Great question. Unfortunately deepfakes have gotten even better lately, to the point where people are using them to do things like fake celebrity endorsements. I think the cat's out of the bag and short of banning everyone from owning GPUs, there's not much we can do to stop things like deepfakes. Interestingly enough, requiring licenses for powerful GPUs was one of the proposals put forward by OpenAI, I can't find the document right now, so if someone has a link that would be appreciated.

I share the same opinion of Emad Mostaque, founder of Stability AI, who says that the best course of action is to make sure everyone has access to the tools, and educating people on how to use them and where the dangers are. I know it's not a magical solution, but I think the risks are even bigger if we don't build AI out in the open.

No one can. It will be used for evil

We cant, but we can normalize public key cryptography and make it commonplace for people to sign their content.

Salut!

Yes, I use Github Copilot on a daily basis and I couldn't live without it. Every developer I've spoken to in real life shares the same opinion (it's so good that people are paying for it out of pocket, not even waiting for their companies to buy it for them). Haven't seen this much adoption of a product by programmers since the early days of Slack.

Jasper is pretty good for copyrighting, it's built on GPT-3 but has a lot of extra functionality. I know some product and marketing people who are using ChatGPT for things that they view "low-importance" commercial boilerplate, pages like "Our mission", etc.

The most important tip when using ChatGPT (or Bing chat) is to specify the agent that you want to instantiate. Think of these chat bots as Simulators - they can be simulate both a good copyrighter and a bad one with similar difficulty, so it's important to specify that you want a good one :) I'm oversimplifying a bit, but remember that the chatbot doesn't know what world it's in and you need to help it help you, if that makes sense.

Great point. It does seem like many people are either quick to dismiss AI tools, or call for banning them (which are sort of opposite views), especially in the art community. And I can see why we might feel threatened, art has been a uniquely human activity and until recently it didn't seem like machines would be able to do it.

Recently, there has been significant progress in the Stable Diffusion community using ControlNet, which allows you to shape the subject however you want. Here's a nice demo of what I'm talking about.

Another very useful tool that increases productivity when generating art is outpainting, which allows you to generate plausible continuations of images. Works reasonably well and of course you decide what to generate.

If you're willing to put in some effort, there are guides on how to use ChatGPT as your home assistant. Not sure I would trust it though. If you can wait a few months, the existing home assistants will get much much better, as none of them are currently using anything close to the state of the art AI tech (especially Siri).

I was talking to someone the other day about this. There are some tasks that AIs can do better than the average human and one of them is audio transcription. OpenAI Whisper is incredibly good at transcribing audio in multiple languages, even when the sound is pretty bad. It's really, really impressive. So it can already be used as a sort of "gig worker", where you give it the audio and it outputs the text (there are already a myriad of self-hosted solutions so you don't need to run the model yourself).

But if you're referring to the job in terms of everything it includes, things like working with your team, interacting with your boss, doing various real world tasks, we're not quite there yet. This might change in a year or two though. Also, some (perhaps an increasing amount of) jobs are reducible to gigs, so in those cases it's more clear cut.

My favorite one has been Rewind (no affiliation to them). Unfortunately it's Mac only. It's basically a rewind button + search bar for everything you see on your screen and everything that gets picked up by your microphone. Everything happens locally and it's pretty fast, but I had to disable it because my poor Macbook Air couldn't handle it. I'm sure it will improve in time.

So in my case, if I'm trying to reproduce a bug and I forgot what I did, I just swipe to rewind and play it back. Also being able to search every text that ever appeared on your screen is pretty insane. If you had told me 2 years ago that this would be possible today I wouldn't have believed you. That's why when someone tells me how AI can't do this or that, my answer is to check back in 2 weeks.

There are several tools out there, the most well known one being Originality (full list here). OpenAI is also working on tools to fingerprint content generated through their tools. Ultimately, I don't think this is a winnable battle (students can do multiple rounds of rewriting with different tools until the fingerprint is gone). But hopefully I'm wrong. If not, the best way to handle it is to update our education system to incorporate the new tools, just like we did when calculators came along. I remember my math teacher saying "You won't always have a calculator in your pocket!". That didn't age well.

None. You cannot trust AI, by definition. AI does not understand the concept of trust.

Great question! I see that your username has Spanish in it, so if you know a foreign language my best tip for you is to choose an AI that you like from There's An AI For That and build the equivalent in that language. There is a total lack of non-English AI tools and most of the demand is international.

Next best tip is to pick a super simple language generation use case (like "excuses generator") and fine-tune GPT-3 on 100 good examples on it. Most people are not bothering to fine-tune but it makes all the difference. Remember the Linkedin viral post generator? Something like that. You can use a free SEO tool like Ubersuggest to find niche ideas, and this way you'll keep getting SEO traffic once the initial buzz is over (I don't think search engines and SEO will die anytime soon).