The trade-off between quality and speed of development has always been a tenet of software engineering. Agentic coding changes the equation a lot, but the equation is still there.
> The trade-off between quality and speed of development has always been a tenet of software engineering. Agentic coding changes the equation a lot, but the equation is still there.
The quality and speed tradeoff doesn't matter if other incentives aren't aligned. In today's product-driven world, where usage stats are compiled in real-time, features are added and then pop-ups, nudges, etc are added to software to get product usage up. These features may or may not add quality, relevance, etc. But the "success" is measured in usage, forced or not. This is essentially Microsoft today (especially with copilot), but even after every major apple OS update, I get "what's new" pop ups for every Apple app (notes, reminders, mail, etc) on every platform (iPad/iPhone/Mac) that I own. I opened the email app to check my email, don't get in my way!
Higher TPS by itself wouldn't remove the humans in the loop. Humans in the loop aren't there to make things faster, but to make them right - current LLMs can't do that on their own in most real world projects.
On the contrary, I would say that higher TPS would make humans in the loop much more productive, because the current long wait times mean that you lose focus and flow.
Schengen is pretty strict in that it's a full people/money/goods zone all together. It'd require Canada pretty much all going in on EU laws.
Much more likely is looser employment visas and maybe allowing Canadians have an easier time moving for things like retirement, etc. I'm not holding my breath, though.
This is pretty a common misconception. Schengen area only refers to the fact that there are no border checks between two countries which are in that area - it's just like an internal movement. This applies to anybody, so for example if a Canadian enters the Schengen area, they can move to any other country within that area without checks and without requiring an additional visa.
What you're thinking about is the "single market" - which, even if the name could deceive, also applies to people, including the right to go work anywhere within the single market for citizens of single market countries. Which is what many brits always hated, as they wanted single market for services and goods, but not for people.
Single market doesn't imply Schengen though - eg I as an EU citizen have the right to move to Ireland for work, but I still need to go through border checks as Ireland is outside of Schengen (because they have free movement with NI).
As someone who is very much in favor of this, let's remember that for now we're only at the announcements level. Any binding changes in the EU require a long process - if anything else, because there are 27 member states, each of which have veto power. But I'm hopeful that facts will follow these announcements.
That's true, however, under all the recent pressures, everything is much more fluid. Especially because 'associate member' does not exist yet as a status, I think they could do things much more incrementally compared to full memberships.
I have no idea about timelines, but the current LLM architecture has no mechanism which could emulate online learning in a way similar to how it happens in humans and other animals. As another commenter wrote, there is no neuroplasticity while these systems interact with their environment in normal usage; adding information/constraints to the context partially mitigates this, but it's a completely different mechanism.
The RL phase is the most similar mechanism I know of that comes to my mind, but I'm not sure it could be adapted to fill that gap.
My totally unsubstantiated theory is that this is the missing link towards what most humans would consider AGI. I don't see this as intractable, but it may require some substantial change in architecture.
We've had that concept for quite a long time now, in the form of Lora [1] and similar fine-tuning techniques.
It first got popular for StableDiffusion to teach the image generation models new concepts.
We could easily live in a world where you can train / build Loras to encompass your entire code base history, company knowledge base, new skills, etc.
Then the models would start with a baseline that already has all the important knowledge without needing to cram it into the context.
This still isn't on the fly learning, but you could imagine daily or weekly training runs to regularly incorporate new knowledge.
I think the main reason this hasn't happened yet is that the shared batch based efficient serving architectures used today wouldn't support that structure well.
Yes, that could be a way to achieve that. After all, even humans have short term and long term memory, and it looks like sleep is a very important "tick" to connect the two, so it's not all just continuous.
For sure, doing it for all users would be economically unfeasible. I wonder if the labs are experimenting with something similar, though.
Chain of thought reasoning is really good, true, but agents really took off with tool calling and being able to integrate with a bunch of stable tools.
In some sense, creating good new abstractions externally and learning those is a form of "learning", on a very large timescale. The AI model is not necessarily doing the whole "look at the whole space holsitically and find a key invariant", but if you let other people do that you can enable new capabilities that were previously unknown.
Given how important matrix multiplication with a huge number of fixed parameters is becoming, there is an enormous incentive to design much more efficient architectures where this very simple compute is colocated with memory. Inference cost would come down a lot.
With the size of these matrices I don't think they are even meaningfully colocated with themselves in memory. You'll end up with some dataflow TPU architecture anyway because you'll have to stream the second matrix to multiply against.
Exactly where I see this going as well. Sure, a smartphone might be a nice place to introduce such tech. But matrix multiplication is literally where all non-labour jobs are going - this is the bedrock for efficient (time, energy) machine learning and inference.
If AI really is going to eat all our jobs, then matrix multiplication in memory is almost a requirement.
reply