> Of course, the AI talent war may end up being an expensive and misguided strategy, stoked by hype and investor over-exuberance.

To me, that's a pretty good explanation.

The world is crazy with AI right now, but when we see how DeepSeek became a major player at a fraction of the cost, and, according to Google researchers, without making theoretical breakthroughs. It looks foolish to be in this race, especially now that we are seeing diminishing returns. Waiting until things settle, learning from others attempts and designing your system not for top performance but for efficiency and profit seems like a sane strategy.

And it is not like Amazon is out of the AI game, they have what really matters: GPUs. This is a gold rush, and as the saying goes, they are more interested in selling pickaxes that finding gold.

▲

bee_rider 6 days ago | parent | next [-]

I guess Amazon can also probably afford to wait until somebody comes up with an application for AI that is, like, something Amazon can actually sell or use…

Customer service bots? Maybe. Coding bots? I bet they use some internally. Their customers don’t really need them, or if the customer does, the customer can run it on their side.

	▲	janalsncm 6 days ago \| parent \| next [-]
		As I’ve said before, the kind of AI that makes money is called machine learning. Pricing ads, recommending products, improving search, optimizing routing. In general these fall into the category of things humans cannot do at the scale and speed necessary to run SaaS companies. Many of the things LLMs attempt to do are things people already do, slowly and relatively accurately. But until hallucinations are rare, slow expensive humans will typically need to be around. The AI booster’s strategy of ignoring/minimizing hallucinations or equivocating with human fallibility doesn’t work for businesses where reliability is important. Note that ML algorithms are highly imperfect as well. Uber’s prices aren’t optimal. Google search surfaces tons of spam. But they are better than the baseline of no service exists.
	▲	wiether 6 days ago \| parent \| prev [-]
		You mean something like Kiro? https://kiro.dev/

▲

janalsncm 6 days ago | parent | prev | next [-]

AI is huge, it’s just not the only thing happening in tech right now. I say this as an MLE but it seems really unbalanced that LLMs have gotten trillions in investment when other groundbreaking innovations like battery improvements or fusion power or gene therapy have gotten substantially less attention.

Disagree re: DeepSeek theoretical breakthroughs, MLA and GRPO are pretty good and paved the way for others e.g. Kimi K2 uses MLA for a 1T MoE.

	▲	bigbuppo 6 days ago \| parent [-]
		Big money investors know that real tangible products that have real tangible benefits aren't usually decimal-point-shifting-your-net-worth jackpots. They make money, sure, but factories can't be built in a day. Also, if they can make AI work as it says on the box, they'll be able to get rid of all those pesky employees and turn their companies into pure money-printing enterprises. Pay no attention to the cracks that are showing. Nevermind the chill. Everything is fine.

▲

energy123 6 days ago | parent | prev [-]

I don't agree, based on my experience trying all Deepseek models on real world software tasks.