povik
Joined 10 years ago
Karma
119
Submissions
43
Member since
May 19
Recent submissions
Comment · 20 days ago
Comment · 20 days ago
The article singles out paying users. So there is a way to pay for OAI/Anthropic models without having the model trainers look at all you do, it's just not well known.
Comment · 20 days ago
> ChatGPT and Claude will never be private
> These companies only offer closed versions of their powerful models (despite the word "open" in "OpenAI"). You pay them for chat and/or API access, and they are allowed to look at everything you do.
How should I reconcile this claim with 'Google Vertex' and 'Amazon Bedrock' being offered as providers with zero data retention on https://openrouter.ai/anthropic/claude-opus-5.5#providers?
Comment · 3 months ago
If a model demonstrates being good by winning on prediction markets I may want to ask the same model questions which are too niche to have markets. That said I don't know if niche questions are enough of a revenue stream to be interesting for model providers.
Perhaps they may enforce a knowledge cut-off for information retrieval and price the service based on how recent the cut-off is, and also use the cut-off as a way to guard their advantage on the markets.
Comment · 3 years ago
Hi, I wrote the Linux kernel drivers for speaker output on the new ARM64 Macs, so I can shed some light on this.
> I wonder how they model the energy output of the speakers.
Simple. The speaker codecs (special made for Apple, but e.g. compatible to https://www.ti.com/product/TAS2764) give you current/voltage measurements you can calculate the power from. It's what they call I/V sense.
> In principle you could have a function that calculates the energy of the spectrum and integrate that over time.
Right. You don't want to do the calculation in frequency domain since that would be wasteful. You can integrate the power of the samples straight away (here by multiplying the instantaneous voltage and current, if we didn't have that, we could at least estimate it by assuming a constant resistance of the speaker coils).
> If you put out too much energy, what do you do? Reduce the overall volume, apply a frequency cutoff, or a low-pass-filter?
You model the coil's temperature. If it comes near a dangerous level, you reduce the volume on the particular speaker. You don't need to worry too much about this since usually you don't hit the limits, you are merely guarding against especially nasty sound input.
> I feel there are so many variables here, and it is going to be hard to replicate what Apple does.
We don't need to replicate what Apple does. Though we are stealing the parameters for the coil temperature model Apple is using (at least for some machines).
Comment · 3 years ago
> When the next Apple chip comes out, it's back to square one. It's entirely Apple's fault.
That's simply not true.
Source: wrote a couple of drivers for M1 (now upstreamed in mainline Linux) that work without a single change on M1 Pro and M2
Comment · 4 years ago
Yes, exceptions should be written into the law in the first place. I assume the government is very comfortable with getting to draw the line themselves.
Comment · 4 years ago
I happen to come from that part of the world where something radically different has been tried out, and will also tell you capitalism is good (considering the alternatives).
Comment · 5 years ago
Technical nit: if average is 48 hours, median cannot possibly be higher than 96 hours.
Comment · 5 years ago
How are you so sure? Is the environmental impact of the pet rocket program of his really so significant?
Comment · 5 years ago
Also at top of the list: “We care about your privacy” (and not that we are required by law to give you these options that follow)
Comment · 5 years ago
> It seems to me this is simply a statement that an electron is a fermion not a boson, right?
Not really, that's a property of any quantum particle, bosons also. They don't occupy a single point in space like you would imagine classical particles do. They also cannot occupy a single point in space relative to another particle.
> It's a fermion, but why should it be?
Well there's the spin-statistics theorem in quantum field theory which starts from some likely assumptions and then shows that particles with non-integral spin must be fermions while particles with integral spin must be bosons. Other than that I take it simply as an experimental fact. I don't think there's a nice reason for it one could give (today, that is).
Comment · 5 years ago
> I don't really understand the physics of why electrons don't collapse into the positive nucleus, since positives and negatives should attract
My (student of physics) answer to that would be that they are pretty much collapsed as much as they can. It’s just that under quantum mechanics that least energetic state is not the one in which the electron is perfectly co-localized with the positive charge. (As that perfect co-localization is not even physically attainable.)
Comment · 5 years ago
Although you can use a lot of latex commands to enter math, which is useful when you are starting with texmacs after exposure to latex
Comment · 5 years ago
Why would the commentators have a stake in the matter?
EDIT: strangely Azure shows up as ZDR on openai's models but not anthropic's: https://openrouter.ai/openai/gpt-6.1-sol#providers