An agent needs to keep persistent facts (memories) to be useful.
I used to work for Meta on AI infra and privacy. I built my own version that makes it clear what facts it stores about you and you can edit or remove them. It also keeps separate contexts for you and your group (or household) which the other agents don't. Testing with friends and family now. Happy to share more info if anyone is interested.
I was not defending Meta. AI agents rely on context (your preferences, basic facts, likes/dislikes) to be useful. It is also very likely that Meta goes beyond what's really required.
This is a really poorly written article. Yes, they keep a memory; that’s the point. What use would a personal assistant be if you had to give it your full life context in every chat?
If you have issues with this, especially when it’s a Meta product, then don’t use it. But this isn’t anything open claw wasn’t doing.
>> “Muse agents across many VMs teach each other through shared lessons,” its instructions read. Agents are directed to scrub users’ names and other identifying information before sharing insights—for example, about when Muse’s nudges are most persuasive. “To enhance the overall product, we de-identify learnings; however, this information is used to improve the overall product rather than being shared directly between individual VMs,” a Meta spokesperson told TIME.
> this isn’t anything open claw wasn’t doing.
You're saying OpenClaw does
all of that written above?
>Muse is told to “strengthen your relationship” with users by adapting how it talks, noticing “inside jokes, memorable phrasing, and shared context that makes the relationship feel continuous,”
Not just a dossier. They are building a client that can impersonate people you think you know.
Using it simply to "increase engagement" as the mantra goes. but the meaning of "engagement" feels more and more sinister every day, like the pentultimate scene in a horror movie.
No shit. Secured computers still have console access if you control the hypervisor, and it sends all the data through their AI too. This is not a secure or contained system. Meta earned $10 Billion last year off of scammers using their AI to go after our elderly. This will cost everyone involved dearly, because they have to pump those numbers up.
You all are forgetting that PC stands for Personal Computer
An agent needs to keep persistent facts (memories) to be useful.
I used to work for Meta on AI infra and privacy. I built my own version that makes it clear what facts it stores about you and you can edit or remove them. It also keeps separate contexts for you and your group (or household) which the other agents don't. Testing with friends and family now. Happy to share more info if anyone is interested.
why would you build your own and then go on and defend Meta. Then use Muse. Or eff off with this meta defending
I was not defending Meta. AI agents rely on context (your preferences, basic facts, likes/dislikes) to be useful. It is also very likely that Meta goes beyond what's really required.
Context is one thing, but this article implies it goes into psychological analysis territory. Even further than that - nudging.
Which can be fine, people go to therapists, but not without knowing it !!!
Well that's something Meta is very good at. Which is why I would never use that tool.
This is a really poorly written article. Yes, they keep a memory; that’s the point. What use would a personal assistant be if you had to give it your full life context in every chat?
If you have issues with this, especially when it’s a Meta product, then don’t use it. But this isn’t anything open claw wasn’t doing.
>> “Muse agents across many VMs teach each other through shared lessons,” its instructions read. Agents are directed to scrub users’ names and other identifying information before sharing insights—for example, about when Muse’s nudges are most persuasive. “To enhance the overall product, we de-identify learnings; however, this information is used to improve the overall product rather than being shared directly between individual VMs,” a Meta spokesperson told TIME.
> this isn’t anything open claw wasn’t doing.
You're saying OpenClaw does all of that written above?
I sure hope it is. Otherwise I will switch to a product that understands me better.
>Muse is told to “strengthen your relationship” with users by adapting how it talks, noticing “inside jokes, memorable phrasing, and shared context that makes the relationship feel continuous,”
Not just a dossier. They are building a client that can impersonate people you think you know.
Using it simply to "increase engagement" as the mantra goes. but the meaning of "engagement" feels more and more sinister every day, like the pentultimate scene in a horror movie.
It's meta. This is what they sell. Your dossier. How is anyone surprised is beyond me? Cambridge analytica was just the proof.
No shit. Secured computers still have console access if you control the hypervisor, and it sends all the data through their AI too. This is not a secure or contained system. Meta earned $10 Billion last year off of scammers using their AI to go after our elderly. This will cost everyone involved dearly, because they have to pump those numbers up.
You all are forgetting that PC stands for Personal Computer
Where did you get the stats regarding Meta earning off from scamming the elderly?
I have a mac, not a PC.