Live data from Hacker News

Amazon R&D spending is more than that of all companies and government of France

twitter.com

141–150 of 161 posts

Re: Amazon R&D spending is more than that of all companies and government of France

#141
post #2

I suspect its just a tax break related policy, ive been doing the same shit at US companies (still in EU), and some European companies (in EU) and the US one were the only one asking me to file reports for hours classified as R&D, maybe european ones have no incentive to do that Id say, considering that, and considering that for these corps taxes are lava, i’d go to look at incentives on taxes about R&D filing

At least for public funding (subsidies) EU companies are required to have full hourly accounting of time spent.

Re: Amazon R&D spending is more than that of all companies and government of France

#142
post #107
post #28

Earlier quoted context omitted.

Can confirm, filed some “R&D” hours at a company some years back that did software development but nothing I’d call “R&D”. They straight-up told us it was a requirement for some tax benefit or other. Like they did have us do it only on non-client work, but… it was not really R&D the way anyone means that.

The D in R&D literally stands for development. So 'software development' would fit.

What is R, D or non of the above is a spectrum. Public funding and tax favoring regimes in the EU mostly use the TRL framework for classification.

https://en.wikipedia.org/wiki/Technology_readiness_level

Re: Amazon R&D spending is more than that of all companies and government of France

#143
post #54
post #24

Using the very same numbers, Amazon's R&D would be 11% of all US R&D. I suspect the numbers are wrong.

Ornas others pointed out, Amazon is boosting its RnD reporting.

I mean, sure, maybe they are cooking their numbers.

My own hypothesis was that the first table and the second table don't measure the same thing. Which would mean the headline is comparing apples to oranges.

Re: Amazon R&D spending is more than that of all companies and government of France

#144
post #136
post #119

Earlier quoted context omitted.

You would, yes. Would you believe what drives more sales and what you think you want could be different?

I don't think your argument holds water. A while back I read some article about how search is handled in companies like Amazon and surprisingly one requirement they have is appeasing key sellers. There were major political battles driving the decision on what search results should be boosted in specific search keywords just because a representative for company A insisted quite strongly that should be the expected res…

I'm not saying it's optimal in any way, just that it's a significantly harder problem than people with their ten seconds of thought about it implies.

Re: Amazon R&D spending is more than that of all companies and government of France

#145
post #67

Earlier quoted context omitted.

It is a human problem, there's not necessarily a technical solution. How do you define a Adidas shoe? Something with the same name could easily be fake

An Adidas shoe is a shoe produced by Adidas.

Awesome, is your suggestion for the algorithm just : $query = $producer ?

Can you see where this would not work, or how it can be exploited ?

Re: Amazon R&D spending is more than that of all companies and government of France

#147
post #135

Earlier quoted context omitted.

France has a giant research tax break that is very wildly misused by corporations who put development in it. There are even service companies with “research centers” now. They simply bill employees without a client to the state for doing “research” that mostly consists in completing tutorials on the latest trendy technology at the HQ.

So instead of spending $0 on not having that employee, the company would rather hire than employee for $50k to get back $25k in tax savings? This doesn't seem like a profit-driven decision.

The thing is $25k of software production is left at the company, and when the employee is not needed anymore, then the collectivity has to pay for unemployment

Re: Amazon R&D spending is more than that of all companies and government of France

#148
post #108
post #54

Earlier quoted context omitted.

Ornas others pointed out, Amazon is boosting its RnD reporting.

Why isn't the rest of the US boosting their reporting?

I suspect the same reason why not all US citizen pee in a bottle

Re: Amazon R&D spending is more than that of all companies and government of France

#149
post #135

Earlier quoted context omitted.

France has a giant research tax break that is very wildly misused by corporations who put development in it. There are even service companies with “research centers” now. They simply bill employees without a client to the state for doing “research” that mostly consists in completing tutorials on the latest trendy technology at the HQ.

So instead of spending $0 on not having that employee, the company would rather hire than employee for $50k to get back $25k in tax savings? This doesn't seem like a profit-driven decision.

That $0 is on a day by day basis. "Riding the bench" in between project bookings is common, especially for junior consultants with limited skills.

Bench / sub 100% utilization is one reason consultants cost so much.

Re: Amazon R&D spending is more than that of all companies and government of France

#150

And yet Alexa still sucks compared to modern LLM's

Seems to me like products with entirely different use cases. Alexa's not so good at generative text content, but ChatGPT isn't very good at triggering kitchen timers and home automation shortcuts via voice command.

It's quite capable of doing so though: https://johnthenerd.com/blog/local-llm-assistant/

There is nothing technically stopping ChatGPT from hitting up endpoints with JSON other than OpenAI's forced limitations.

It has excellent voice to text, excellent text to voice, excellent comprehension (it can understand vague commands like "set all the lights in the house to a Christmas color theme") and it knows how to generate JSON in a particular format that you can specify. That plus an output scraper that knows where to send the JSON to is pretty much all you need to make this work. It's so close to a commercial product that I can literally taste it. And yet here we are, having to hack together our own solutions... with local LLM's that are about as good as GPT 3.5.

Post reply on HN