
continually learning in a state of delight | ex sr MTS | interested in ai epistemology | robotics sys, infra, ml @AMRWelding | father of 2
synthesis
Joined May 2022
- Tweets27.2K
- Following5K
- Followers1.1K
- Likes336K
malachi retweeted
I might be paranoid, but I think we all need to prepare for a large-scale temporary breakdown of the internet
Something like a network partition caused by a major BGP-like incident, or so much infrastructure being down / compromised that large parts of the network stop communicating with each other for a few hours / a day. We have been operating under the assumption that the internet is a reliable layer, but I think that assumption might break in the next 12/24 months
For defi protocols, we need to prepare for a scenario where Ethereum mainnet does not finalize for a day, with 2-3 competing forks being built in parallel
You can think about:
- Containing exposure. How do you trigger your circuit breakers across different networks? Can you have geographic redundancy?
- Maintaining critical operations. If you rely on bots, what happens if they are all running in a network that becomes unreachable?
- Avoiding unsafe transactions. For example, if two forks A and B are being built, and you send a swap transaction on A, but B becomes canonical, your transaction on A could be replayed on B. So you need to be mindful of replay risks, inconsistent chain states, etc.
- Off-chain code. How robust is your offchain code to reorgs and temporary loss of finality?
- Chain finality / L2s. What happens if the L2 sequencer is unreachable? How does finality work on non-EVM chains you use?
This might sound like sci-fi and far-fetched. But given everything we have been seeing over the past days/months, I think such an incident, while still unlikely, is now plausible
You don't need to go full doomsday, but I would recommend including this as a worst-case scenario in your incident response plan
malachi retweeted
Still a lot of work to do here but achieved SOTA on temporal coherence today!
2.5x improvement over SOTA in both depth error and jitter!
malachi retweeted
I would expect B300 pricing to continue declining.
A staggering number of neoclouds have approached us @liquidcompute looking to remarket long-term B300 capacity originally committed to Chinese-affiliated offtakers into the US and Western Europe due to concerns of stricter legislation regarding compute servicing to China.
liquidcompute.com/share/b300…
malachi retweeted
with model factory, one can manage a portfolio of Prime Agents, and they manage the training runs. AI researchers are now quants
malachi retweeted
Elon Musk reveals some of his best data center builders came from In-N-Out, because aces "making epic burgers all day long… are gonna do a damn good job of network cables"
"And then we have an in-house team, not just contractors. We have contractors too, but we have an in-house team, which we call sort of ArmyX, that is productionizing the building of the data centers."
"And doing the basic principles of productionizing things, which is where you break the tasks down into things that are easily trainable within a few days. So you don't need people that have, say, been doing networking cables for their whole life. You divide the tasks into things where you can train someone new who may have been, like, a Starbucks barista, or stocking shelves at Target. But by the way, people who are good baristas or good at stocking shelves are great workers. And also, hire someone from In-N-Out."
"Those are some of our best people. They've never done data centers before. But let me tell you, if you get some aces out of In-N-Out who have been making epic burgers all day long, perfectly, they're gonna do a damn good job of network cables."
Elon Musk says TeraFab should make its first test chip around June next year and reach volume production toward the end of 2028
"So, yes, I think it's going quite well. We expect to have the sort of research fab in operation, and hopefully make our first test chip around June of next year. So I guess we're about, like, eight or nine months away from that."
"And if all goes well, we will achieve volume production towards the end of 2028, at least for phase one of TeraFab. And then we'll go to phase two and onwards. But I think you'll see some very rapid evolution in the way that chips are made."
malachi retweeted
Reality is,
If you're serious about carving your name in history, you can either write a timeless piece worth reading, or live a life worth writing about.
Outside of that, your life oscillates from one meaningless distraction to the next until you just disappear.
As if you never existed.
malachi retweeted
Introducing Frostbite: a new class of CPU built around interaction nets.
Given only interaction net theory, our agents co-designed the ISA and chip architecture from scratch, running large swarms to design, verify, and optimize the CPU down to the final GDS.
On our first set of target workloads, Frostbite is 4.6x more performant per watt than modern traditional CPUs at similar area.
malachi retweeted
In Space-Time, sometimes you really need to see things in multi-dimensional movement to determine what they are.
malachi retweeted
I love this so much because I'm a sensitive young man attuned to the higher concepts like color, resonance, ensembles
malachi retweeted
sharing this tweet for absolutely no reason why do you ask 🌞
Did this offensive lineman’s stance tell help decide the Super Bowl?
I classified every Patriots early down snap for the whole season entering the Super Bowl (situations where they could plausibly run or pass) and found a glaring tendency from right tackle Morgan Moses.
Entering the big game, with Moses in a two point stance:
• Hands together in shotgun: 90% pass on 202 plays
• Hands apart: 78% run on 238 plays
The photos below show plenty of examples of the contrasting looks. In the video, watch Seattle’s pass rush tee off on three critical early down sacks, seemingly unconcerned about the run. Moses had his hands together on all three plays, and each possession ended in a punt.
There also appears to be communication among Seattle’s defenders before the snap. I can’t confirm they were communicating about this particular tell, but I strongly suspect they at least knew about it entering the game.
A tendency this strong can let defensive linemen pass rush immediately and help coverage defenders anticipate pass rather than hesitate over their run responsibilities.
I honestly can’t believe this wasn’t corrected or adjusted over the course of the season.
I had planned to share this before the Super Bowl rematch, but a player on another team encouraged me to hold it because they wanted to use it themselves against New England.
While working with the Raiders ahead of the 2025 opener, I had come across this tendency from Moses’s previous season with the Jets. We used that information to contribute to a strong defensive performance in a major upset win in Foxborough in Week 1.
These details may seem small, but the edge they give a defense can make a huge difference in who ends up holding the trophy.
malachi retweeted
"Wouldn't it be funny if we put women in government and they started doing insane things like banning videos of women saying they're happy being wives?"
"That's hilarious! Could you imagine? What kind of a society would be crazy enough to do that?!?"
malachi retweeted
Replying to @reneromero08 @elonmusk
Our leaders should read Nietzsche and Elon has read Nietzsche
It is not causal. He’s an unelected king
He could and should step into the role and confront his responsibility to society, if he wanted to.
Idk why he elects not to @elonmusk
The actual message of plato’s cave, if you can escape the wall-shadow versions, is that the world of discourse and ideas and custom is false, and the hard material world of uninterpreted perception and natural consequence is the real, true, and good one. You can escape the matrix
malachi retweeted
the problem with the iranians is that they're much smarter than the arabs
malachi retweeted
(4/n) The first takeaway is that visual representations emerge from this on their own!!
The model learns to represent large background regions with just a few tokens, while retaining finer structure in semantically rich foreground regions.
What’s more, if we put text and images together, we even see a dense, segmentation-like effect! The model automatically learns to connect words with fine-grained image regions—without word–region annotations.
Same noisy image, different words: “bed” lights up the bed, while “oven” lights up the oven! HiT learns these word–region associations without segmentation masks or region annotations.
malachi retweeted
Thoughts after talking to lots of engineers in London in last couple of days:
• Everyone's moving into the same direction and getting more and more agent-pilled, but all on individual speed. No group of engineers right now is uniformly agent-pilled.
• Becoming agent-pilled is a one-way door. You only realize this once you've gone through. Many think they "get it" but haven't made the switch yet.
• Wild how many orgs are still treating agents as "thing that writes code on my computer" instead of robot-engineer that can help with whole SLDC.
• Teams are drowning in PRs and code review. No process has been changed yet, even though this is the biggest dev productivity lever of last 30 years, engs still work off tickets like nothing happened. Wild.
• Barely anyone realized yet that, hey, even if the model isn't good enough to write the code the way you'd write it [doubt it], they can surely write all of the peripheral tooling to speed you up.
• Very few engineers have imagination to come up with better tool. Instead of reviewing code in diffs, line by line, think: if I could wave a magic wand and make tooling disappear, what tooling would I want to gain confidence in this change being correct? Full instrumented/traced end-to-end runthrough? Demo videos? In-production probes? High-fidelity and anonymized production data fed through test runs? New testing environments with the greatest debugging MCP support one can imagine so agent can inspect everything?
It's time to dream big.
But seems like most people's dreams don't reach further than "this agent works off the tickets assigned to me 10x faster than I can."
malachi retweeted
How does this enable cross-embodiment learning? A human hand and a robot gripper look nothing alike, but they can move objects in the same manner.
The scene flow is similar between embodiments, so this allows FloMo to learn from action-free human videos.
malachi retweeted
Separation logic and IRIS should be part of the standard computer science curriculum.















































