“My prediction is next year we’ll have over a thousand, maybe a few thousand, Optimus robots working at Tesla.”
https://x.com/SmokeAwayyy/status/1801400924096839732
This was a prediction made during the Tesla 2024 Annual Stockholder Meeting.
I take "working" to mean autonomously deployed in factories and performing productive labor.
Update 2025-06-06 (PST) (AI summary of creator comment): The creator has clarified their interpretation of productive labor:
Robots dancing as entertainers for the human factory workforce would be considered as performing productive labor.
1,000
3.00@pietrokc I think I'd have to accept "dancing as entertainers for the human factory workforce" as productive
@DavidFWatson oh wow, ok, I just looked at your all time profit..... Thank you for your contribution to the platform I guess....?
@DavidFWatson and yes hahaha, it was TSLA stock price prediction that I missed by $8-9 dollars I believe
Side commentary; it's weird that Elon seems to want this to happen. The point of Optimus is supposed to be that humanoid robots will be "good enough" at a huge range of tasks, because they can interact physically with the world in ways similar to a human. The point is not supposed to be that humanoid robots will be better at assembly line manufacturing!! That's where specialized robots obviously shine. Am I missing something?
@diracdeltafunk I don't really think you're missing something, though presumably he understands what humans are currently doing at Tesla and whether it would be useful to have a few thousand bots around.
I did notice that he focused mainly on "everyone will have a bot to do chores" and not "everyone will have a bot mining raw materials and assembling shit 24/7" at the last shareholder meeting though
@diracdeltafunk yall are definitely missing something, you probably never walked through a car manufacturing line, even with the specialized robots we have, there millions of humans work work in assembly for specific tasks that a human hand and opposable thumb is better for. Lol
@DavidFWatson what kinda retarded reply is that? I was saying in general, but Tesla has dozens of thousands of people across their factories that work in the assembly line
@DavidFWatson I think they employ less than 100k, so we're talking 10% replacement. And that assumes they don't have to downsize due to falling sales
@DavidFWatson they have over 30-40k that work in assembly lines, I am not sure what ur point here is?
And lol, bros never read their balance sheet
@MolbyDick To be precise, my assembly line employment count was 70k. 10% replacement would be extremely ambitious. Furthermore, I'm aware of their current financial position, my point is that any reasoning about their future actions includes taking into account risk to Tesla's business.
@DavidFWatson Tesla may be struggling with Auto sales, but they have made it clear they made it clear the core auto business is not what they are, Tesla energy alone will overtake auto sales business long term,
I believe cheaper models and Unsurpervised FSD will help with sales, but overtime majority of Tesla car sales will be to fleet operators. Not to mention teslas revenue mix will continue to grow in other areas, like Energy, Software, rideshare & eventually Optimus which will become their bulk of their revenue in the next 5-10 years.
@DavidFWatson I don’t know if I’d be @ 30% by 2027, like I said I believe cheaper models and FSD will drive higher sales, not to mention it will be hard to quantify because regular people will be able to buy cybercabs and other models and put it on the robotaxi network when they feel like it.
@DavidFWatson I said “over time”
No 2027 also do individual buyers who put their cars on the network considered fleet?
@MolbyDick we're talking about this in the context of this Optimus robot event which expires at the beginning of 2027
@DavidFWatson lmao wat? I made a general comment, and bc it’s under a market for a specific take you think that’s what I meant
@MolbyDick I said effectively "one reason they might not have 10k robots before 2027 is that their sales might fall"
you said "ahh, but they'll be selling to fleets, soon they'll sell a majority of cars to fleet operators"
I said "want to bet on that" and created a market expiring at the end of 2027
and you said "wait, no, I didn't mean they'd be selling to fleets so soon"
If you don't think they'll sell tons of cars to fleets by the end of 2027, how could that help them pay for robots by the beginning of 2027
@DavidFWatson try re reading my comment again? That is not what I said, you missed out on the “I believe cheaper models and FSD will drive higher sales, but overtime majority of sales will be to fleet operators”
I’d work on your English comprehension, coming from someone who english isn’t his first language
@MolbyDick Either you were responding to my claim about robots before 2027 or you weren't. Given that you seem to have hit the 'reply' button, I assumed you were.
@MolbyDick I'm glad they make clear that they want to be other things in the future. Does any of that help them invest in robots before 2027?
@MolbyDick I am confused on what ur saying about “help them invest”
A company that has 36B of cash in hand prob has enough to bring the manufacturing line and supply chain online lol
@MolbyDick I was imagining 10B cost to get them built. Everything has opportunity costs, you don't blow that much money on falling production of a product. But it's true, they could probably do it if they wanted to
@DavidFWatson lmao wat, 10B? Are you getting those numbers out of ur arse hahahahahahahahaah go read their latest earnings report bud, look at their capex
@DavidFWatson oh bud, ur not serious are you? The cost of building the Optimus manufacturing line would appear in capex, which they’ve started in q1, their cost to start the manufacturing line and start building the robots this year will prob be under 1B
@MolbyDick yes, and I'm sure they plan to build lots of robots with that, but I'm not convinced that's on a trajectory to deliver the robots in 18 months. I'm saying they'd need to be spending more to deliver them that fast in a usable state
They could probably deliver 10k robots in boxes with this spend. But they're not going to be "working on the factory"
@DavidFWatson what are you talking about? The actual cost of each robot in the beginning will prob be around 50-60k as they ramp they will get down to 25k
They are going to build 5k this year, and then ramp their manufacturing line next year. I am confused, do you understand how manufacturing lines work?
@DavidFWatson they are doing one shot transfers from RL (in sim) to real world execution
This dance video for example https://x.com/tesla_optimus/status/1922456791549427867?s=46
Was one shoted, if your familiar with Reinforcement learning, you’ll quickly understand the getting these bots to useful tasks by EOY extremely doable
@MolbyDick I'm quite familiar, and while the dance isn't hand tuned, and it's great to be doing learning in simulation, dancing demos are much easier to accomplish than anything that involves interacting with objects. That's why Boston dynamics does so many of them. Next step will be getting good results from simulation RL for tasks that involve interaction with objects. But a robot that needs engineers to build a 3d physics simulation for every single task is not cost effective.
But! If they wanted to hire engineers to build 3d physics simulations to do RL training for every single task on a factory floor, that would be the kind of thing I'm talking about: very very expensive, but doable in 18 months with billions of dollars.
@DavidFWatson The real goal is good one-shot performance from a single observation of the task done by a human. But that is not what we're seeing in the twitter post above.
@DavidFWatson I think you lack a basic understanding of RL https://x.com/tesla_optimus/status/1925047336256078302?s=46
Being able to record something doing something then train the bot in RL IS the whole point of this, idk where your getting the billions of dollars to build th RL system in si lmao? What ur saying is literally what they are doing, right now, so instead of teaching the bot via teleportation g, you do it via video…. Do you get it yet? 🤣
@MolbyDick I think you underestimate how many iterations need to be run in simulation to achieve the real world outcome you’re seeing. For that matter, I think you don't realize that the RL simulation environment is set up by hand, it's not somehow built up based on video of the example task.
That's why they do so many dance videos! There's only one element to set up in the simulation: the floor. It's really simple to set that up.
@MolbyDick I read what tesla, it's 'one-shotted' except for the millions of iterations of RL in a realistic physics simulation.
@MolbyDick I really don't think you understand how Reinforcement Learning in Simulation works. And it's probably going to lose you some mana.
Just not nearly as much as your inability to integrate political risk into stock market calculations has 😆
@DavidFWatson you clearly don’t understand what Tesla meant by “one shotted”
https://x.com/ashishkr9311/status/1922483092415033495?s=46
If you can’t extrapolate from here, you’re in for a whirlwind in the next 6 months
@MolbyDick ur whole premise of it costing “10b” tells me everything I need to know about ur comprehension on this matter 🤣🤣
@MolbyDick This sentence:
"All of these use the same pipeline: RL training in sim followed by sim2real (zero-shot)."
it means they did an unlimited amount of reinforcement learning in simulation of the exact dance being demonstrated. And then the real-world robot could perform the dance on the first try "zero-shot".
It's very impressive! But it requires RL training for every task
@DavidFWatson Your mana profit tells me what I need to know about your ability to predict future events
@DavidFWatson it completely dismantles your retarded take that it would take 10B for them to build the bot. If you know what you just quoted to me means, then you realize how wrong you were, are you going to admit u were being retarded or pretend this doesn’t completely negate ur point 😂
@DavidFWatson check back in 6 months, mister “it will take 10B for Tesla to be able to make the bots” 😂
@DavidFWatson I do prove it, to my net worth, and that’s all that matters.
How about we make bet on an actual money market?
@MolbyDick I’m happy here thanks. You're talking to me in the comments of a prediction market that you've already sunk thousands of dollars into. No, I will not be joining you elsewhere.
@MolbyDick I don't understand what's o wild about that? They expect to spend 2b/year, and I think that if they wanted to accelerate timelines enough to do this they'd need multiple years of progress, in 18 months.
@DavidFWatson what’s wild is that you have absolutely 0 fucking clue what ur talking about.
They will make 5k of robots this year and 100k next year.
I don’t think you understand what the limiting factor is or the bottleneck. why would they need to spend 10B? Before you said it was about training the bots, now you’ve realize how retarded that is, so what’s ur new goal post? What do you think Tesla needs to spend 10B on?
@MolbyDick it’s still about training. But like, look at my exposure to this market, to other Tesla ones. I hold all these opinions quite lightly. I’m not certain about anything
@MolbyDick if you wanted to deploy Optimus faster, it would be more expensive to make that happen then to deploy slower.
Most of the cost would be software, since that’s the part that seems furthest behind.
@MolbyDick but it’s all academic. They said wise they plan to spend. We’ll see what they accomplish with it. Maybe they’re actually very close to it being able to do arbitrary tasks. But I don’t think so.
@DavidFWatson ur saying empty words. Why would it be more expensive (10B) more to deploy it faster, and what do u mean faster? Faster then their current go of 5k this year, 100k next week, and 1M by 2030?
@DavidFWatson answer the questionS.
What do you mean by Academic?
What do you mean by they will need to spend 10B to deploy it quicker?
What are you saying? Get specific, or is it that you have 0 clue what ur speaking about?
They’re not going to spend 10B, so it doesn’t matter whether they’d succeed if they did
My claim is that they would be spending more on huge RL in simulation across shitloads of domains to get something useful across their factories from of their current pipeline in 18 months.
but that’s not what they’re doing! they’re trying to make breakthroughs that make such an approach unnecessary.