I wonder how much work it would be to actually handle them being underground/overground at appropriate places? e.g. the District Line and/or Picadilly near Hammersmith. Would be cool!
Agreed - "underspecified prompts" being listed as a failure of the tooling is not a strong case. Even interns can understand ambiguous asks with a bit of help, and understand when they need to stop and ask instead of just carrying on. They are often working fairly independently on ambiguous tasks before the end of an internship, too.
So is the argument that frontier models are not just junior engineers, but first-month interns with no capability of progressing beyond that level?
Of course they are. It's a good reminder that non-profits are very often just an arm of capital, whatever their legal status is, and that they will exercise capital's authority to try to suppress workers' rights.
Nothing really wrong with being a Luddite. The Luddites weren't anti-technology or anti-progress. Rather, they opposed deployment of technology in the workplace which they believed would reduce their pay, reduce the quality of their output, and make it easier to exploit vulnerable people such as children. They were, basically, doing the same thing as software engineers who argue that AI mandates are making their lives worse and making the quality of their employers' software worse.
Can't speak for GP, but I wouldn't - privacy is already eroding at a startling rate, and more KYC for things that really don't need it is just a further affront to human rights. (See also the FCC's recent request for comments on requiring government-issued ID to use a cell phone.)
I'm concerned that developing better metacognition is really just throwing more finite resources at the problem. We surely don't have unlimited compute, or unlimited (V)RAM, and so there must be a wall here. If it could be demonstrated that this improved metacognition was coming without associated increases in resource utilization, I would find these improvements to be much more convincing... but as things stand, we're very much not there.
(There may be an argument here re: the move from dense to MoE models, but all research I am aware of suggests that MoE models are not dramatically more efficient than dense models - i.e., active parameter count is not the overriding factor, and total parameter count is still extremely important, though it does seem to roughly follow a power law.)
And yet, the ruling class seems quite happy to punch the poor - and this is not dystopian? Let's not get into the tolerance paradox here, because if someone is already getting punched, and the puncher refuses to stop... well, yes, it's okay to punch the puncher.
Software engineers are in the same class as the people below them - the working class. The entire concept of "middle class" originates from a time when the middle class were non-nobility who were, nonetheless, sufficiently powerful that they needn't worry about things like "keeping their jobs", whether because they were their own employees (as were nearly all doctors, lawyers, etc.) or because they had sufficient social capital not to worry about such trivial things as paid labor.
I want to be clear here: Eton boys were (and are) predominantly middle class, not upper class. In the US, we allowed the idea to be perverted, perhaps because we do not have nobility, and so there is no true "upper class". Given this, the reality is that we are bifurcated into a working class and an owning or capitalist class - though, many would argue (correctly, in my view) that we are in a feudal regime now, rather than a capitalist regime.
To put perhaps too fine a point on it, software engineers are house slaves, and, yes, CEOs and billionaires have done a good job of convincing the field slaves that the house slaves are their enemies, and of convincing house slaves that the field slaves are inferior and just want to take what the house slaves have without working for it.
So... a tiny fraction of people get to capture the value again, and at even greater environmental (and thus societal) cost than before? Wow, what a world.
We've had enough advancement to change the economy for many decades, but the powers that be have insisted that, despite the lack of need, we continue to toil doing completely unnecessary work, because that's what's required to extend their fiefdoms.
Not that the singularity has any relevance here, either - except maybe that the robots take over, and the billionaires have missed the boat? I don't know.
Humans are also non-deterministic, though. Why does replacing one non-deterministic actor with another matter here?
I'm not particularly swayed by arguments of consciousness, whether AI is currently capable of "thinking", etc. Those may matter right now... but how long will they continue to matter for the vast majority of use cases?
Generally speaking, my feeling is that most code doesn't need to be carefully-crafted. We have error budgets for a reason, and AI is just shifting how we allocate them. It's only in certain roles where small mistakes can end your company - think hedge funds, aerospace, etc. - where there's safety in the non-determinism argument. And I say this as someone who is not in one of those roles. I don't think my job is safe for more than a couple of years at this point.
I mean, Tesla gave up on quality self-driving many years ago when Elon went hard against LIDAR. He's never relented, either, and I don't foresee that changing.
On the one hand... the operator who started the agent is responsible. If you fire a gun into the air, the bullet will probably not land on anybody. It's pretty unpredictable what will happen unless you have perfect data and forecasting. But it's still your fault if it kills someone when it comes back down.
But on the other hand, with so many tech companies pushing their ICs to use AI - and pushing more and more to use agentic tools - I'm generally in agreement that the fault lies more with the employer than the employed. You can't tell someone to use a tool, then punish them when the non-deterministic tool they didn't want to use in the first place misbehaves.
Maybe if we had decades of history showing these tools could be operated safely this would be different. But asking an entire industry to adopt new tools with minimal training and real-life experience, and asking them to adopt them at scale and in production, is not the same as asking them to use DynamoDB instead of RDS.