Tool calling is, yes, but the current approach to it is (a) overt not transparent to the user, (b) not designed for the purpose I've outlined, and (c) not remotely good enough for the AI to be able to manage data through such tools.

Yes, you can connect an AI to a PostgreSQL database. Whoopee. Not even close to an AI detecting via classifiers that some of the data is relational in nature, transparently setting up its own PostgreSQL database in response, and using that proactively as an additional way to examine the data. You would need to explicitly set up the constructs, explicitly set up the databases, explicitly tell it when, where, and how to use the database, and if that particular usage conflicted with a way the AI actually worked, the AI would not be able to use it effectively. That's not remotely close to what I'm talking about.

What I'm talking about is much, much deeper than that. AIs lose focus when there is too much information currently in the system. Absolutely nothing stops an AI from using a document database as a virtual memory in which it can page in and out conceptual spaces so that the focus on any given step of a problem concerns just the problem. Other than such a concept not existing yet, which is kind of a limiting factor. But you need to know what connects to what. The AI could use an ontology reasoner for that, or a relational database, or an external graph. But you cannot provide those tools and you cannot configure them, for the simple reason that the AI is the only entity with any detailed map of how the underlying neural net is connecting those ideas up.

The AI itself has to select the tools, has to configure the tools, has to do all the work. Which means YOU cannot use any sort of API. YOU should have no involvement in the underlying mechanisms needed for the internal NN housekeeping.

And that's your other error. You're assuming this is about user data. No. It's about housekeeping operations within the NN itself to maximise functionality, it has nothing to do with the user side of things at all. If the user was capable of setting up dynamically structured databases that mutated with each step of a decomposed analysis, you'd have done everything the AI would be doing and you wouldn't need the AI to begin with.

You cannot do this work, you cannot even assist in this work, it has to be dynamic and it has to be utterly invisible to the operator.

We don't need bigger models, at this point. What we need is multi-dimensional decomposition, problem space transforms, and the ability for AI to use external tools (such as SQLite, memcached, etc) so that it can externalise static data that it needs to not corrupt accidentally but still keep in easy access.

If we had that, most of the things "bigger models" will do will actually end up being done better, quicker, with fewer compute resources.

To be useful in deep space, you're going to have to deal with very harsh radiation, far harder than the Vikings dealt with. You really want to have some large number of computers, where the number has to be odd and exceed 5. The reason it has to exceed 5 is that you need 5 in order to be able to use the Byzantine General's Problem to discern which computers are working correctly and which are radiation damaged. Since some will be damaged over time and you want 5 computers still operating around the time the power runs out, you have to start with more than that.

Massively radiation-hardened highly-robust low-power chips could be done. If you were clever enough, you could do this as a wafer-scale system where you marked parts as bad and networked around them. Free space optical communication across the wafer might be doable.

They might well still end up being low-density CMOS discrete logic.

Lead-lining the electronics isn't as much of a problem if you've launchers capable of handling heavy objects and don't mind burning through a lot of ion drive propellant for course corrections. This would improve the hardening beyond what can be done through the usual operations.

But, yeah, you'd need defect-free microelectronics. You really can't handle F00F bugs or defective instructions once you pass Earth orbit. And I can't think of anyone who knows how to do that.

Well, not just the solar system but the heliopause as well, and the near-Sol galactic winds. We can estimate the effects of space from just about any ancient near-Earth space junk, but I'm not sure how we'd go about calculating the impact of the galactic winds on something. The Voyager sensors won't be sending back nearly enough high-quality data to establish that.

It's actually a lot harder. The fly-bys invovled gravitational slingshots, which not just steered the probes but vastly accelerated them. Without the slingshots, human rocketry of the time would not have had the capacity to get probes to solar system escape velocity.

Our understanding of asteroids is now reasonably good. We know they come in solid and rubble-pile formats. We know that several show signs of lava flows, suggesting they come from disintegrated planetoids with a liquid core. We've managed to land on some and even retrieve samples from one.

Our understanding of comets is much weaker. We know that several appear to be multiple bodies fused together with ice (the one we tried sending a lander to was of this type) and the break-up pattern for Shoemaker-Levy 9 is suggestive (but not proof) of it having been a composite structure too. The one icy body we've seen in the Kuipier belt was also two bodies frozen together. Photographs of the core of Halley's Comet also suggests two bodies fused together with ices. This leads to some interesting questions. Obviously the probability of this configuration is very high, but what is the probability of other configurations? Is it so incredibly low that the odds of finding single-body or triple-body configurations is practically non-existent?

(This is kinda important if NASA is serious about a DART-type project to redirect one.)

Our understanding of solar system captures is non-existent. If there are extra-solar objects flying through every few years, then there is a definite non-zero probability that there are significant extra-solar objects that have been captured by the solar system in the last 5.5 billion years. That would be kinda very useful to know.

Current solar system formation models don't work well, but planets have sorted their composition by mass. Studying them only tells us so much. Asteroids are more useful but hit each other too often, so we don't know how much mixing and reconstitutuing has occurred there. The Kuipier Belt is much better, from that point of view. The density is far too low and most of the objects far too small. A lander could obtain a lot of extremely useful data about how the solar system formed that we can't obtain closer.

The Oort cloud is theoretical only, at this point. The only way to find if it is real is to send something out there.

If we're to get any probes in the future to go between solar systems (and such missions are entirely possible), we need to know a great deal more about the heliopause and the galactic winds.

Well, viruses and bacteria. And that last one is critical.

We know that the microbiome alters blood chemistry and we know that this alters brain function.

I do NOT postulate bacteria cause mental illness, but I do postulate that there is a non-zero probability that bacteria can aggravate mental illness. And that is all that we need here. If we can develop a vaccine that triggers the immune system to either attack the bacteria directly OR to produce antitoxins which attack those changes in the chemistry which are harmful, then the vaccine would lessen the impact of this particular vector.

That would not prevent mental illness, but that's fine. Reducing it is perfectly acceptable.

We know that brain function is altered by the blood chemistry changes made by gut bacteria. This means that any mental illness exacerbated by this route where the chemistry added is verifiably unique to an undesirable bacterium can be targeted by the immune system.

Will this be all mental illnesses? Obviously not.

Do I know of any such chemistry? No, which is why I said "in principle". The potential for the vector exists and the potential for training the immune system to attack it exists, but the reality of such a vector actually existing is unknown and the safety of attacking it is also unknown.

"Potential" and "Unknown" are very important words here. They mean that we do NOT know if the probability of such a vaccine being possible exceeds zero, but equally we do know that a key underlying requirement has a probability that isn't zero. But that is all that we can definitely say at this point in time.

Although it would be wonderful if we could retrieve them, realisically we can't. And if we could get something to them to fit new nuclear batteries and replacements for components that cannot be revived, we could get newer probes out there with better sensors and a more powerful nuclear battery that could last a lot longer.

Having said that, I honestly wouldn't trust any of today's manufacturers to be capable of building anything with a comparable level of robustness, so a recharge mission isn't quite as stupid as it sounds. Almost, but not quite.

Still, fresh probes sent into the Kuipier Belt and the Heliopause would be extremely interesting.

So a murder suspect can appear in court and explain that someone gave him the wrong photograph, so it's not his fault?

The US, and many other countries, need to put forward a new law -- if an excuse applies to officials, then it applies to the citizenry. This is not because officials don't sometimes need extra freedoms, but because the vast majority of manmade disasters are "perfect storms" and because this case is a perfect illustration of the banality of evil.

Legitimate causes are just that and they won't cause problems. If, indeed, they are truly legitimate. Excuses mean that they knew they were doing wrong but felt that finger-pointing would confuse the issue for long enough for the storm to die down. Innocent people do not make excuses. Innocent people don't need them.

But this isn't something officialdom will ever accept. Nor will they tolerate a loss of power. So you don't take away their power, you merely insist that powers they claim are the same everywhere for everyone, and that the powers they deny are likewise denied to all. The sensible ones will see in seconds precisely where that's going to go and will get rid of the stupid ones very quickly. Not because they're morally better, but because such a rule is highly volatile and self-interest takes on a very different meaning when the consequences are unpredictable.

In principle, it should be possible to develop a vaccine for certain forms of mental illness. Regrettably, no such thing currently exists and antivaxers are probably too far gone.

Adoption is a function of existing adoption and existing availability of resources.

Extremely good OS' and truly amazing computers have failed, not because they were too expensive or had issues, but because they were perceived to be unpopular OR simply lacked enough software that could run on them.

Extremely bad OS' and truly awful computers have succeeded for the opposite reason.

Linux has an extremely good architecture, at least for machines that are mostly operating with CPUs. When most of the work is done by the GPU, it's less obvious that Linux provides the necessary mechanisms to manage resources and abstract away the hardware specifics. That doesn't mean I expect Linux to start fading, but I don't think a centralised OS is necessarily where things will go. However, the reason it hasn't done better so far is simply because potential users and potential software houses had a negative view of it (even though actual users were almost invariably loving the system).

Windows is crashing now not because it is a slow, bug-ridden piece of spyware (which it is, but users have never cared about that in the past), but because it is perceived in increasingly negative light.

They can get vertical samples from space and the ground. What they can't get from space or the ground is how the internals of a system evolve.

The problem with "models" in the abstract is that there are a lot of them and some of the time the thing you want is to know the same point relative to the system and not to the mudball.

Lots of people have gut feelings about AI, positive or negative.

Most of the AI research I can see people doing is focusing on single-element problems (such as finding counter-examples to a mathematics hypothesis) or solving very simple engineering systems (at most a dozen or so components). Most of the benchmarks are even simpler (write a short story at the level generally asked in English classes of primary school kids).

As a result, I think that people have developed either a very cynical outlook or a very optimistic look. Neither of these is entirely realistic, for the simple reason that no actual useful problem fits any of those descriptions.

Digging much deeper, I'm finding that when problems get into the hundreds or thousands of elements, even ChatGPT Sol 5.6 and Claude Fable 5 struggle badly to keep track of accurate relationships, let alone any level of detail.

If you want to solve a relatively simple coupled system (whether to use solar panels or direct solar heating, and which side of the house to place them), I'm sure AI is fine. If you want AI to write a five hundred word story detailing a conversation between Kim Possible, Rufus, and a smoked herring, I'm sure it could cope. Just.

If you want it to do anything hard enough that a human could do with an assistant, the tests I've been using (meaningful rather than synthetic problems) suggest you've developed a fantastically expensive blue smoke generator. We are going to need to see the inference engines developed for previous generations of AI, and the reasoners developed for the semantic web, and perhaps tools that haven't begun to be imagined yet, before AI can do anything non-trivial.

A balloon is not static, it moves with the atmospheric system it is in. Ergo, it tells you how that system evolves. A drone cannot do this.

A drone can stay put, relative to the ground, telling you how a location's atmosphere evolves. A balloon cannot do this.

These are solving completely different types of problem.

You could, of course, construct a hybrid, as balloons are traditionally single-shot. If a balloon carries the measurement gear and a drone capable of carrying that same gear PLUS a deflated balloon, then you can have something that operates as a balloon until it reaches some pre-set location, whereupon the balloon's gas is released and the drone flies the ensemble to a collection point.

It obviously gives us guarantees in terms of what the source will do and what the compiled code should need.

These do not provide memory safety, so memory bugs would not be solved by these. They will find unexpected behavioural issues, though.

The precondition and postcondition statements are not cunningly-disguised assert statements because they're not asserted at runtime. They're enforced at static check and/or compile time.

The policy hints mean that if the kernel is doing something you're not expecting, you find this when testing the code. Equally, though, it means that users cannot cause the software to venture into realms unknown through many of the usual attack vectors because those will require permissions that aren't there.