How are you defining "statistical inferences" as distinct from "logical inferences"? If you're defining fuzzy logic (e.g. not necessarily yes-or-no answers but allowing for ambiguity in conclusions), then we can agree conceptually, but your choice of wording is, I have to say, weird if so.

Blue Origin is far, far from having "caught up". They've had three launches, with a 33% mission failure rate. They are now where SpaceX was fifteen years ago, but with a much worse record (Falcon 9s first mission failure was launch 19, though it did have a partial failure on launch 4 -- primary payload successful, secondary payload failed). New Glenn has done better on booster recovery, but they weren't the ones learning how to do it.

And even if SpaceX never manages to make Starship fully-reusable, they can always punt, build a lighter, fully expendable second stage and have a launch platform that blows every other heavy lift vehicle in the world away.

The Chinese are moving pretty fast but they're also a generation behind.

Maybe, but the problem is that the electronics have to run at those temperatures and not have solder joints start popping, or other fun failures.

As far as I know, it is just real-time. And it didn't even slow down at the flashing red light. So either it recognized that someone was waving it on or it didn't see the flashing light at all.

I *think* that number actually is counting them, though it's hard to be certain. I'm pretty sure servers are outnumbered by PCs by a large margin.

Cameras and LIDAR. I am not a self-driving car engineer, but from what I understand, it seems likely that it is possible to detect water with even just cameras, at least under the right circumstances, and with cameras plus LIDAR under a lot more circumstances. But doing so would require proper training data; it's not like there's a "Ooh, that's water" recognizer built into the hardware or whatever.

More to the point, they would have to train it how to recognize that some particular sensor return pattern (e.g. zero LIDAR reflections off the ground) is a problem, and do so in a way that doesn't over-correct for flooded potholes, an inch of water in the street, etc. Presumably, detecting water is the easy part; detecting the depth of water is the hard part unless you know exactly how high the curbs are.

I'm a little spooked when I see Tesla FSD beta demos where the car plows right on ahead through slightly flooded streets as though there's nothing there. It makes me wonder whether it saw it and ignored it or just didn't see it. It's the same feeling I got this morning when someone was signaling me to go ahead at a flashing red light and FSD beta (12.6.4) went right on ahead. I wonder if it somehow saw the hand gestures, or if it just didn't see the flashing red light at all.

That's what makes all this stuff fun.

It's basically a $75 Raspberry Pi knockoff board with a $20 SDR radio, a plastic case and some dolphin branding that's likely to get you in trouble.

Any hardware or device driver engineer will tell you that every hardware problem, once shipped, is a software problem. :-) (Translation: Hardware bugs, once the hardware is in the wild, have to be fixed with a software workaround.)

Nothing in self-driving cars is a quick fix other than geo-blocking something. You have to train a visual or LIDAR recognition model to recognize the problem situation and then train the path finding model to flag it as a bad path. How hard that is for this particular case, I have no idea. And there's probably other stuff beyond that; I'm not an expert in this area.

very much putting him on target to sign more EOs than Truman did nearly a century ago.

To be fair, Truman was ending a war and starting two new ones and Trump is valiantly battling uh... 51 concurrent national emergencies.

My understanding is that LLMs are built on a foundation of ANNs, and that indeed the backpropagation used to train ANNs is a statistical process;

Two responses. One, that's discussing individual-neuron scale processes rather than collective processes; and this was a discussion about inference, not training. Human neurons also learn by error minimization (Hebbian learning). But this does not describe the macroscopic processes that result from said minimization.

* During training, neurons develop into classifiers that detect superpositions of concepts that collectively follow the same activation process. Individual neurons weight their input space and subdivide it by a fuzzy hyperplane to achieve a classification result.

* In subsequent layers, said input space is formed from a weighted combination of the previous layer's classification; thus, the superpositions of questions being formed are more complex, as are the classification results.

* In a LLM, this iterates for dozens of layers, gaining complexity at each layer, to form each FFN

* The initial input space to a FFN is a latent (conceptual representation), as is the output; the FFNs, in result, function as classifier-generators; they detect combinations of concepts in the input space, and output the causally-resultant concepts into the output space

* FFNs alternate with attention layers dozens to hundreds of times in order to process the information, each layer building on the results of the previous one.

The word to describe that is not "statistics". It's "logic".

In a LLM, the first few layers focus on disambiguation. If there's a token for "bank", is this about a riverbank, a financial bank, banking a plane, etc? As the layers progress, it starts building up first simple circuits, and then progressively more complex circuits - you might get a circuit that detects "talking like MAGA", or "off-by-one programming errors", or whatnot. In the late layers, you have the general conclusions reached - for example, if it were "The capitol of the state that contains America's fourth-largest metro area is...", you've already had FFNs detect the concepts of fourth-largest metro area and encoded Dallas-Forth Worth, and then later taken that and encoded "Texas", and then finally encoding "Austin". And then in the final couple layers you converge back toward linguistic space.

Anthropic has done some great work on this with attribution graph probes and the like; you can detect what circuits are firing, and on what things those circuits fire, and ramp them up or down to see how it modifies the output. They very much work through long chains of logical inferences.

The old method of building chess players was heuristics. You have a bunch of engineers who hopefully know something about chess come up with rules about how good a position is. That's what "heuristics" are.

The thing that made chess computers better than any human and go computers better than all but the very best (or probably all by now) was using neural networks that learn their heuristics by experience. And yes, human mathematicians absolutely do learn heuristics, i.e. "gut feelings" regarding good approaches, promising mehtods, and quite often claim to be "close to a result." They also frequently come up with suboptimal results in the form of proofs that are not quite correct or are overly complicated, and are improved by the original author or others.

Must have.

And that would be totes different than human mathematicians who pick a problem and work at it it and only it until they succeed.

Dem gaps are a closin hey?

Yeah, there's not enough crack in the world for that to make sense. The cost-to-orbit is well documented. Artemis 2: $4.1 billion for a launch that can lift 95 tons to LEO. SuperHeavy/Starship: $90 million for a launch that can lift 150 tons to LEO. Ignoring minor details (e.g. that nobody is willing to expend a SuperHeavy to do trans-lunar injection), the reality of the matter is that the cost per ton to orbit for NASA rockets is 72x what it is expected to cost for SpaceX rockets.

And that's consistent with Falcon Heavy costs, too, so you can't hide behind "Yeah, but SuperHeavy and Starship don't work yet", because unless your payload is huge, SpaceX is still almost two orders of magnitude cheaper.

I wasn't replying to you.

LOTS of people here have been skeptical that AI can do X where X is pretty much anything, and certainly where X is "formulate novel math proofs."

PS: I don't disagree with you that Sam Altman claiming something isn't good evidence. That was not the subject of the post I replied to and has nothing to do with my reply to not your post.

I use every style imaginable, including photos, in my tests. Same result every time.

One time I even did it with a Calvin and Hobbes comic, pretending than an AI made it. Responses included things like "The illustration also looks like shit and barely makes sense. Hope that helps.", "God damn this sucks so bad", "This also fucking sucks", and "The only punchline here is casual, pointless cruelty. if you think this is funny then you're literally a psychopath."