adinisom

There's room for all sorts. While I've met engineers happy just building the widget with all details specified up front, my career has been taking incomplete specs and deciding how the product works to create holistic designs.

Domains cost money to maintain. They can lapse, change, or be sold. Sometimes people make mistakes. So to my way of thinking they're great as a short-term named identifier but not as a long-term stable identifier. Would be nice to have both.

Soft disengage -- if the GPU fails it continues to drive via on-device model, alerts the user, and then disengages.

More generally, Comma's safety is handled by a microcontroller. The microcontroller limits steering torque commanded to a safe range for level 2 driver assistance and acts as a watchdog for the rest of the stack. Failure of openpilot causes it to disengage and sound an audible alarm.

I'd like to see DNS ownership combined with the time range and an owner-unique key so you can go to the website you intended to visit rather than whoever happens to own the domain at the moment.

Will be interesting to see how Google implements a more open DSP wake-word detection. The requirement for it to work without the app holding a default role suggests needing to recognize multiple wake-words for each of the non-default apps that uses one.

From an openness perspective this is excellent. Technically it seems challenging with a DSP designed to detect a single thing using as little power as possible. Currently this balances doing as little work as possible to detect plausible utterances of the wake-word on the DSP while minimizing the costs of spurious wake-ups on the CPU. At the very least multiple wake-words seems to need the DSP to do more work and wake up the CPU more often.

Thanks, beat it following your plan. Very cute that at times the game punishes the player for being too productive with tractors.

So far have not made it to phase 3.

Despite improving production (more tools! more assistants!) and bonus tractors (work harder! heroic labor brigade!) it's not enough to make up for phase 2's punishing 1000 tractors in 60 days, or roughly 16 tractors a day. Even if we give ourselves a full 120 days, that's 8 tractors a day and we seemingly cannot accumulate tools and assistants early enough to do that.

Great seeing where Speech with Sam and, more generally, you ended up.

In Stratton Oakmont v. Prodigy, Prodigy ran a message board and would have escaped liability for defamation by a third-party save one problem: they exercised editorial control by moderating content.

Hacker News faces the same risk.

Respectfully I prefer a world where Hacker News exists.

Appreciate you sharing that. Agree with the overall concern with children using tech and I'd start with banning smartphones in school classrooms. Two observations:

Why stop with image boards? After eliminating all photo uploads, awful people still produced and shared illegal content. So then we eliminated unlicensed camera use. 'Course the real problem still happened.

Hacker News is made possible by Section 230... should it not exist?

There's some logic to wanting to assign responsibility for other's actions upon those who enable it. Like you say, to incentivize intermediates to police their users and those they do business with.

But the problem with deputizing intermediates is that it's too effective. It creates incentives to over-police and we have less rights against corporate policing than we do government policing. We would not have the internet we have today without the user generated content and moderation that section 230 enabled.

If you want to pull another cable:

- secure a string to the old cable

- pull the cable out the other end, pulling the string through

- secure the string to the new cable

- pull the string out the other end, pulling the new cable into position

The picture in the article shows what looks like keypoint matching (ie, SIFT, SURF, FAST) between the query picture and the database. This can give an exact location if a picture of the location exists in their database.

They contrasted this with their prior technique which is more of an image classifier that can identify general location from image features. This approach does not require their database to contain a picture of the exact location.

A while back I looked at matching pursuit. At first it seemed very complicated, but after staring at it a bit realized it's simple.

- Start with a list of basis functions and your signal.

- Go through the list and find the basis function that best correlates with the signal. This gives you a basis function and a coefficient.

- Subtract out the basis function (scaled by the coefficient) from your signal, and then repeat with this new residual signal.

The Fourier transform is similar using sine wave basis functions.

The key that makes this work in situations where the Nyquist theorem says we don't have a high enough sampling rate is ensuring our sampling (possibly random) is un-correlated with the basis functions and our basis functions are good approximations for the signal. That lowers the likelihood that our basis functions correlating well with our samples is by chance and raises likelihood it correlates well with the actual signal.