Eigenλime

@elonmusk This is actually dog shit elon. Stop fucking around, we have real work to do.

This slaps 😂

@yishan Most of these replies are completely wrong - it was literally rl + small networks + isaac lab

What’s the core technology that wasn’t available before that’s now enabling bipeds to make progress in ~2 years that took Boston Dynamics a ~decade to do originally? (BD did not open source its control algorithms so they aren’t copied)

@adcock_brett Bro… this is like a week dicking around in isaac lab tops lmao

Uneven terrain

@jloganolson They were too busy wondering if they “could”, to ever ask if they “should”

Unitree G1 crawl policy deployed to hardware! Plenty of room for improvement, but it's a start.

Some time in the future, I'll boot up my ol' x220, pop open a shell, and with a face full of grays ask codex 25-high, "what was I like...back then..." and take a trip down memory lane.

@Stone_Tao Do you believe in classical sims, or do you think world model is the only way to get sufficient diversity in objects/texture/vision

tasks like this are why I’m still extremely simulation pilled demonstrations can be a great way to bootstrap a model. To get the necessary cycle time and success rates you need very optimal data which currently can only come from sim (and maybe world models in the future although it would likely be OOD and difficult to generalize accurately) I think the only time you could get optimal demo data here is if you literally record the angles between bones in the humans hand and the electrical signals signifying touch to the brain (good startup idea??)

@KyleMorgenstein @chris_j_paxton Do you find the isaac lab physics (ex physx) to be better for sim to real transfer than mujoco? I feel like I here stories for both

@eigentime5454 @chris_j_paxton in addition, there’s a huge amount of behind the scenes work being done to make Isaac lighter weight and faster (for example my efforts with uv) so the Isaac team isn’t standing still on that front either

Isaac lab is powering basically everyone right now

@AnimaAnandkumar Lol…jesus…“godmother of ai + science”? How insecure does one have to be to have that in their self bio on x 💀

Our paper from many years ago arxiv.org/abs/1912.02279

Had a conversation a while back about how you’re only a resident in an area when you’ve bought a house in said area. When you think about it long term, we’re all just visitors.

@vinayxmahadik @piotrstomasik @deedydas Does this apply to iso’s? I thought even if you excercise iso’s, as long as you dont liquidate for a year, they get taxed as long term cap gains

@piotrstomasik @deedydas You are right - I haven't done this yet, but it seems you can structure the loan as a non-recourse loan where you just surrender the stocks at the end to avoid taxable income. However, in most cases, the company gets acquired, and you may not run into this?

PSA: If you join an overvalued startup, you can actually LOSE money instead of getting paid unless you avoid this mistake. Often, at startups, you are granted options that you "exercise" at the strike price to own them. This 409a price or fair market value (FMV), the price at the annual appraisal or after the last round, is what you pay to buy your options. The FMV is usually far less than the "preferred price" which is what investors paid at the last round. Great founders know how to keep the FMV low for long enough to make life easier for employees. For later stage, say $1B+ startups, you might have to pay $250-500k in some cases just to own your equity. Startups will tell you "don't worry, it's already worth so much more!" To make it easier if you're short of cash, some offer low interest rate (4%) loans to buy your options. Be very cautious of this. At the peak of 2021, there were certain companies that bust in valuation- their preferred price was worth next to nothing -which left the employees working jobs where they owed more money to pay their loans than their salary. They were losing money working a job. Side note: After ~$3B+ valuations, typically startups transition to RSUs to prevent this.

@shpetim @EricVallieres84 Do you make a lot of money @shpetim ? You seem to be really smart across a broad range of domains (so not just experinced, actually smart). I hope you make a lot of money

How high volume we talking? Lights out high volume would be Enshu & Toyoda. If youre not operating lights out or in a tier 1 shop then the other machines are great choices. The higher end machines will have the best seals on the ball nuts. The waylube/grease will be hard lines not polytube. The solenoids will be from Parker or Bosch not some mid tier brand. The doors will have igus guides and rollers not just generic sliding parts. The cabinets will be extra spacious with proper thermal control.

Which machine tool manufacturer is the go to for high volume manufacturing. Horizontals with pallet pools most likely. Local options are: HASS OKUMA MAKINO DN SOLUTIONS MAZAK Re-shoring manufacturing is happening

@signulll It was lillian. Tbh lillian is worth a B

i work at meta. hr systems. mostly comp processing. quiet job. stable. sometimes i daydream about retiring in portugal. today a package hit my queue. base + bonus + equity. looks normal at first glance. then i open the details. $1,000,000,000 over four years. plus signing. plus year 1 minimum: $100m. i stare at it like it’s a typo. check the name. triple check the level. researcher. coolcoolcool. so now i have to enter this into workday. i paste the first number, field throws an error.“value must be below $99,999,999.” lol. okay. i try splitting it. base in one bucket, equity in another. nope. i try scientific notation. it rounds it down. the system can’t HANDLE a billion dollars. i call someone on payroll infra. tell them i’ve got a 10-digit comp packet. they think i’m joking. i forward the req. they go quiet. “we’re gonna need to escalate this,” someone mutters. “to who?” i ask. “god, maybe.” next thing i know zuck’s chief of staff is in the thread. now i’m in a thread with zuck. because of a number. then i find out the guy declined the offer. just said no. no negotiation. no counter. just… no. i’ve been maxing out my 401k for 11 years & this man said no to a billion dollars like he was skipping dessert. i close the ticket. delete the draft. go for a walk. & then i rethink everything.

@yacineMTB Those are correct numbers - either him or a friend

just missed out a job offer from meta for 8m TC 🤦 it was a 2 hour exploding offer and my phone was on silent so i didnt see it in time

@EndWokeness This is coming from a proud ass American btw.

@EndWokeness Also don't come at me with americans can do the work -> they can't. Why do you think Nvidia is worth 3 trillion dollars? Because of Chinese students who graduate and work fukin 20hr days to make the GPUs work. Most (not all) American's can't or won't do that.

469 American students in China 280,000 Chinese students in the US We educate our enemies while they spy and rob us of our intellectual property. Now it's finally coming to an end.

@AlternateJones This was a good one, was on the fence, but bought 650k at 250 overnight

Called it again 🎯

@DanielleFong Lightcell merch wen?

got inspired to fire my alumina 3d printed vortex holder :)

@kevin_zakka @ChongZitaZhang Have you tried multi-node with torchrun? Any obvious limitations?

@ChongZitaZhang FYI, brax PPO uses pjit to automatically scale to all available GPUs. You should give it a try, a 2x4090 setup made me so productive 😃

This is cool - PPO also differs