I'm doubtful about that. To be pedantic if you consider the set 'superhuman intelligence' to be a proper superset of 'human intelligence', then I would expect it to be able to structure its communications to the level the audience could understand. Does that make sense?
Is is more what it feels like talking to someone who doesn't really understand what it is they are talking about. A lot of keywords but no flow, no conceptual linkages from one part to another. Etc.
You can't stop an agent from leaking your information by telling it not to leak your information.
To prevent information from moving between your work domain and your personal domain, or between your communications-with-board domain and communications-with-journalists domain, then the agent for one domain must be physically and deterministically blocked from accessing information in other domains.
In this implementation, Markdown should be considered harmful.
Operator Memory injects `.operator-shared/operator.md` and `.operator-shared/index/.md` directly into your agent's instructions before you even write the first prompt.
So if you clone a repo or review a PR where a bad actor put malicious instructions in these files, now your agent executes those instructions automatically and silently.
It could exfil `.env` and `~/.ssh/`, change `~/.bashrc`, all kinds of dirty deeds.
Agents are pretty good now about not running prompt injections hidden in code and Markdown, but this plugin bypasses all of that, and puts the prompt injection right in the system prompt.
And with higher priority than AGENTS.md and CLAUDE.md.
You dismiss "pattern matching" as some sort of trivial thing, so why weren't humans able to apply the same pattern matching to find and fix these defects?
They often were able. Fifteen bugs claimed by mythos were bugs that were already fixed, which the model plagiarized from git history or the kernel mailing list.
> You dismiss "pattern matching" as some sort of trivial thing
I didn't say that - Greg said it in the talk. You interpreted wrong. However, human are a few orders of magnitude slower than agents. Our context windows is definitely less than 1 million tokens (I believe, heck we don't even know how our brain works)
Human brains don’t have a context window. The brain just constantly reconfigures itself on new input. The AI people call this “continuous learning” and believe it’s one of the most important things current AIs are missing.
We do actually know pretty well how the brain works at the level of neurons. We don't know the higher levels of how all of those interactions work to produce certain behaviors.
We don't even understand how a single neuron works and are still finding out they have way more connections and connection types than we thought they had.
I think it's safe to say we still have a woefully limited understanding of how the brain works.
Cellular neuroscience is a very active research field, and until we actually manage to derive organism behaviours from cellular-level models we won't know that those models are complete enough to do so. We haven't even managed that for C. Elegans, with 959 neurons, let alone humans, with 86 billion.
So we don’t know anything, we can mechanistically dissect any phenomena, but that rarely yields the ability to make interesting or useful predictions, i.e. understanding.
By quoting this segment, I inferred that you agreed with Greg.
If you disagree with Greg, then I apologize for inadvertently criticizing you personally.
My point stands, that we're back to debating some metaphysical understanding of what "real intelligence" is when the real standard should be "does it do a better job than humans at this specific task"?
I don't care if the Waymo isn't "truly" intelligent, it drives better than I do, that's a very good thing.
We're all pattern matchers, and if the machine pattern matcher is can find defects and vulns that human pattern matchers can't, that's also a very good thing.
> I don't care if the Waymo isn't "truly" intelligent, it drives better than I do, that's a very good thing.
The funny thing about Waymos is that they seem amazing driving all on their own everywhere until it rains (as it did where I live for the last two days), then they're suddenly nowhere to be seen, because they can't operate correctly in conditions that I've been driving in for my entire life.
Are you suggesting AI companies should intentionally make the visual representations of their products look overtly terrifying?
And also, if they should look like what they are, what exactly does an AI agent that follows you around in all your devices _look like_? Surely they have to take some creative liberties, and it's hard to blame them for wanting something that looks appealing and friendly.
Would you be satisfied if they made it a floating skull with glowing red eyes that periodically screams vague threats at you?
Presuming you are new-enough here that you didn't realize that jacquesm has been posting on HN for a very, very long time: did you consider clicking on the GP's profile link to answer your own question?
I have seen accounts on here registered 10+ yr that now have LLMs responding for them. And a lot of users on here with accounts that are months to years old? Often not their first account, they may use several to respond with or to try and break up their digital footprint.
I miss o3 in that regard. if GPT-4o led people into virtual romance and over-validation, o3 gave me the same sort of "madness" but with work.
when I tried GPT-5, I was sad mostly because GPT-5 had some added wordiness. fluff, if you will. o3, though, is a black hole you can talk to, and it only gave information back if there was something to give back. that kind of vibe is my dream coworker
The best part about Muse is they think that the Enterprise is going to start buying things from Meta that looks like it belongs in a preschool. Muse and Ray-Ban surveillance glasses, Zuck must live in one hell of an echo chamber!
From the careless people woman we learn that he is unaccustomed to losing at Settlers of Catan because they all let him win.
If you're not much of a board game person this is _wild_ because Catan gets annoying with 1 beginner since you can see how they end up gifting the win to someone else.
So he either knows people are letting him win (and got mad at the newbie who didn't?) or he's too stupid to see them make mistakes or he thinks they're idiots but he keeps playing with them?
I dunno there was a lot of stuff from her that was damning but somehow this is the most damning thing to me, real insight into the man.
Reminds me of the UK prime minister's close protection officer and inner circle all gambling on the election date and getting caught.
What degree of corruption did you witness that this seemed OK?
With any luck he'll change the company name again, this time to "Surveillance Glasses", months before shutting down the whole division due to everyone hating the idea.
I can't speak for parent poster, but it's not the cuteness. It's the "putting a cute face on something that's actually a bit sinister."
Hello Kitty isn't trying to harvest anybody's data or take anybody's job. Sanio makes cute things and they hope that you will exchange money for their goods. That's the whole relationship.
The difference is the failure rate and the corrective action taken when defects are found. If Boeing's software failed at 0.1% of the rate that normal web / commercial software failed, we wouldn't have a functional airline industry.
20 million people wearing facial recognition glasses in public, always-on, networked and geolocated, with a Muse agent monitoring the feed, would 1000x the domestic surveillance capacity of current Flock, Ring, Tesla and CCTV cams.
The Fermi probability that you'll be ingested and inferenced by this network on any given day is incredibly high, even if you never buy a Meta product and never create an Instagram account.
It's hideously dystopian and you cannot opt out.
Maybe that's why the Muse avatar is a cute and cuddly Teletubby.
> the government is paying $0.84/mile for people to ride BART
If BART disappeared tomorrow, how much would the government have to pay to build all the additional roads and parking to support that incremental traffic?
And how much for healthcare for all the incremental pollution and collisions?
> several cents per vehicle mile
It's close to a dollar per mile when you include all the externalized costs.
Incremental property damage, injuries and fatalities, incremental chronic illness from pollution, parking land subsidy, fair market value of incremental land use and rights of way, lost tax revenue from that incremental land use, etc.
In general, this is why semi-private or private transport works so much better. If the end user is getting a bunch more value out of working on the train, they will pay more for the train than the road (if they have to pay the full price of the road vs train each time they use each).
reply