Using FFT to dot matmul is much more memory intensive, IIRC.
CUDNN supports FFT to do matmul as well as convolution/correlation and can also be configured to automatically use the best algorithm.
In some cases the FFT method has the incidental side-benefit of data reuse, like in the case of FIR filters, where the data allows for partitioned convolution.
Yeah, if I'm reading something, I want to have an indication of how long a page is and where I am currently. I don't want to have to move my mouse somewhere to get that information.
My experience with this class of Chinese manufactured inverters are that they all use TI TMS320F28xxx series DSPs and usually without any protection fuses burnt. If you look hard enough you should also be able to find unencrypted firmware and flash it with the standard TI tooling.
I have all my untrusted devices (including cameras) on a VLAN that does not have internet access, but is reachable from my main VLAN (but not the other way around). Then I have frigate and homeassistant running on the main VLAN that can connect to the cameras. I use wireguard to connect when I'm not home.
It makes sense. If you are an investor and have a strong belief that a specific product or idea is a good one - you might want to decouple your odds of success from the people/team/company executing that idea.
What's interesting is that the people who are able to predict and come up with the idea (e.g. a researcher using AI in 2021) are often not the best ones to execute on it (typically, lack of experience or capacity to handle the pain of growth while marketing).
What's most interesting is that most people aren't just "one type". In your life you go through multiple roles. Just like how most people, regardless of their income at age 20 will be earning top 35% income by age 35, people move up in their roles (and struggle at a young age to understand that this will happen to them). It's all about timing and age.
I believe some investors go for multiple teams purely because the ratio of those types is different - like a different risk profile. They invest in one with 80% AI boffins, and one with 80% business folks, and one will likely win in the circumstance of the moment (in the ChatGPT era - mostly the business folks. In the data driven AI era - mostly the AI boffins)
The general is that income, like wealth, is correlated with age, not ransom. Most workers under 18 will be making near minimum wage. Workers in their 20s are probably still starting their careers. People fall off again when they are older, as they either retire early or in some professions just become less capable.
So when you put it all together, a lot of people with an under average career will have an over average income at 35: Just not an over average income within the cadre of 35 year olds.
The lack of correlation with income at 20 comes from how many careers require training that doesn't give good early income. A future doctor, barista or AI programmer are not likely to have. a great income at 20, but their incomes and wealth diverge rapidly as some have longer training with different outcomes. The doctor will hit the 1% after residency. The AI expert might start making money earlier. The barista is probably ahead in his 20s, but it's unlikely their income grew quite as much, although many a barista is working on doing something else. So again, looking just at percentile of income at different ages is going to lead to mistakes as different life curves are being aggregated together.
Yep. I have responded in another post with some stats/surveys. Exactly, the aggregation of people hides the trends, and the fact that a 35 year old is not competing with other 35 year olds, but also with other 21 year olds working at cafes, contributes to this.
Human beings are the only creatures capable of self-evolution, using our minds to analyse the world around us and the world we have nurtured within us. Over time, we can stagnate into ossified animalistic patterns of selfishness, or we can expand our curiosity, consciousness, and, hopefully, our realization that compassion for others is the source of our own happiness. We can either find ways to better integrate ourselves into a better future for those around us, or wall ourselves off from the world around us. We would do well to remember that carpenters don't make hammers, nor computer scientists their own food, soil, packaging, tractors, or trains.
The use of the vast sums available to successful tech investors gives them a greater point of leverage than the rest of us not so endowed. And the more power is given, the responsibility is required, though few acknowledge or heed such wisdom.
At the end of the day, no matter how confident many folks are, who really has the humility and intelligence to know if a person is a genius? Very few, I guess. Very, very few, indeed.
"Most people will be earning top 35% income by age 35" seems like a patently false statement. Clearly more than 35% of the working population is over 35.
- Firstly, I admit I can't find where I heard the age 35 specific claim to reference it. I have listed two sources below and found an even stronger claim by Thomas Sowell (that most people at one year of their life in America earn top 20% income), but it wasn't that statement I was looking for, so I will keep try to find it. But income decile movement surveys in the UK supports this, as do dynamic income surveys conducted in the US, so I'll share those to you:
- Income is highly mobile. When tracking what income groups have, within groups there is great variation. Half the people with top 20% of income drop out of the top 20% over 5 year periods, with a significant proportion of those dropping out (30%) dropping to below average income. In other words 15% of the people earning the top quintile of income will earn below average income in 5 years time.
... Meanwhile, showing upwards mobility, 20% of the people earning the 20% lowest in the UK moved to the top 40% of income by just 5 years later in this study (Figure 19 / corresponding table) - https://www.gov.uk/government/statistics/income-dynamics-201...
- You seem to assume income goes up over time, however in western countries, income tends to peak at 30-50, hence 35 in the stat. It peaks at age 37 and stays flatish until it declines from age 44 in the UK, for example: https://www.statista.com/statistics/824464/mean-disposable-i...
- Historically, older people earned more in the 40s and 50s, but due to COVID or the simple growth in proportion of people economically inactive or in part time roles at those ages, you can see how in the UK in the last 20 years that trend disappeared: https://www.ons.gov.uk/peoplepopulationandcommunity/personal...
- Children and pensioners essentially earn nothing for the purposes of this stat, but as they still spend (or indirectly through parents consume) money, they're in this total. Pensioners are reasonably part of this stat because they are spending in the economy, but are seldom members of the "working population"
The main resource I would suggest on income mobility is "Chasing the American Dream: Understanding What Shapes Our Fortunes" which is based on the Panel Study of Income Dynamics (PSID) [https://onlinelibrary.wiley.com/doi/abs/10.1111/jacc.12716] - it does date to the 70s/80s, but it focuses on: can you achieve top income, can you have income security. Their major finding is that A) Most people through randomness and effort end up earning above-average incomes during the middle stage of their life, and that B) Wealth, but not income, is cumulative.
This is also historically constant, for example Thomas Sowell wrote about how most Americans at the time would be in the top 20% of income for at least one year of their life in "Economic Facts and Fallacies" and "Basic Economics". (obviously the UK quintile study notes decreased lower income exit mobility which is bad for poorer people and means dynamic income mobility is reducing).
Exactly. I think this is a brilliant strategy by YC. They know that some ideas have great potential. They just back multiple teams and hope that one of them will win with their execution.
We need to define conflict of interest. Question: is an investor who gives money to one organization, but is not involved in the decision-making, conflicted if they give money to another organization? Are they self-conflicted (undermining their own likelihood of success)? Are they contractually or legally conflicted? Have they breached the trust of people they invested in? Are they ethically conflicted?
Answers to these questions are non-obvious. Attempts to simplify the set of relevant questions means imposing a worldview.
On the ethical question, a consequentialist would say it depends on the outcomes. Like many consequentialist analyses, this is complicated. Consider this: Investing in a similar company might validate the market and make it more likely for the company and/or its people to reach viability.
>This creates an extremely obvious conflict of interest.
That's not necessarily bad though. People like to throw out these terms that sometimes have negative connotation as if they are inherently negative. If you think the conflict of interest is a bad thing, you need to elaborate on why you feel it's bad, not just pretend it's prima facia bad.
It’s not a marriage. Have you ever watched any racing of any sort? One team frequently has multiple contestants. First and third is better than just third.
The conflict of interest is simply dead and forgotten in an era where the president-elect has his own cryptocoin, his own social media company, and appoints his billionaire supporters to improve efficiency in the parts of government that directly oversee those billionaires’ own businesses.
Yes, political corruption is a drag/loss on everyone except the corrupt ones. Worse, it can shift a system out of its viable operating zone. Corrupt individuals in a market can destroy the market.
But what is the connection to the parent comment? No matter how corrupt one part of government or a market becomes, it doesn’t excuse further corruption. If anything, more corruption makes additional corruption more likely to break the whole system.
I think it's reasonably straightforward they were making the point that conflict of interests are no longer taken seriously, along with many other related things. Politics tends to be a trailer of social views, not a leader.
Individuals who voted for Trump don’t necessarily want to throw away a social contract. Many of them do support societal norms, albeit different ones. And some don’t even think in these terms; they are more motivated by other factors.
Along with many, I think their collective actions point in a direction that (a) undermines democratic rule and (b) enables Trump’s corruption, but they seem to be relatively unaware or disagree with such effects.
Many of them think Trump will combat one some types of corruption (the “deep state”).
Overall, I’m more inclined to think many/most Trump supporters have reasonable core values, especially at the individual and family levels, but due to their information sources and mental processing, their overall choices don’t bode well for us, together. The biggest breakdown I think has to do with epistemic values: how does one find truth.
I don’t think most people, of any party, have the individual ability and discipline to make sense of a modern world in a rational, scientific manner. This isn’t something easily achieved, after all.
About me: I strive to not “blame” individuals in the traditional sense, because I reject free will as a meaningful concept. (Roll back the clock and a person will the behave the same in a deterministic universe. And if the universe has intrinsic randomness, we can’t ascribe free will to that randomness.)
So instead of blaming individuals, I focus on systems and their statistical effects.
I don’t think I made the assertion that anyone wants to throw away a social contract.
I said it was in the bin — it has been thrown away, and not by recent Trump voters, but it started long before that, maybe when ethics became something you didn’t learn until or unless you went through a professional program.
Having talked with a few religious Trump voters that I consider intelligent, they have a larger picture of the world, whereas I think a lot of Harris voters (being vaguely irreligious/atheist) are implicitly making the USA their whole world. Trump voters do not see the risk the same way, and have faith elsewhere. For many Harris voters this was an existential issue. That all said, those Trump voters I spoke with still seemed apologetic.
However it came about, there clearly is a misalignment in the stated social values world, and that is why that social contract doesn’t exist anymore. Whether that is because family forces, or tribal forces, or identity forces, or something else carried the day instead, well, maybe the future historians can sort out.
Two key points, which we probably agree on: (A) There are various flavors of social contracts. (B) Acceptance of any one social contract is a statistical thing, not a binary thing.
The most basic idea of a social contract is the legitimacy of a government to exist and function. This might mean to carry out various agreed-upon roles. And/or it might mean to operate according to some defined procedures, like the rule of law. The vast majority of people either consciously accept this or live as though they do. Of course there is variation on the proper role of government: does it include national defense, maintaining order, creating fair markets, collecting taxes, protecting civil liberties, internalizing economic externalities, investing in R&D, providing a healthcare floor, fighting corruption, etc?
If that were true, there'd be no point in ever applying a second time to YC.
Anecdotally, people do get accepted after 2-4 failures. Maybe YC was on the fence about those founders and their willingness to slam themselves against a wall repeatedly tipped the scales, or maybe they invest in a certain kind of founder and when they run out of those they invest in "promising" ideas.
Somewhere in my HN comment history from a while back is a response to a person claiming that modern phone cameras can’t detect IR illumination and remotes.
I took a bunch of modern iPhones and Android phones, from colleagues in an IT dept, and demonstrated they can in fact see a bunch of different IR remotes and illuminators with the rear camera.
I could find zero cameras that could not see the IR.
I’m not sure where people got the notion they couldn’t.
The ultimate answer is "it depends". And upon what it depends is the particular IR wavelength the camera emits to "illuminate" a scene in IR for night photography.
My cell phone's back camera will show IR light from IR remote controls (I've used it for just that to verify that a remote is transmitting). But I also have an outdoor IP camera with IR illumination in my back yard. The same cell phone camera sees zero IR emitted from the outdoor IP camera (even though it quite well lights up a fairly broad area of the yard at night).
So for my phone, if a 'spy cam' were using the IR wavelength the IP camera uses, I would never know it was present by using the phone camera. If it used something closer to the wavelength used by IR remotes, yes, then the 'spy cam' would light up via the phone camera.
"I’m not sure where people got the notion they couldn’t."
I make IR devices. My phone is the only one in the warehouse that can pick up their emissions - everyone else's cameras have IR filters with what appears to be a sharp ~750nm cutoff. I'm the only one that will pick up 800-1064nm with my cheap Samsung, and so I'm the only one doing the testing on those diode assemblies.
Ok, I'll admit that I have not tested that many phones, but all the phones I had in the last ~8 years would not detect my remotes unless I used the front camera. That could very well be because all the remotes I tested operate on 940nm instead of 850nm (the two common options), and the IR filter in my devices are so that it would cut the one but not the other. Or that they just have a much lower power level. Either way, most modern phones that double as cameras will have an IR cut filter of some sort, otherwise some photos appear weird - like the red glow from a fire will appear purplish-white.
I don't but it's easy to test, just pick up a TV remote and press a button while pointing it towards the camera. It should look like a flashing white LED.
I should also mention that both IR illuminators and TV remotes are usually either 850nm or 940nm, I have not looked into that aspect of it. I imagine that it's possible that your camera can detect one but not the other...
Easy test - check your black levels against someone wearing a black fabric or a reflective black surface. If you get a fair bit of grey in the image, odds are very great that you don't have an IR filter installed.
