This is a cool idea, very well put through for everyone to understand such an esoteric concept.
However I wonder if the core idea itself is useful or not in practice. With modern memory there are two main aspects it makes worse. First is cost, it needs to double the memory used for the same compute. With memory costs already soaring this is not good. Then the other main issue of throughout, haven’t put enough thought into that yet but feels like it requires more orchestration and increases costs there too.
Building robots at that scale without any indication that the market wants it is weird. I wouldn’t want to say atupid because with musk there is no rational thought. However this is not cars where the concept exists and we know people spend 100k towards a car. We don’t know if people will even spend on a robot that doesn’t do shit. Figure is looking at 100-150k robot if built at scale, so u less they revised this estimate down drastically, what does a 20k robot do?
Well from this article I got the feeling the intended customer is industrial, not domestic. There's a lot of talk about how much a robot can lift versus hydraulic systems.
But I do also get the feeling that maybe Musk is just off his rocker and everyone else is copying what he does just in case he actually a genius
There are plenty of positions were you have normal humans doing only a handful of tasks.
Checkout youtube on some chinese factories building like rice cooker and co. They have like 10-50 stops were one person only does like 1-5 things. Putting tape on, screwing something together etc.
I can see it as the last niche were the real big specialised and for purpose build robots are just not economicly
Musk has a very spiky character sheet. He is, in some dimensions, extraordinarily stupid, and I believe his ego makes a lot of big decisions. But something that might fall into the genius category is this: building things speculatively, primarily for the capabilities that you anticipate developing along the way, the nature of which are not yet known. But this increases your odds of having capabilities in the future that others lack, which looks a lot like a venture capital oeuvre.
To condense that, i might use a phrase like "blind-buying future option space"
Whether Musk deserves that credit is a moot point. I haven't trusted a thing he's said for years, and studying him for revealed intent can't get past "clown on drugs" without violating occam's razor.
They’re meant to clean your house, do your dishes, do your laundry, some gardening probably mowing your lawn and weeding. The possibilities are endless. It would 100% be worth at least $50,000 if it could do those things even 85% of the time.
The industry has a lot more money and easier use cases.
A robot like Optimus will not be a household robot for years to come. Why? If it falls, it will crash into some kind of glas from doors to windows etc. If it falls it might crash a human or animal underneath it. It might trip on a toy or stairs and crash into a wall.
I would love to have one robot but 50k? Who buys something for 50k? A normal person has to save up for a car and they need a car, for a household robot you need a lot of income to justify 50k. You will buy a car, flat, kitchen, etc. before you will buy a 50k robot.
10k perhaps is more realistic but than it has to be good. Like if you are alone, I don't think you will recognize normal housework as such a bad thing that you will buy a robot for a small flat.
For families, the robot has to be very good and really save.
If you have a partner not working, you might not be able to afford a robot and that perosn has time anyway to do all of that.
I can imagine having a robot for elder people and some remote service using these robots to do stuff for them but 50k is costly.
I'm not bullish on household robots for the next 10 years at all. Now you have another problem though, if they become really good in an industry setting, guess who will lose their jobs? yeah exactly the people whou should be able to buy these.
There's zero chance I'll let a robot into my home to do my dishes when I know it'll be filming and audio recording 24/7 and sending that data directly back to Elon who will mine the feed and sell that data or use it against me every chance he gets. It's so much worse that the idiots who bug their own homes with "smart" assistants and ring cameras. You'd have to be crazy to let someone else's drone into your house.
Give me a similar device that I own and control, one without access to the internet, and I'll be happy to let a robot do my dishes, but otherwise such robots are strictly for fools (and sadly there are plenty of those).
An autonomous robot that I program, I update (or don't) as and when I see fit, and does not need to connect to the internet or to anything, for $20k that does dishes, or helps me lift things at the shop, and returning data to it's maker IF AND ONLY IF, AND WHEN, I CHOOSE (or dont)? Great - take my money!
An ambulatory machine with eyes, ears, touch sensors, continually watching, listening, observing, mapping, recording everything it encounters in my home and/or shop and sending all that data back to it's manufacturer "for improved user experience"? HELL, NO!
The latter, supplied by Musk, even if he's paying me $1 million per year to 'host' it? I'll buy the equipment to destroy it as soon as it comes onto my property.
I don't think there is any supplier I could even begin to trust when they require a connection. Can anyone here think of one?
I don’t see the allure here. It’s nothing but an ai assistant that’s already on every phone. The interaction angle is not clean at all. I hear Apple is going to release something similar. Very curious to see how that goes, I am extremely pessimistic on that one. Let’s see
I wonder if the message of eff doesn’t resonate with the younger generation who did not see the OS wars first hand and instead always saw Microsoft as a cloud provider and Apple and Google as the OS providers.
This is actually really nice from anthropic. They are aggressively owning the entire development stack for every swe. They become the default development platform. Automatic recurring revenue too and I am sure they will come up with more categories of subscriptions too.
Meta is in a weird spot. They caught up late to the game and instead of releasing llama as a chat bot they open sourced it, precisely because they lost the mind share. They thought chatbot is not their product and I am sure they are regretting it now. Mark is obsessed with becoming the android of something and he poured billions into the metaverse thinking he is first and failed. He then open sourced llama and wanted to be the android of llms. He ended up enabling groq but it didn’t benefit meta directly at all. They have no revenue or mind share path from llms but continue to pour billions into it. The only 1-1 mapping is with the glasses but that is a tough fit for the company given they are extremely allergic to privqcy and security.
> He then open sourced llama and wanted to be the android of llms.
Well the original llama did kick off the era of open source LLMs. Most original open source LLMs were based on the llama architecture. And look where we are now OSS modles are very close to frontier.
It may not have benefitted Meta but it commoditizatised LLMs.
The llama weights were leaked. It open sourced itself.
You are right though. Meta could have been in lockstep releasing ChatGPT features into some chat bot on Facebook.com but instead it seemed like their FAIR arm was hell bent on commoditising this stuff by publishing their research models before the Chinese companies took the lead in that.
It’s hard for me to be mad at FAIR even though I general disagree with the outcomes that Meta produce for their users.
How is this not already common knowledge for existing llms? They are all trained with all the literature available and so this must be standard, no? Is the real danger the agentic infrastructure around this?
yes and it's not hypothetical. the system card describes Mythos stealing creds via /proc and escalating permissions. that's the exact same attack pattern as the litellm supply chain compromise from two weeks ago (fwiknow), except the attacker was a python package, not an AI model. the defense is identical in both cases: the agent process shouldn't have access to /proc/*/environ or ~/.aws/credentials in the first place. doesn't matter if the thing reading your secrets is malware or your own AI: the structural fix is least-privilege at the OS layer, not hoping the model behaves.
So with all the bluster we are able to roll the clock back successfully to pre Obama stage of negotiations? Essentially starting from discussing if Iran should have nuclear capability or not and then adding new stuff like Iran controlling the strait and collecting toll on it. Awesome, so much winning!
Has anyone played with the released versions of Claude and tried to create exploits? I cannot imagine it not being able to craft one if guided, unless the tooling around it doesn’t allow it
This is not over yet and it may just result in an established fee for each shipment through the strait to Iran. We won’t/havent hear from Israel which is the key player here. They just do what they want to do because they know the whole world will look the other way.
However I wonder if the core idea itself is useful or not in practice. With modern memory there are two main aspects it makes worse. First is cost, it needs to double the memory used for the same compute. With memory costs already soaring this is not good. Then the other main issue of throughout, haven’t put enough thought into that yet but feels like it requires more orchestration and increases costs there too.
reply