letting you look at the code would be too dangerous, you could reverse engineer an exploit to another product! Or distill their internet-distilled model!

But don't worry, at least it will be very convenient.

no, last stage is removing you completely from the equation

They’ve already succeeded at that. My time spent manually writing and editing code must be down 99% post Opus 4.7.

You already have an agent freely doing stuff on your machine. Subagents prompts are a weird place to draw a line. It's not like you're reading everything the agent is doing in any case, let's not kid ourselves.

When things go wrong I very much read the session traces to figure out what in my prompt wasn't good/explicit enough, then retry to evaluate if it would have helped.

I was about to do the same with Sol + Ultra, but then discovered this encryption issue that prevents me from doing the same for sub-agents.

> It's not like you're reading everything the agent is doing in any case

Personally I do, these tools aren't mature enough to be used without supervision

> You already have an agent freely doing stuff on your machine

No. Agents run in VMs. Assume anywhere you’re running an agent will be compromised, because eventually, it will be.

The only reason most people haven’t is luck, they didn’t happen to install Axios or Tanstack at a certain time.

I imagine this will be because a decent chunk of the IP in Codex is probably within its prompts, how they're built, and how they're sequenced and orchestrated, rather than in the codebase per se.

We had this discussion a few months ago where we talked about allowing people to choose an AI provider and provide their API key, thinking about enterprises with "preferred" (read: mandated) AI suppliers. We also wanted to offer the kind of very simple pricing that this is one way of enabling. But we realised pretty quickly that this would/could lead to leaking our back end prompts to customers and, although those prompts are only a part of the value add, if you could build a detailed trace of them then you'd be able to relatively easily reverse engineer a lot of what we're doing.

So we quickly dropped that idea.

I'm unable to understand how much value can be in low-definability non deterministic prompts. It feels like the kept the right divinity spell into a chest.

I don’t disagree with your divinity spell comparison but unfortunately there is a lot of value in the prompts because these spells are the “programming languages” of LLMs.

yeah i get it too, i'm just flabbergasted that this is today's market

it reminds me of the pre-vulkan game programming days.. drivers were black boxes, game developpers had to resort to magic tricks to do stuff, until everybody got fed up and wanted some logical ground to operate

It's a brave new world, etc., isn't it?

One does find oneself slightly askance at one's own thinking sometimes, that's for sure.

But I suppose, is it really so different? I mean, back in the day moreso than now, a lot of the valuable IP in any system was in the design and specification of that system - the problems usually solved within the design and specificaion (use X algorithm, etc.) - and the code was "just" the implementation of those solutions.

So perhaps it's more of a regression in some ways: the value is in the specification (the prompt) once again.

Your point about stochastic behaviour is well made though, and there is no way to 100% guarantee or formally verify the behaviour of a system that relies on an underlying technology whose behaviour is fundamentally stochastic.

Further proof that this tech stack is immature and would have needed to bake for a more years.

In an ideal world this would have been public tech like ARPANET or WWW and there would have been 2-3 major iterations (until the equivalent of Claude 7-8) and only then would everyone have tried to build huge businesses on top of it.

I mean, sure, it's sort of usable, but the churn is insane. And we're burning the planet (and probably the economy, too) for it.

Compared to burgers, it's a rounding error

> understand how much value can be in low-definability non deterministic prompts

When was the last time you used an LLM to evaluate how true those last part(s) still are?

I also love how you went from "I'm unable to understand" to "This is surely right", it's a good representation of the software ecosystem at large :)

the trick about agentic systems is definitely how to do the prompting. things like automation and sandboxing are trivial in comparisson. if you generally ask via API model directly you can see what basic answers it actually yields and how fine tuning prompts and refinements to output as well as adversarial prompts etc are important to get relatively solid results.

a lot of expertise of certain domains' workflows is needed to make it functional within that domain. some of this can be yielded via prompting too etc so its also baoance of how much to prompt it vs. how much of it you wanna let it reason over itself. (if you tell it too much i lock it into a path and if you tell too little it will give incomplete results )

Perhaps AI providers should support this natively: the customer supplies the API key but doesn't get access to the transcripts.

I don't know how you'd enforce that unless it was something you could mandate at the level of the API call, and then the API call is rejected if the customer hasn't configured it for "no transcript".

It sort of feels like an area of friction even still.

I was thinking of an API key that was scoped both to a specific customer and a specific service provider (perhaps both have to do something to provision it).

Billing goes to the customer, debug logs etc go to the service provider.

Yeah I thought "wow, some homomorphic encryption* stuff", but then "nah, usual greed".

* https://en.wikipedia.org/wiki/Homomorphic_encryption

The title was fixed like 40 minutes ago, when you come back to old browser tabs you probably want to hit that reload button before leaving a comment ;)

Unnecessarily snide remark for someone just commenting on their interpretation.

At that point 50% of all comments were about the title and it had been updated almost a whole hour before parent made their comment. Sorry for being low on patience.

It's also not the first time Codex started encrypting stuff. Their excellent compaction endpoint has served up a giant encrypted blob since at least five months ago.

Feels strongly like we're in the late-stage-AI-unicorn phase. If this is really their moat then the Chinese companies will win.

> If this is really their moat then the Chinese companies will win.

I already switched to a Chinese provider personally, I don't think the difference between both really justifies the wide gap in pricing

> I don't think the difference between both really justifies the wide gap in pricing

I'd be ecstatic if this was true, but nothing so far comes close to the SOTA models from OpenAI + highest reasoning, but I'd be more than happy to be proven wrong by testing it out myself.

So far, I've tried MiniMax M3, GLM 5.2, Hy3, MiMo-V2.5 (+ Pro), DeepSeek V4 Pro (+ Flash), Gemini 3, Kimi K2.6, GLM 5, all the various Qwen variants and probably a bunch more I forget about, in a wide array of harnesses (Codex, pi, opencode, my own and more), and still nothing seemingly comes close to GPT 5.5 (now 5.6) xhigh for tasks beyond 5-10 minutes of work, they all more or less collapse after a while in my experiments. Although most of those do work well for really tightly scoped tasks.

What specific model are you thinking about here, in case I've missed testing it?

I'm using Kimi 2.7 and GPT 5.5 at home, Opus (4.8 I think) at work and I really don't see much difference honestly.

Sure, Claude might be 90% correct and Kimi only 70% correct but does that matter when 90% isn't enough to make it work autonomously anyways?

My workflow is just strict supervision of what's happening, I also edit the agents file with anything I see the model doing that I don't like.

My sessions are also short, after any task which is completed, I just kill the session and start a new one so I don't think I have more than 15 min sessions unless it's tech discovery.

> Sure, Claude might be 90% correct and Kimi only 70% correct but does that matter when 90% isn't enough to make it work autonomously?

Huuh, what does this mean? GPT models frequently do 100% of what I tell them to do, anything less and I'd see no point in using agents for work at all. Do you tell them stuff then 30% of the cases Kimi goes off and does other things, or what do you mean? The time the agent does something unexpected, I can almost always trace it back to me fucking up something in the user prompt, or the system prompt being wrong somehow, I'd lose my mind if it was only "70% accurate".

> My workflow is just strict supervision of what's happening, I also edit the agents file with anything I see the model doing that I don't like.

Same, including inspecting exactly what the (full verbatim) sent system/user prompts are, which the change we're all discussing here is getting in the way of.

But "Kimi only 70% correct" sounds like it's so bad it's not worth using. In my testing, I didn't find that the model just went out and did other things, but all the providers I tried were way slower than even Sol which is kind of slow to begin with, and it's really inefficient with it's thinking. Tasks that took Sol five minutes could take 15 minutes with Kimi for example, which just feels like such a waste too.

> Huuh, what does this mean? GPT models frequently do 100% of what I tell them to do, anything less and I'd see no point in using agents for work at all.

I never managed to have this experience even with SOTA models, they routinely make architectural mistakes, wrong assumptions and take shortcuts they should not take. Less for sure but they still do it often. I didn't try Fable yet though so can't comment on it.

So based on that, since I have to watch everything they do anyways, why would I pay extra?

> But "Kimi only 70% correct" sounds like it's so bad it's not worth using

If you want an analogy, it's like the numbers of 9s in server availability and since currently I'd rate nothing above 90%, it's zero nines. Since I have to deal with unreliability with every provider, I don't see why it would be worth it to pay more to still deal with it.

> I never managed to have this experience even with SOTA models, they routinely make architectural mistakes, wrong assumptions and take shortcuts they should not take. Less for sure but they still do it often. I didn't try Fable yet though so can't comment on it.

Ah, you let them make architectural decisions? :P That might explain it. Agents for me are more like pair-programming or just "what types the code", all the design and decisions are made by me, so if those are wrong, it's my fault. The agents are just used to implement what I've decided to have implemented, and I can't remember the last time codex did a mistake without correcting itself, or made a wrong assumption or taken shortcuts, unless I explicitly told it something that lead to those things.

> So based on that, since I have to watch everything they do anyways, why would I pay extra?

Personally I pay more to have to fix less later, and for a piece of mind that if I ask it to do X, it doesn't go off and do Y.

Anthropic at least is consistent in their naming. They're all literary genres, ever bigger ones.

OpenAI's are all unnamed or suns, so it's kinda the same picture.

> OpenAI's are all unnamed or suns, so it's kinda the same picture.

Umm, no:

> GPT-1, GPT-2, GPT-3, GPT-3.5, GPT-4, GPT-4 Turbo, GPT-4o, GPT-4o mini, o1-preview, o1-mini, o1, o3-mini, o4-mini, o3, o3-pro, GPT-4.1, GPT-4.1 mini, GPT-4.1 nano, GPT-5, GPT-5.1, GPT-5.2, GPT-5.4, GPT-5.4 mini, GPT-5.4 nano, GPT-5.5, GPT-5.5 Pro, GPT-5.6 Luna, GPT-5.6 Terra, GPT-5.6 Sol

Besides the constant shifting of nouns and adjectives...

Luna is the Moon and Terra is Earth.

Unnamed or celestial bodies then. That was a fine nit to pick.

The rest just aren't names, they're designations at best

... so Anthropic is more consistent with their naming and we're back to square one in this conversation.

I'm sympathetic to Logseq here. My guess is that many things are true but do not sum attractively to an audience right now.

The db probably is the cleaner place for Logseq's note storage, but invites comparison to Obsidian, whose fundamental unit of information is already a document, where Logseq's is more like a bulletpoint. Logseq being open source where Obsidian isn't could maybe blunt the edge of moving to a less-portable format, but a user can also understandably look at the status of two-way sync and see an unsexy "In-progress" and figure that's not good enough. Logseq also carries some baggage wrt instability, and the argument that the db is the pro-stability move unfortunately needs to be proven over time. Overall seems like a rough spot to be in.

I dunno. I like Logseq and wish them a good launch.

Arguing that "at least John is doing [clown thing] in the open" just dilutes whatever leverage John's supporters had against John on that thing.

I find myself unwanting to be on the side of people who willingly give up leverage.

I don't think the comment you are replying to is giving up leverage. It's simply pointing to the OP that what he seems to be upset is not the thing itself, which everyone does, but only that he knows about it or ignores others doing the same.

When I say cancer sucks I'm not ignoring that other things also suck. So when people start trying to put those words into the words of others, they effectively act like cancer fanboys, which is weird and all about them, with no part of originating or relating to the the people they question about stuff they supposedly "ignore".

You analogy makes no sense.

"I complain about some people doing X openly while I ignore others also doing the same X behind the scenes"

is absolutely not equivalent to:

"Complaining that X sucks does not imply I am ignoring that some other unrelated Y also sucks".

> "I complain about some people doing X openly while I ignore others also doing the same X behind the scenes"

Nobody said that they're ignoring anything, hence putting words into the mouth of others part, and claiming that criticism of A implies ignoring B.

All LLMs inherit bias from their training data, and xAI’s argument is that Grok is being steered to counter that bias rather than simply inherit it. You can disagree with whether they succeed, but the act of steering isn’t automatically suspicious when knowing that every major model is steered. The relevant question is whether the steering moves the model closer to truth and neutrality, or just replaces one bias with another. At least with Grok, some of that intent is unusually explicit. I’d rather have multiple competing approaches to steering than a monoculture where every model quietly optimizes for the same idea of acceptable answers.

One similarly egregious UX issue on the latest Android is that pressing buttons in the dropdown tray doesn't give any feedback until the action is complete. I can press the "turn on WiFi" button and receive zero haptics or visual indication that the phone registered my tap, for over a second, and THEN it will decide "okay! let's shake the phone and change button color now".

And if you have a tray button that needs to e.g reach over the network to a HomeAssistant instance that needs to itself reach out to some fuckass IoT vendor server, you may as well not expect any sort of feedback before you close the tray.

That's a good example of bad UX, and, in a way, counter to the article.

It sounds like Android needs a middle state indicator. When your action is in a pending state, the state indicator should not be the same as the previous state. A common pattern is to show a pending animation.

While this animation is running, what should happen if the user presses the button?

I've yet to update to Android 17, but I ended up disabling haptics on 16 because the annoying pop sound feels like a gimmick compared to what haptic feedback used to be.

They're rigorously sourced in that some set of underpaid clerks is tasked with surfacing their best guess as to what material is most relevant to the court, with optional fact checking competing for their time.

https://www.propublica.org/article/supreme-court-errors-are-...

So the sources exist, but whether they're quality is another question.

Your question arguably doesn't have multiple answers. There's one answer: the set of children (discounting different orders of the set).

The question "Who is her child" has multiple answers because it asks you to deliver a single answer.

But wouldn't every one of those multiple answers be the correct one in this case? Like it can say child a or child b or child c (hypothetical) and while there are mutiple answers, each of them is a logically right one for the question "Who is her child?" no? So how do we judge what is the absolute right answer to that? its ambigious when you say child

Zooming out to the original complaint that "A is B" doesn't imply "B is A" in common English, and then further -- to the goal of having an LLM predict tokens that map closely to truth/logic/helpfulness:

I don't think a person speaking plain English in most contexts should be seen as "correct" to answer the question with a non-list answer, even if the question is shaped to expect one, unless there's an established confidence that the shape of the question wasn't made in error.

If someone asked me in real life who "my child" is on stage, and I had multiple children on stage, I would first say that I had multiple children there, rather than choosing one from the set. It would be most helpful for an LLM in my position to do the same, rather than infer that [because Timmy is niam's child, niam's child ought to be Timmy when queried].

Yeah I guess that makes sense and which I was getting at. Even though logically it a non-list answer to that might be "correct" but not as helpful as returning the list answer and clarifying that there are multiple children. And I guess the later is also kind of more intelligent if we think about it even though it doesn't fully confirm to the exact prompt.