People thought neural networks were just an interesting thought exercise a few decades ago and not for practical ML problems, and look what happened since then.
Population in developed countries is already decreasing, so who knows what happens after that? Unfortunately a lot of the foundations of our economy are built on top of an ever-increasing populace.
No, absolutely the opposite. LLMs are terrible at things that require judgment and justifications, because they don't reason. They come up with something that sounds plausible.
That's not good enough when you're dealing with matters that can lead to civil or even criminal liability. Errors can be incredibly expensive to fix, if they can be fixed at all.
With a CPA or attorney, you at least have recourse if they screw up. You don't with LLMs.
You have the same problem that you have with legal LLMs; an LLM is incapable of providing legal or regulatory-involved advice, and anyone using an LLM for such purposes (even leaving aside hallucinations) forfeits any justifiable reliance defense. There's a role for LLMs, but no one with legal responsibility over reporting could or would possibly rely on an LLM for complex regulatory and rules analysis, not when there's the risk of your wardrobe being replaced with orange jumpsuits.
Yeah exactly. This is where an LLM could really shine. The trick though is consistency and that it’s often more on the basis of how the organization typically treats something and rationale to its applicability to GAAP. The creation and consistent adherence to internal standards and providing them and proving them to auditors is the key and LLMs would need infra to accomplish this.
I think F1 got significantly more popular in the past few years with Drive to Survive on Netflix, and then most recently with the F1 movie on Apple TV.
It’s a sports league with history and has been around for a while, but I think significant popular mindshare only happened in the last 5 years.
I think that heavily depends on regions. In Germany it peaked with Michael Schumacher. Later drivers like Vettel were successful, but didn't attract the same mainstream attention.
But in global terms F1 tried to grow it's reach to China and US. (Which then turned to "night time races" for their traditional European audience.
Are Amazon and Meta the ones losing out the most here, in terms of the companies building foundational models?
Probably more understandable for Meta, since they've been leaving the B2B space since Workplace has been sunset. Amazon losing out on this is pretty rough for AWS though.
Is Amazon trying to build a competitive foundation model? From what I can see AWS is instead focused on hosting and re-licensing Claude, Cohere, DeepSeek and others via Bedrock. And it's pretty likely that a large chunk of this $200M will anyways go to AWS. So I'd hardly call them a loser here.
Amazon has a number of foundation models under the name Amazon Nova, which they claimed were SOTA on release but I haven't heard much at all about them since.
Most of US government runs significant workloads on AWS now and that’s only increasing. They’ve cornered govt cloud infrastructure (with Azure, GCP, etc. very far behind) so not sure this matters in grand scheme of things.
Anecdotal based on industry experience, no citations.
My hot take - this isn’t that much different than English speakers not being able to write in cursive anymore. It’s just not something that’s as practiced as much now that we have digital input methods.
iOS development has been around for quite some time now. Most senior iOS and Cocoa developers probably started with Objective-C before slowly migrating codebases over to Swift.
#2 is a slippery slope if you don't do it properly.
You might look end up looking at lots of different slices of your data, and you might come to the conclusion, "Oh, it looks like France is statistically significant negative on our new signup flow changes".
It's important to make sure you have a hypothesis for the given slice before you start the experiment and not just hunt for outliers after the fact, or otherwise you're just p-hacking [1].
Fundamentally, you can't use the same data to both generate and validate/disprove a hypothesis.
Srgmenting and data dredging is fine provided you run a new test with fresh data to validate if there is a causal relationship in any correlations found.
I agree, as per example and point number one, if your goals was to increase conversions, you were successful. You can then go to the next step, slice the data up, and iterate on another change. If you fall into the box of over-analyzing you will probably find all sorts of irrelevant patterns.
