coderenegade

This is a shame, and hopefully it reverts soon. I switched from Claude to Codex just before the Fable release because Claude doesn't follow instructions and goes off the ranch too frequently. I really don't want to have to switch back to what I consider to be a fundamentally worse model.

This is my experience with Claude as well, and the reason I switched to codex. Codex just seems far more efficient, despite the smaller context window, and it actually follows instructions.

And those of us working in the slopfields will probably be employed until the day we die. Love it.

This. German really isn't that difficult for an English speaker. And if someone really wants to make a home and build a life in a foreign country, why would they expect that they don't need to learn the native language? That seems stunningly entitled to me.

No, because the thread is about preventing crimes before they happen, and asserting that this is unquestionably good and an unambiguous moral position. It isn't, even if we accept that widespread enforcement of the law is generally something that we want.

Fair enough, I was under the impression that enough data was sufficient.

1) The only difference between a just law and an unjust law is time. I have family members who needed to wear dog tags to drink in certain pubs because of the colour of their skin, and I'm not old. And yet such laws were widely supported (for various reasons) for a long time.

2) lazy governments often apply sweeping blanket bans and fall back to police discretion. Carrying a hammer? Technically illegal in my home country, and the police can stop you. Whether or not they have have reasonable cause to charge you is a different matter. If you look like a tradesman you probably won't be bothered, but if you don't, you're breaking the law.

3) Laws can be mutually contradictory until legally tested and a precedent is set. If you're unlucky, you could be the one who has to test it. Regulations are notorious for this.

I'm not saying that we don't want to enforce laws, because generally we do. But when and how laws are enforced requires nuance. And certainly the idea that always enforcing them is a net good is very, very far from the historical reality.

It's definitely not. The law is not, and never has been, a moral bar. A healthy legal system requires laws to be able to change over time to reflect society, and sometimes this requires that people be able to break the law.

It would depend on what the data says. By the time putting such systems into civil airframes is even a conversation, there will be an enormous amount of data generated by unmanned systems and military vehicles. If the data shows that a given system is acceptable, I don't see any reason it wouldn't be allowed.

That's not the same as saying that current architectures are acceptable, because clearly they aren't. But both neural net and stochastic controllers have been tested in aircraft since at least the 1990s. Probably earlier, but that's as far back as my knowledge goes, anyway. I don't know of any neural nets doing active control type tasks, but I don't see any intrinsic reason they couldn't be used as long as you can show that they're good enough.

When I last looked into it (admittedly a number of years ago now) this was true for remotely piloted aircraft, but there wasn't enough data on systems that pilot themselves with humans doing the navigating.

I'm not convinced that humans are particularly effective in emergency situations, especially when they're expected to go from monitoring systems that work 99.99% of the time, to suddenly aviating with alarms going off everywhere. I guess there's not really much of an alternative as it currently stands, but seeing how effective agentic AI is at chewing through data and analyzing systems makes me think that eventually, such systems will be used if they ever get to the necessary level of reliability.

It's unlikely; the energy density just isn't there, even for a hybrid system. Because of the way flight profiles work, you're still talking about enough battery to provide meaningful power for half a flight, which probably isn't feasible in the near term.

And it's not like the industry has missed this possibility. I've seen papers on systems as exotic as combining high temperature fuel cells (SOFCs, PCFCs) with the Brayton cycle to achieve a hybrid powerplant. The bar for novel propulsion systems in aerospace is extraordinarily high, higher (imo) than it is for automotive. The exception is unmanned systems, which are new enough and varied enough that there's been an explosion of activity, which has been exciting to see.

Aerospace as a discipline has tried just about every propulsion system under the sun. What you're proposing has already been flight tested on an unmanned vehicle, albeit with many smaller props for a larger effective prop disk than the two you're proposing. This is actually better, because electric motors want to use rpm control, so you need to keep the moment of inertia of the propellers low. The efficiency penalty of smaller props is overcome by having many of them arranged closely together to create a large effective prop disk.

For a hybrid system to be worth it, you need to claw back more efficiency than you lose in going from mechanical energy to electrical energy, and then back again. For cars, this is generally the case, because they're always accelerating and decelerating. Their wide operating band means that the engine will always be a game of compromises, which is why sticking a motor and battery in the loop and decoupling the engine from the wheels is beneficial. But planes aren't like that; they go from setpoint to setpoint, and they stay in a given configuration for long periods of time. They have very narrow, highly optimized operating bands, so hybridization just isn't as effective.

The philosophy in aerospace is more to build a better engine rather than to have more engines, and this extends to every aspect of aircraft design. Engines are already built to contain catastrophic failure, and the planes themselves remain functional for all but the most extreme situations. We're at the point where essentially every lost aircraft is a compound failure, with significant human factors contributing to the event. It's likely that we're on the pareto front of what engineering can reasonably accomplish, and the only gains in safety either barely nudge the needle of what we would notice (better materials, say), or difficult for the market to accept (removing pilots altogether).

Aerospace RND has been looking into hybrid propulsion systems for a long time. If there's one thing they aren't shy about pushing, it's the ability to go higher, faster, more efficiently. Such systems aren't used because they aren't yet good enough.

The economy is human-serving and human-legible because humans naturally create economies. Even in the most fanciful pie-in-the-sky projections of AI, human economies will still continue to exist and function, even if it's in the form of bartering or using side currencies the way US dollars are used in many developing countries today. You can't stop people from exchanging goods and services, and the need for that will exist until the end of time.

What's more likely to happen is that the economy might split. Organizations that have no need for human labor or input are essentially islands unto themselves. The only remaining economic link is the substrate -- the land we all inhabit -- is shared.

I'm not sure how that works out (and indeed, that's the worrying part), but what I do know is that human economies will continue. It's even possible that a split might be a good thing, because right now, our currencies span such vast scales of value that it's almost impossible to reconcile them all. Governments use economic health to both drive and act as a signal for the effectiveness of their policies policies, but what happens if the value created by organizations that only employ a handful of people vastly outstrips everything else? You could lose famines, plagues and homelessness in the noise, because the people economy no longer matters. And it's arguable that this is already happening in many countries, which is why so many voters feel like they're not actually being represented, i.e. they're not, because they already don't matter.

Fwiw - you already use defensive everywhere. Python, Java, etc. come with garbage collectors. It's verified that the code is executing your intent.

Sort of. Garbage collectors can be fallible too, especially where release optimization is used.

I'm honestly having a hard time visualizing the technique some of you guys seem to be using.

I have a few black wattle rounds that have been sitting around for years. I have a go at them whenever I feel like I need to be humbled. There's also a fallen tree at the bottom of our property that blunts chainsaws. It's been there for years and nothing seems to eat it. I harvest what I can from it, and a good sized chunk will burn through the night.

Series hybrids can compete with parallel hybrids, because the full decoupling of the engine from the wheels claws back some of the efficiency you lose through the energy conversion. It's series-parallel hybrids they can't really compete with, because those are able to do the same trick, but they also lose less energy in conversion, because the engine does some of the heavy lifting.

Series hybrids are great for packaging, though. Parallel and series-parallel commit to certain packaging decisions like having a transmission, or a long, monolithic unit, because it's the mechanical coupling that buys them smaller motors and potentially better efficiency. Series hybrids don't care about any of that, so even though you have bigger motors and potentially higher losses, you have more freedom over where things go.

Personally, I think there's a massive untapped market in converting old cars to hybrid engines. You wouldn't try to upgrade the old engine, you'd design a smaller and more power dense package and rip all of the original gear out. Because electrification lets you cut the size of the engine down so aggressively, this is probably a feasible strategy. As you pointed out, series hybrids are probably best suited to this because of their packaging flexibility. As others have pointed out, there's tremendous potential there for replicating original driving characteristics using software and the electric motor. And if we're being honest, off-road vehicles probably should get rid of the transmission and low range, because electric motor torque is just better. As is, the potential for cars is enormous, but we're getting the worst possible outcomes thanks to legislation.

It's hard to know for sure. There are good information theoretic reasons to suspect that general models will always be better than smaller expert models, but maybe a MoE can claw some performance back, albeit with redundant computation. The properties of conditional entropy, for instance, always favor more generality. This assumes that the harness isn't a factor, or is at least equivalent across different models.

Yep. These days, simplicity is a massive part of my development style. I don't want to be looking at a codebase, even my own, and thinking "shit, this guy was way smarter than me".

There are good information theoretic reasons to suspect that general models will be better than specialized ones, because knowledge and skills often overlap different areas, sometimes in surprising and unintuitive ways.

And yes, I'm aware that that statement might seem to fly in the face of much of the past two years of industry development, where specialized models have been in vogue. I think they'll settle to being appropriate for low cost "good enough" applications, but I'm less convinced they'll have anywhere near the fidelity of larger frontier models.

It is, but they have different use cases. CadQuery uses a geometry kernel that does boundary representation, which you need for path generation for modern manufacturing tooling. OpenSCAD produces a standard mesh representation (i.e. triangles), which is insufficient for cutting and subtractive manufacturing, but often fine for additive manufacturing (3D printing).

The context here is that at one point, IBM was an innovator and global leader in the technology space before it got outcompeted. It was the first company to cross a $100 billion market cap. If you think of the IBM of 50 years ago as being roughly analogous to Apple today, the difference is pretty clear. Google is much closer to Apple than it is to IBM, and I don't see that changing.

IBM got to where it is today by being complacent and not keeping up with innovation. Google is notably at the forefront of innovation, in a driving seat, and that innovation directly stands to benefit one of their core businesses in a way that the market is probably only just beginning to understand. It's an entirely different situation, imo.

Google is in an incredibly strong position. They're a top tier AI vendor, and in a world where content creation is largely commoditized and outsourced to AI, advertising companies will determine what gets seen, and what gets buried in the noise. They control both generation and visibility of what gets generated. Facebook could be in the same position, but they aren't as strong in AI. OpenAI wants to be Google, but they don't have the advertising reach.

Yeah, they aren't perfect or always necessarily the best in a given area, but to compare them to IBM is probably missing the forest for the trees.

The policy is how you select your actions -- in this case, the next token. It can be random, but it doesn't have to be. "Deterministically choose the best action" is a valid policy (we would call it the greedy policy), as long as you have some other means of injecting stochasticity so the model explores the space. Uniform random is also a valid policy, as is always selecting the same token (it obviously wouldn't be very performant, and would defeat the purpose here, but it might be fine in, for example, a multi-armed bandit scenario). Most of the time, the policy is a parameterized distribution, and we want to learn the model parameters that maximize some measure of success (the reward component).

Off-policy versus on-policy refers to what data the model is trained on. On-policy training is where the training data is collected by the policy. Off-policy training is where the data was collected by a different sampling process (e.g. we have a standard dataset that we're going to use for supervised training).

The distillation risk has been brewing for a while now. In a very real sense, the model is the data, so if the data is locked down because of how valuable it is, it was only a matter of time before fully open access to the models would be revoked.

There's also an additional economic concern that rarely gets mentioned: because no one has cracked continual learning, keeping models up-to-date and filling in gaps in performance requires retraining on an ever growing dataset. Granted, you aren't starting from scratch each time, but the scaling required just to stay relevant looks daunting.

I don't know where any this goes on a societal level, but I've believed since the release of deepseek r1 that access to frontier models would eventually be locked up behind contracts, since the only moats protecting the models themselves are purely artificial. It remains to be seen how effective China is at pushing the envelope, and whether they are interested in providing unfettered access. And on top of that, it remains to be seen how well these models actually turn out to scale in the long run.

You only get good at the things you actually do. Our ancestors had to maintain a minimum level of fitness in order to be able to eat -- a level that most people today never reach, because the modern world has removed that need. Thinking is a skill just like any other, so what happens when people no longer have to exercise that skill to survive? It's a scary thought.

The problem is that if you don't stick to truth and make an attempt at objectivity, others will step in to fill the void. This is how you sow division and undermine trust in science.

If anything, the golden age of third places coincided with the golden age of suburbanization, which was obviously heavily car dependent. Their death almost certainly has more to do with financialization making it harder for small businesses to stay afloat, a drop in demand due to competition for attention, and decreasing work-life balance eroding people's ability to socialize.

In my grandfather's day, one income was enough to support a household, and there was less free work being done on the job, which meant fewer hours and being less drained at the end of the day. And yes, people spent less time commuting, meaning they had more time and energy for socializing after work. But communities were also more decentralized, and population centers had fewer people in general. A big part of the problem is that modern cities can be massive, and invariably funnel people to a handful of work districts, which just doesn't scale. When you double the distance to the CBD, you quadruple the number of people coming in (give or take, it's not exact because we tend to increase density close to the CBD as a response to this). Take it from someone who's lived in a place where cars aren't really necessary, the logistics of urbanization are still a crap experience when you're crammed into a train carriage during rush hour. It's common for people to commute for 90 minutes on public transport in Asian megacities, for example.