BenoitP

Hah! I knew Sudgy would be your first link. There's relatively little content about GA, but once you're hooked you begin to consume it all.

It's like you never got to study Relativity, conformal geometry, electron spins, quaternions, etc then someone comes with a simple cheat code which introduces you to these topics gently. It's like what category theory wants to do for mathematics, but simple.

Here are additional resources:

GA playground: https://enkimute.github.io/ganja.js/examples/coffeeshop.html...

A physics engine in 100 lines (Gravity, Hook, and damping laws are just one line each!), and you can go from 2D to 3D to 4D by changing a single parameter: https://enki.ws/ganja.js/examples/pga_dyn.html

Other resources: https://bivector.net/

And anything by David Hestenes: https://worrydream.com/refs/Hestenes_2002_-_Reforming_the_Ma...

On child #2, can confirm. We actually planned for it. By planning for no activities. Just bunkering and resting for 2 months straight.

We began partial daycare on Thursdays + Fridays in winter. Daycare is like Cancun for bacteria. Dry cold can weaken the skin, and cold is good for bacterial protein stability.

The cycle begins: kid starts incubating on Thursday. On Saturday he's sick, and you hope for a good lottery. Diarrhea? Puking? Maybe not. But a runny nose most probably. These may ruin your nights, still.

Sunday they are sick, and you begin to feel unwell. Monday the kids is recovering, but you're still in it. By Wednesday you should be good though. And then it's Thursday again.

On another topic air filtering is great for the kids I'm sure; but I can't help wonder if it's just kicking the can down the road.

This. On top of the MCAS modification being necessary in the first place.

Higher torque generated because of more off-center thrusters surely raised eyebrows in the stall risk mitigation team.

In the end, that's 3 teams where management made the topic being fixed by another team. That's systemic. And with a single confounder: cost measures being prioritized over safety.

With the current QA problems, it cannot be made more clear that that single culture element is the root cause at Boeing. And ethics dictate that no more experiments be made, and the top 200 of management be forcefully removed.

I'd argue you both could be right. Your comment could lead to a definition of intelligence. Organisms capable of causally influencing deterministic systems to their advantage can be marked as intelligent. The complexity of which would determine the degree of intelligence.

Your point is great in that it pinpoints also the notions of agency scopes. In all the causal DAGs it feels like there are implicit regions: ones where we can influence or not, intervene or not, observe or not, where one is responsible for or not.

An intelligent agent is one capable of modelling a system, influence it, and bias it. Such that it can reach and exploit an existing corner case of it. I talk about a corner case because of entropy and murphy's law. For a given energy, there are way many more unadvantageous states than advantageous one. And the intelligence of a system is the complexity required to wield the entropy reduction of an energy source.

If you want to build yourself an order of magnitude of how much performance is left on the table in CPUs, consider the following:

* A single bit Adder with Carry can be made with 9 transistors, and has to wait for 4 stages for the signal to propagate into. Then you have to stack 32 of them in series for a 32 bit carry. And wait for the signal to propagate.

* Then you first have to route the adding instruction, which is code treated as data. To do a single addition, you have a whole meta pipeline to run. With a lot of side-support: prefetching, cache coherency, loading, writing, instruction pointer, stack pointer, etc.

* Then you have to have the data available, or else you'll be waiting for RAM. That can be as long as 200 cycles, all the while you could be doing several additions in parallel; wasting 650 picoJoules, when the addition in itself is just a few picoJoules.

As silicon scaling laws hit their respective walls (latency, clock limits have been hit; bandwidth, memory sizes remain growing) the only way we'll be able to eek out more performance is to go ASIC on a lot of stable code.

What about governance in The Hive? For Cathedral and Bazaar, it's simply the two opposite corners: control everything, or control nothing. Just like we use binary because it's the min and max of voltage: it's simple. We don't use intermediary voltages. It costs energy to maintain that.

To align with the colorful analogies in the article, governance in The Hive is 'draw the rest of the fucking owl' here. Ahem, what is distributed consensus?

I have quite a few ideas about that, but they remain to be tested (Define a voting pool, let everyone vote for everyone, run a PageRank, publish the top 20 Queen Bees, let singleton decisions be taken in Queen Bee committees, decay votes for renewal). Let's call it "Human Staking"? Surely some research has already been done about that.

Can someone knowledgeable about this chime in:

> The invention of group theory. In proving that there are no general algebraic solutions for the roots of quintic equations, Abel invented (independently of Galois) what later became known as group theory. In addition to Galois, the topic was also studied in the same period by Joseph-Louis Lagrange (1736–1813).

How are quintic equations related to group theory?

> It is too complicated.

I wouldn't say this. It can bring complications, yes: but only because like a lot of expressive tools people will over-engineer things with them (probably like this sentence)

To me, Typescript saved JavaScript from under-constraining. It's bringing me some peace of mind.

If you like playing with First Order Logic, I cannot recommend enough this playground site:

https://incredible.pm/

You can even introduce your 10 year old to FOL with it

The field is evolving quickly and it's pretty to see.

Another deeper reason to me is that it employs some neural network techniques -which have differentiability-. This means that you can backpropagate errors from a domain to another. In the future, we'll have videos as embeddings that can understand written signs, and interpret the voices and what they refer to in the scene.

This is not AGI, but will be much closer to the sensor fusion us humans do.

Seems like it's quite doable even with standard equipment. I follow an FPV (first person viewer) youtuber and he did quite a good series on his Everest trek. Did not go higher than base camp, though. Here it is:

https://m.youtube.com/playlist?list=PL5Y2yyiC4P5QZLGC7SJcKwC...

----

On another note, the Mars Helicopter drone worked just as well. It just had to ramp up to crazy rpm. I guess it's all about energy. 1m at sea level is about the same energy as 1m at Everest height. Drones can accelerate a lot so I guess 3x higher rpm (atm density is a third of sea level) is no problem.

I wonder if a set of learned function (can|does) reproduce the truth tables from First Order Logic.

I think it'd be easy to check.

----

Anyways that's great news for differentiability. For now 'if' conditions expressed in JAX are tricky (at least for me), and are de facto an optimization barrier. If they're learnable and already into the network, I'd say that's a great thing.

They do have a saying about catching a falling knife at Wall Street. This is definitely not my expertise though, I'll let others comment.

This is the future. As we try to go 3D and start stacking dies, the limiting factor of heat is getting even more blocking.

It's not about consuming less electricity. It's about dissipating less of it as heat inside the microprocessor. The future will made of tiny porous cubes that are tall sandwiches of RAM and CPU/GPU/etc

Seems quite close to the 'active learning' ML topic: let the model choose the question that will bring the best information return.

And this kind of decomposition using simple knowledge pieces (logical axioms?) is IMHO what we have to do to bring LLMs to senses. These pieces should light up in the intermediate embeddings inside the LLM. It won't really be more intelligent, but it'll model reality better.

Well, I consider Lorentz' work to be old hat. I can't find an older example though.

https://en.m.wikipedia.org/wiki/Lorentz_group

Ah the good old days of Michael Arrington. TC was my gateway drug to silicon valley produced news, including HN.

He had the habits of pulling stunts like showing up at meetings he wasn't invited to. And the ways of a savant lawyer at extracting information from people. I've always suspected he knew well in advance about the iPad; and he convinced investors to back him up to produce an equivalent product, partnering with a laptop company. The product got shipped, only just at 8-10 times the weight of the iPad though.

He'd have made a fantastic COO, or investor relationship manager IMHO if he had not chosen to do TechCrunch. I hope he's doing well at his cryptofund.

You should jump straight to geometric algebra. It's actually simple there. Quaternions and Complex Numbers are about just some not so special multivectors.

Here are 2 45mn videos you'll not regret:

https://youtu.be/htYh-Tq7ZBI?si=qi9HfU40qKcc_HNc

https://youtu.be/60z_hpEAtD8?si=1M3dDAh7vPjx2Q8Y

In the "Multivector Derivative" section you may plug in an arbitrary multivector function of your liking:

https://en.wikipedia.org/wiki/Geometric_calculus

At some point there has to be acknowledgement that the design envelope is unsound.

Newly fitted engines have moved the center of thrust far away off the center of mass. This creates huge torque; especially as is it often done in emergency situations: stalling avoidance advises to push thrust to the maximum. If you've practiced a bit with plane simulators (the ones where you design your planes), you know these things are uncontrollable.

And us software engineers should know better: software-patching design errors sometimes just doesn't work. You can't just "polish it". Without active control you can't do it (which by law is reserved for the military casualty risk tradeoff)

MCAS-patching it yet another time isn't the answer.

React! This has been on my Christmas wish list since Observable was first published.

Thanks mbostock!

This should be seen in light of the Great Differentiable Convergence™:

NERFs backpropagating pixels colors into the volume, but also semantic information from the image label, embedded from an LLM reading a multimedia document.

Or something like this. Anyway, wanna buy an NVIDIA GPU ;)?

For context (because he's too good to brag) OP is among the original creators of Apache Flink.

Question for OP: I'd bet Flink's Statefuns comes in Restate's story. Could you please comment on this? Maybe Statefuns we're sort of a plugin, and you guys wanted to rebase to the core of a distributed function?

---- Updates:

(SpaceX) 6:45 AM CT The Starship team is go for prop load

(SpaceX) 6:04 AM CT Now targeting 7:50 a.m. CT, official stream says broadcast starting at 7:10 AM CT

---- Streams:

High Quality VLC: Open VLC, Media, Open Network Stream, paste following, Play:

(higher quality) https://prod-ec-us-west-2.video.pscp.tv/Transcoding/v1/hls/G...

(lower latency) https://prod-ec-us-west-2.video.pscp.tv/Transcoding/v1/hls/G...

--- When

Broadcast usually starts 30mn beforehand (6:30 AM CT)

Launch window: 7:00 AM CT - 9:00 AM CT

---- Streams:

High Quality VLC: Open VLC, Media, Open Network Stream, paste following, Play:

(higher quality) (will post as comment when the broadcast starts)

(lower latency) (will post as comment when the broadcast starts)

NASASpaceflight: https://www.youtube.com/watch?v=mTkhv4fvOgA

Spaceflight Now: https://www.youtube.com/watch?v=lFkqZF-Ss7o

Everyday Astronaut: https://www.youtube.com/watch?v=8VESowgMbjA

(FR) Techniques Spatiales: https://www.youtube.com/watch?v=rwbSOvi34KI

---- Mission profile:

Attempted booster flight back near the shore!

Low wind, sunny weather, we should have good images

https://www.spacex.com/launches/mission/?missionId=starship-...

---- Links:

https://twitter.com/SpaceX

https://twitter.com/elonmusk

https://old.reddit.com/r/spacex/comments/1d6wx6w/rspacex_int...

There's the Stowe Janey Scerri Table.

https://en.m.wikipedia.org/wiki/File:Stowe-Janet-Scerri_Peri...

OP is also the author of the popular dimensionality reduction algorithm UMAP.

I guess the pipeline was embedding documents with an LLM (or even plain old word2vec average over the abstract might do it), and then reducing that to 2 dimensions with a cosine similarity metric with the help of UMAP.

I have no idea about colors and local cluster naming though. Maybe that's handcrafted.

Pretty interesting!

A big surprise for me was to find the explainability cluster quite far from the causality one. But I guess it stems from a cultural facet: causality is mainly the purview of statistics (with Pearl at the helm) with a strong medical sciences focus; when explainability is more of a reaction to algorithm used in the industry (trees, GLMs, neural networks, etc.); which you deploy for performance, and only then you care about knowing the why.

What a wonderfully crafted piece. But a part of me can't refrain from saying it would have looked way more elegant and succinct in 3D Projective Geometric Algebra.

Most of the last sections (all intersections) feel like corner cases, when in PGA they are one and the same.

That'd be great if it could absorb the surplus, but right now at 12GW production it is curtailed at 0.29 GW. This means neither pumped storage, nor batteries, nor exports can deal with it.

On one hand it might create a new market, hopefully H2 electrolysis that replaces coal or gas in industrial processes rather than Bitcoin farms. On the other it also signals that any additional Solar is going to evict about as much capacity into such curtailments.

Anyway, California is the record-breaking market and all eyes are on it to helps us guide the global energy future.