person3

Having used Windows XP this bothers me because it's just slightly off.

> Non-technical teams are now shipping production code and many of our workflows are being automated

Given Coinbase is a financial platform this doesn't make me feel great. Hopefully they're contributing in areas that don't affect security or money.

This might be the most brain dead way to waste tokens yet.

I'm trying not to be negative, but would a human ever read any of the content? What value does it have?

It is challenging. In my CS degree grading for programming questions fell into two areas

1. Take home projects where we programmed solutions to big problems. 2. Tests where we had to write programs in the exam on paper during the test.

I think the take home projects are likely a lot harder to grade without AI being used. I'd be disappointed if schools have stopped doing the programming live during tests though. Being able to write a program in a time constrained environment is similar to interviewing, and requires knowledge of the language and being able to code algorithms. It also forces you to think through the program and detect if there will be bugs, without being able to actually run the program (great practice for debugging).

This scared me - I thought he died for a minute.

I can type like 45 WPM on my phone keyboard right now. I'm definitely faster on a full size keyboard, but I'm not sure if I would be faster on this small keyboard. I know a lot of people wanted keyboard like this when the iPhone first came out, but now with a lot of practice using mobile keyboards, I'm not sure it's needed.

One of the main arguments for hardware keyboards was you could type without looking. I don't really look at my phone keyboard when typing, I roughly know the spacing of the letters. Plus auto correct is really good at this point, so when I do make mistakes the phone usually just corrects them.

The only use case I could see for this is if the keyboard had control/alt/esc keys - in that case shelling into a machine on my phone might become slightly more efficient than an onscreen keyboard.

> 1. I want _nothing_ to do with user data. Nothing. Toxic nuclear waste. The idea of keeping the waste on hand needs strong justification.

Additionally, with regulations like CCPA in some jurisdictions, this isn't even optional anymore. At some point you will need to hard delete user data.

I don’t know a lot about how it works, so forgive me if this is a silly idea. I wonder if attestation could be done using real Apple devices, while leaving the private key on the user’s android. So similar to the old beeper to get the signed attestation, and send the result to the phone. Still could be secure since you can keep the private key used to encrypt messages local on the users device. I guess the issue might be a cat and mouse game if detecting beepers flock of Apple hardware to try and disable them all… (given many people would be using the same Apple devices)

This is new to me, so I might be wrong, but I don't get why they share revenue with the creators of these GPTs. They are basically just prompts that consist of a few sentences. There's no value add, and the more ChatGPT improves the less prompting will be required. These GPTs feel closer to bookmarks than an actual program.

This won't even work to solve the problem they're trying to solve. If I'm a scraper or someone that wants to drive fake ad impressions, what stops me from faking the attestation info? There's some mention in the original article about the attester validating the attestation data is signed on the client, but that just pushes the problem down the stack a bit. Someone could still spin up VMs, and just automate the scraping in a real environment that passes attestation. The author is claiming this will ensure only humans are viewing said data, but it doesn't really ensure that, it only adds a couple steps.

I also find it funny that the authors point to mobile platforms as an example of how this will work well. Last time I worked with ad tech, mobile ads were flooded with fake impressions, and I highly doubt that has changed. The funny thing about players like Google is that they want to be able to tell advertisers they're doing a lot to prevent fake impressions to get them to buy ads, but they don't really want to solve the problem because it would cost them a lot of money. So they kinda play the line and develop tech like this that sounds fancy but doesn't actually stop the problem in practice.

It's a bit odd though. What if you built your own ML model and trained it over a set of data that your wrote yourself. Would the work generated by the AI based of your prompts be not be copyrightable?

The original copyright laws were thought up way before even cameras, and we're still trying to apply them today to generated AI. Why can't we just realize that the world is very different now, and just create new laws? Instead we keep trying to arbitrarily interpret the law in a biased way to try to fit our modern goals as best we can.

I definitely agree, "open" in the name is pretty annoying and wrong. I'm a bit torn though - I tend to think GPT-3 is approaching a level where ML is becoming dangerous. Spam bots + GPT-3 is not my idea of a fun time. So the restrictions they put on usage do seem important.

Regardless of that moral question, I don't think a fully open model on the level of GPT-3 is even possible. Given the required cost to train and the expertise involved, big tech will always be a few years ahead. And it's unlikely they would give it away with how much they invest in creating it. Unless capitalism suddenly ends, I don't see any of the major tech companies parting with state of the art ML.