sibeshk96

Using my previous chess analogy, the world's smartest chess bot has played a million games to beat the average grandmaster, who has played less than 10,000 games in her lifetime. So while they both will have the same elo rating, which is a measure of how well they are at the narrow domain of chess, there is clearly something superior about the how the human grandmaster learns from just a few data points i.e. strong generalization vs the AI's weak generalization. Hence the task-specific elo rating does not give enough context to understand how well a model adapts to uncertainty. For instance - a Roomba would beat a human hands down if there was an elo rating for vacuuming floors.

> that is AI above and beyond what many humans can do, which is "awesome" no matter how you put it.

That's not the point being made. The point OP is making is that it is not possible to understand how impressive at "generalizing" to uncertainty a model is if you don't know how different the training set is from the test set. If they are extremely similar to each other, then the model generalizes weakly (this is also why the world's smartest chess bot needs to play a million games to beat the average grandmaster, who has played less than 10,000 games in her lifetime). Weak generalization vs strong generalization.

Perhaps all such published results should contain info about this "difference" so it becomes easier to judge the model's true learning capabilities.

Great idea! Since you're using Electron/React, did you guys build something in-house for the screen sharing + cursor collaboration? in my experience, nodeJS is really bad at this kind of real-time stuff

Still cheaper than actual caviar

This reminded me of the cosmological fine-tuning argument : that there are few precisely fine-tuned constants which seeded the universe in it's inception, which lead to complexity (life, planets, galaxies) emerging in the universe.

Google has invested in KaiOS too

Kudos for the concept. Intriguing form of counter-surveillance that could actually work in a lot of existing surveillance states. Had a few questions :

1. Who in the system is responsible for maintaining and updating the object detection model? It seems like a centralized point of failure for an otherwise decentralized system. Might want to check out existing techniques for Federated Learning for ways to counter this. (https://www.openmined.org/)

2. How possible is it to mount a Sybil attack on this system? Why did you select PoW considering it's weaknesses? One possible problem could be that early adopters will necessarily have to be co-located for the system to have any value(Hence consolidating value in a certain geographical area, making it very easy to regulate/shut-down).

One of the things I don't get about this concept is how they plan to avoid the surface bottleneck - if everyone's trying to get into the same number of limited holes in the road, then there's more and more congestion and traffic wanting that access. To make a 3d network of roads work, you'd also need a 3d network of buildings, other wise it's not really solving any problem.

Hi HN, co-founder here. We're a team of 3, right out of college, building this for the past 5 months.

Boxware is a web-app that lets you discover, use & collaborate on desktop software in your browser without downloading or installing anything. We do all the heavy lifting so you can access them on-demand - streamed right to your browser. Think Netflix for Software.

In full-screen mode and at 4G speeds, its indistinguishable from using a normal computer.

Fun meta-fact: we built most of Boxware inside Boxware itself. Here's a demo if you're lazy - https://www.youtube.com/watch?v=yCTPDSW2NEk