the question was: what is the endgame for the stated "second class labs" strategy of distilling their frontier competitors then undercutting them on price?
Yes yes, we all understand the game-theoretic race-to-the-bottom you're describing here. Somehow despite linux being FOSS it still powers most of the important computing in the world. Can you explain how that works despite it being free? Once you understand that case I think you'll understand the game-theory behind how large projects can exist in the absence of traditional IP protection.
the obvious difference is the massive scale of data and compute required to develop and evolve these models, and the costs they impose on those building them.
Smaller budgets, slower improvement, less risk. They're not entitled to profits if that business model isn't sustainable. They're not entitled to a change in IP laws to protect their business model. They're not entitled to growing that fast.
Parses emails or other sources, extracts numbers, correlates different transactions, web search, asks questions, stores notes (regex based, very simple).
The hard part is getting good data, I'm sure that lexus nexus or whoever can get API access to my bank account and all my credit cards, but I can't. Email turned out to be the best way for most of my providers. Managed to avoid 2factor auth so far, but it will suck when I need it.
We've got integrations with major UK banks. Curious if you'd like to use a polished product or be more interested in bank-feed-as-an-API type of use case? What banks do you use?
Credit union atlantic. The possibility of you being able to profitably support small regional credit unions is pretty much 0. Maybe with a general browser use AI and an SMS portal. Also you'd need to bypass anti-bot protections, maybe solve captchas.
In my opinion that is the thing holding back almost all of these accounting products that make consumers lives better. I've solved it because all my cards and bills happen to support email, so I can use that as my source of truth for all my credit cards.
The token is stored on a LUKS encrypted drive on kwallet. You'd have just as much luck getting my password from my firefox installs sqlite DB. This is also how email clients in general work.
Also this isn't really an agent. At least not a long lived one. Each email or transaction gets it's own session, the llm can make a few tool calls but must emit json as the final result. Very very short context lengths, very predictable results.
The bot does not have access to my passwords, there are pre-defined scripts that fetch the data that the run an LLM call for each transaction discovered.
There is a big difference between capturing value and creating value. 20 developers can build and maintain a facebook or reddit clone, so why aren't they kicking the asses of the big companies with the much higher costs?
First mover effects, network effects, brand recognition, corruption, rent-seeking over the assets created by the aforementioned
Like, yes the main point of leverage obviously isn't software quality, but also, if AI can't cause one person to be able to solve those other problems at some scale that's currently not possible with no other human help, those problems become problems someone else could also throw the AI at. If running a business relies on having some point of market leverage over anyone who pays your business for something, the reduction of the cost of independently gaining that leverage necessarily means the reduction of the value of your business
I guess another possible means to a single-person company with a billion dollar valuation is hyperinflation
With a video game it's not clear that you're purchasing a revocable license. That's why it's called "buying" a video game. If online stores were clear that you were actually leasing a game, I don't think this would be a problem.
Publishers and storefronts need to be clear that what you're doing isn't buying, or start selling tickets (season passes) that have a clear end date, or use some other mechanism that isn't "buying".
The fundamental problem is that it's unclear what you're buying, and the contract can change at any time. Are you buying an item, a ticket, leasing, a subscription like an MMO, etc. These are all different things, and it misleads consumers when they're conflated.
The terms are also very one-sided, and your "purchase" can be ended by one party at will with very limited notice. Even basic consumer protection like requiring six months notice before ending your software lease would help.
If someone goes to the playstation store to buy a playstation game, you don't think it's clear that they can only play the game on playstation? I think consumers understand what they are buying even if they don't think in terms like revocable licenses and drm. They aren't ignorant just because they don't care about these abstract issues that don't affect their lives at all, practically speaking.
Yes, but first sale doctrine on physical goods means that I can take my Playstation game, which only works on Playstations, and sell it to.a friend after I'm done playing it, or pick up a copy 5-10 years after the game was originally published for cheap from a second hand store like Goodwill. People do understand that, and are understandably pissed off that those same rights don't transfer to the digital realm. If I paid $60 for a copy of GTA IV when it was new but now I can buy a copy for $5 from the used bin, why shouldn't that be true for GTA VI in 15 years after it comes out?
I am understandably pissed that Apple no longer makes small 5-inch iphones because I like them. But other consumers like the bigger phones, so Apple discontinued the mini line.
Is Apple doing me an injustice by not selling me the good that I want?
Is this a question of rights or consumer preferences?
Geez man. I've seen your replies in this thread multiple times. At this point, I have to wonder: why are you so against legislating the actual sale of ownership of video games? How could enacting this theoretical legislation, that you believe is a consumer "preference" (hard disagree, but I'll allow for argument) be in any way detrimental to you? What compels you to lick the boot to a polished shine?
Sort of like how Diary Queen aren’t allowed to call their desserts “Ice Cream” because there isn’t enough dairy, we need to force retailers to start calling them game-licenses? This feels like a whole lot of a big fuss over something like that. I really feel like there must be something deeper going on.
Well the goal is to stop killing games, to make it possible for people to run their own games indefinitely. That's a lot harder than baseline consumer protections though.
Agree. The only way "buying" is understood in English to mean a temporary entitlement is when combined with those terms like "ticket" or "pass."
The choice of language is deliberately made to deceive. If an auto manufacturer tried that, offering to "sell you" a car but the 3-page "Sales contract" had a clause buried in there that said "We can come to your house with 30 days' notice and just take the car back and you have no recourse besides stopping your payments" this would be ruled as grand theft auto (no pun intended) not "the terms and conditions allow it"
So the play here is obvious, use the teleoperation as training data for a more general purpose AI controller. You need that data to make a model in the first place.
What doesn't make sense to me is the cost. Yes, $8000 is probably low for this robot but it's a reasonable price range for something like this. The AI credits though? I know vision LLMs are not cheap, they're not going to run something like Llama3.2vision on every frame. Very curious about the embodied AI architecture that this is going to use, and how it can get cheap enough that it's not going to use $500/month in electricity every month.
The world of computer vision is much bigger than multimodal LLMs. You'd run an ensemble of specialized models for 3d mapping, object classification, path validation, and so on. On a raspberry pi 5 8gb you can run what you need to self drive an RC car on an obstacle track at 10 FPS.
Today, a lot of it is just integrated VLAs. End to end backprop covers a multitude of sins - a single integrated model stack in cross-attention beats "an ensemble of specialized models".
The way one of their employees told me it to me as like a dishwasher.
Of course the dishwasher should be more expensive. When you add up the hours in labor saved and multiply by the hourly median wage, you get something in the $50k to $100k range.
But it's essentially just a sprinkler.
Ain't no one going to to pay the cost of a new BMW for a dishwasher.
Same thing here for the laundro-bots. Their competition isn't against the time saved for a person to do it themselves. The competition is a maid that does your whole house for $70.
> Of course the dishwasher should be more expensive. When you add up the hours in labor saved and multiply by the hourly median wage, you get something in the $50k to $100k range.
Except most dishwashers aren't competing against people washing dishes by hand and making the hourly median wage, they're competing against other dishwashers
>> Very curious about the embodied AI architecture that this is going to use, and how it can get cheap enough that it's not going to use $500/month in electricity every month.
In practice most of the costs would be the teleoperation costs rather than AI inference.
8k is cheap if laundry is fully offloaded but will a regular consumer spend 8k on a device that is not proven? I guess there is a subset of consumers that this automatically targets/caters to.
I'd pay $8k tomorrow for a bot that would 100% do my laundry. That means collecting it from the various dirty clothes hampers throughout the house, bringing it to the washer and dryer, operating the washer/dryer, folding and putting clothes on hangers, and putting them back into the dresser and hung up in closets.
For a bot that just automates an in-house laundry service that washes and folds? Not very interesting since it might save maybe 60% of the time, but practically zero percent of the mental overhead.
This seems like a step towards that I suppose. My house isn't configured to make it an option even if it was a fully-baked product, but if these ever get to the point of actually working without remote teleoperation I'd certainly be in the market.
Unless I was physically disabled, elderly, or otherwise unable to do my own laundry, I couldn't even fathom paying a robot (or a maid) to do it. I can maybe understand it if you don't have a clothes washer, and had to wash your clothes manually in the sink or tub or something, but with a washing machine, the machine is already doing 95% of the work! The rest is not difficult or time-consuming. Laundry isn't heavy, and it doesn't take specialized skill or concentration to put them in the machine, start it, or remove them. Not saying your wrong for wanting something like this, but just observing how different people can be with their priorities.
> but with a washing machine, the machine is already doing 95% of the work!
Not sure how literal you were here or if it's more a feeling thing to you: My washing machine takes 1-2 hours per run. I don't believe you (or anyone else) can do all other attached work in 5 Minutes, or anywhere close to it.
It's way less about the time of doing the thing, and much more about the mental overhead of having to remember to do the thing. And needing to do the thing at very inconvenient times because I forgot to do the thing or lacked motivation when I should have done the thing.
This goes for any recurring chore in my life. I love to garden, but watering plants is hell to me after the novelty wears off. I fixed this by installing an automated irrigation system. I get to do the fun bits mostly on my schedule to wind down when I feel like it (pruning, harvesting, staring at plants) and didn't sign myself up for yet another daily chore to do.
My wife is the opposite. She thrives on "chores" or routine simple items like this. She absolutely loves doing laundry to an absurd degree - kind of a zen moment in the middle of her day she can quickly spend 10 minutes here and there to get done. Same goes for cooking. I enjoy planning and creating elaborate meals I've dialed into "perfection" but take me an entire Saturday to accomplish a few times a year. She loves spending 30 minutes in the kitchen most nights to wind down after work - but really hates "big" projects of any type.
I imagine it has a lot to do with executive function. I enjoy large one-off projects (e.g. designing and installing an over the top totally overkill irrigation system) that are eventually "done" but fall apart on repetitive simple things that never end and just reset to be done X hours all over again. I like to have my "mental slate" clean when I wake up for the day, and I find I accomplish far more when I can configure my life in such a way.
As such, this robot as-is would be somewhat useless to me as I'd have to remember to hang up the clothes or walk them up the stairs to put away or whatever even with it. I'd get very little advantage for the spend.
We can't even get to a washing machine that's also the dryer though. If it seems hard to get people to adopt even that new technology, then clothes folding is never gonna happen.
really? you cannot _fathom_ the idea of paying for a robot to do something you, yourself are already capable of doing? someone should tell all these car manufacturers and the like that their cost-benefit analyses of using robots for work humans can do are completely off!
You have a certain number of hours remaining on earth, and that number goes down forever until it reaches zero.
Get on the cross about... doing laundry, I guess? All you want but it's not crazy to want to maximize the amount of time you get to spend with novel, meaningful experience and minimize the amount of time you spend shuffling piles of clothing from one place to another over and over among the dozens of other mundane chores.
yeah, cool: "sometimes when i feel hungry it makes just makes me more grateful for food"... like, seriously? come off it... what a fucking stunting-qua-realitas stance
They're talking about a household chore and your post compares to an industrial production line. It's not like the Ford executives can walk into the laundry and assemble F-150s.
You’re not replacing the outsourcing it component though, you’re replacing a maid at home doing it for you. In home laundry services are a very different experience since you don’t have to also go pick up and drop off the laundry.
A service like that can be hundreds a month, so pay off period is on the order of years… which could be worth it.
I doubt it will ever be cheaper than doing it yourself, like most things in life. The market is for people unable or unwilling to do it themselves.
Outsourcing can be difficult and expensive in many regions. The lack of an actual human might even be considered a benefit in some cases, such as nursing homes (although you have to weigh the benefits of human contact with the benefits of fewer humans spreading plagues).
Set price too low to be truly rare luxury show off item, but high enough that expendable income is necessary for first movers. Trade in kind by "gifting" to influencer types: the pop science tech nerd ones to legitimize it by scrutinizing current downsides, the effortlessly luxurious ones to establish it as a brand, and a few mom-core ones to seed the aspiration). Develop better versions from the initial data, drop prices a few times a year via holiday sale or via model deprecation, keep current model pricing high. Develop 3rd Gen and introduce "pro" tier. Very tried and true strategy (many step omissions of course) and imo they nailed the price point for initial show off. It's not really affordable for its market but it's also not unaffordable if you consider the costs of what it would replace if it turns out to work!
> So the play here is obvious, use the teleoperation as training data for a more general purpose AI controller.
Strong disagree: the play here is to use teleoperation, claim it's AI, make a shit-ton of money and cash out before the house of cards come tumbling down.
The ultrasound technique there is more like MRI, static imaging, not measuring dynamic electrical signals. Also, regardless of static vs moving, all of this hinges on massive relatively expensive devices (ultrasound is never going to be dirt cheap or miniaturized compared to a smartwatch or even a VR headset) and for the subject to remain perfectly still and probably go through regular calibration sequences. All for <75% accuracy on simple classification tasks. The information mixes in the propagation out of the skull, erasing information. It's analogous to trying to do a row hammer attack on a CPU from outside the computer case.
Also, in the Meta result here, "while actively typing" is actually quite different than passive mind reading because the motor cortex sits nicely near the top surface of the skull, and the muscle memory from past typing makes for a nice well formed signal to measure and classify. It's the same trick over and over since the Brown BrainGate days where you can have people perform or imagine movements and get a decent but not good classification result, and it never gets much better after that trick is exploited. Project dies, VCs and grant writers forget or never appreciated the effect, time goes by, a grad student or corporate research lackey rediscovers it, media puts out an article claiming mind reading is here, and the cycle repeats...
If memory serves the cachyos-nix flakes sets a cache for pre-built kernels, but that cache isn't available until you rebuild. So if you want to use the cache you need to do two steps, add cachyos to your inputs, rebuild, switch the kernel, rebuild.
reply