Ata (@ataghof) header
Ata (@ataghof) profile picture

previously, a few firsts in hardware, then onchain. now, building out human data collection in SEA.

SF / SG
Joined March 2010
Photo by Ata (@ataghof)Photo by Ata (@ataghof)Photo by Ata (@ataghof)Photo by Ata (@ataghof)Photo by Ata (@ataghof)Photo by Ata (@ataghof)Photo by Ata (@ataghof)Photo by Ata (@ataghof)Photo by Ata (@ataghof)Photo by Ata (@ataghof)
Today we launch #MemoriesUpload. Robots need internet-scale first-person video of how people get things done. Collecting this at high quality takes hardware, model and infra working together. That's what we've spent over a year building. And already, 3 of the top 5 frontier robotics labs are training models with the data.
79
135
108
1,178
1,025,839
We scaled a robot model natively to 8,000 timesteps of context, 5 minutes worth of muscle memory, with constant inference cost. Robot policies used to live their lives a few frames at a time (< 0.1 sec), instantly forgetting what just happened. We pushed to 3 orders of magnitude beyond SOTA. Introducing RoboTTT. Test-Time Training (“TTT”) carries a tiny model *inside* the model. Every incoming sensor reading triggers one gradient step on that tiny core, so the history keeps getting compressed into its weights. The hidden state has a fixed size (literally a small neural net), so the robot can “grok” arbitrarily long experience with little overhead. Learning continues indefinitely after deployment. We can then put an entire video in context as prompt! RoboTTT enables one-shot in-context learning from human video: in circuit board assembly, a human demonstrates a never-seen configuration once, and the robot imitates it faithfully. Humans drop things all the time, but we pick them up so fast that we don’t even notice. That reflex to fix is half of our physical competence. RoboTTT shows self-improvement on the fly: the robot is skilled at recovering from its own errors mid-episode, and each fix enters its context to inform the next move. The TTT core distills a general-purpose, failure-to-correction mapping from the training data. One more thing. What excites me the most is a new Context Scaling Curve: from 128 to 8K timesteps, closed-loop performance hill-climbs steadily with no sign of saturation. 8K-context pretraining beats 1K by 62%. What LLM enjoys, robotics should too. Soon, even 1M context is not a fantasy. Deep dive in thread:
72
189
53
1,298
340,689
He worked on Roblox’s world model. I don’t usually share jobs moves, but this one caught my eye. Amongst the people collecting the data to make world models they think @1x_tech has the best world model outside of Tesla. And the cuddliest robot. I watched @karpathy hug one.
I’m excited to announce that I’ll be building robots as VP Engineering at @1x_tech. Rapid progress in world models will create a robotics inflection point. Very excited to work with @BerntBornich and the stellar 1X team to bring humanoid robots to homes very soon. We’re hiring.
10
12
274
40,975
Ever wondered how traditional finance concepts like options trading can unlock insights in the NFT lending world? Let's dive into how perpetual loans on @banx_gg mirror options trading 🧵
1
2
1
5
304
Any LA based founders/devs want some FREE office space to work out of? Got some underutilized space and I’d love to shoot the shit while grinding. Would prioritize fintech/crypto projects
3
1
6
575
warriors in 6
34
7
92
14,521
AARON GORDON SENDS LANDRY SHAMET TO HELL youtu.be/PXzFS1Wk6JI
342
3,653
887
39,652
4,155,736
Has any DAO put all private round SAFTs on-chain with governance voting power? Or will @PsyOptions be the first DAO in the entire world to accomplish this? app.realms.today/dao/PSY
2
8
Howdy #HackerHouseSingapore @ataghof is hosting a workshop about 'Bootstrapping Networks Using Tokens'. Go check it out! 🚀
2
5
Calling all $DUST and DeGod enjoyers... When options for the DUST Covered Call Vault start getting minted and sold, what time would work best to hold the auction? If not listed, let us know in the comments! @degods @frankdegods
30%Monday at 6pm PST
15%Tuesday at 6pm PST
10%Wednesday at 6pm PST
45%Thursday at 6pm PST
20 votes • Final results
4
5
🚨🚨🚨 @CurveFinance frontend is compromised, do not use it until further notice!
222
1,703
375
3,898
In light of recent events, we'll be giving away 10 Ledger Nano S's Retweet, Like, and reply with two friends who are also in need of a Ledger. Let's get those funds secured!
959
907
28
1,238
We can push 3.33% ➡️ 33.3% for $DUST realms.today/dao/PSY/proposa…
2
3
14
.@solana ecosystem's protocols are #PluggedinbySwitchboard! Introducing our brand new Switchboard Ecosystem page at switchboard.xyz/ecosystem Find out in this mini thread: 🤝How Switchboard operates with these protocols & 🔌How you can interact on our page!
9
14
2
39
Psydo = 3 products + 21 contributors + 1 token and a new website to show all that we've built ⬇️ psyoptions.io
35
22
3
62
Earn 24.14% on $BTC, or up to 43.38% with a 1 year lockup with $SRM boosted rewards! Deposit today on psyfi.io
2
4
Interesting look at how much money you can actually make (or lose) staking in the @PsyOptions vaults rn with the boosted @ProjectSerum rewards. read here 👇 @FullStack_tm readfullstack.substack.com/p…
2
1
6
gm is now "grand migration" until further notice Get out of bed, head over to app.psyfi.io to transfer your v1 Vault positions to v2 and have yourself a grand migration! 🛏 💤 ☀️
Out with the old, and in with the new; we’ve got an updated @PsyFinance UI to share with you! Let’s take a look at the core section of the PsyFinance experience: the Vault Information Page. 🧵 🧵 🧵
1
7
3
23
👀
This quoted post is unavailable.
1
3
15