The Future and Us

Welcome to edition #3 of the ‘No Longer a Nincompoop with Nofil’ newsletter. I’ll be candid, if you only read a portion of this newsletter scroll down and read the last section. Trust me, you won’t regret it. There is a lot I haven’t included in this issue so I’ll be sending out an additional one this week so keep an eye out for that 👀. I don’t want to waste even a single line so I’ll get right to business.

Here’s the tea ☕

• Your voice is no longer yours to use 🤐

• Meta’s VR video call tech is kind of insane 🤖

• Elon Musk helped create OpenAI. Now he’s fighting it ⚔️

• The problem with AI no one is talking about 🚨

Being “you” is a lot easier now - for everyone

ElevenLabs is one of the leading companies on AI speech software. They boast the most rich, versatile and compelling lifelike voices, and you know what? They’re absolutely right. Journalist Joseph Cox illustrates just how good their software really is by hacking into his own bank account. Banks across the US and Europe laud voice ID as a secure method to access your account but with new AI speech softwares it’s practically impossible to differentiate between a human and AI. Soon enough it’s going to be really hard to discern what someone said is real or not and as AI gets better and more accessible, creating misinformation will be as easy as a few button clicks.

Take for example the video below. A conversation between Joe Rogan and Andrew Huberman, two of the biggest podcasters in the world. The problem? The conversation never happened. This is an entirely made up conversation designed to promote a product. If you’ve ever heard either of them talk and don’t pay too much attention to their voices you wouldn’t be able tell its fake. This is just the beginning of a wave of deepfake ads and misinformation created online. Taylor Swift apparently said she doesn’t “perform to poor b*****”. In just two days over 1000 videos using the audio were created and garnered over 4 million views on TikTok alone. A tidal wave of misinformation is coming and we have no idea how to combat it. Good luck fact checkers 🫡.

I might owe Zuck an apology

Last week I made fun of Meta implying no one really remembers them anymore. Well turns out they haven’t been doing nothing and have actually made significant progress with their VR tech. The initial two avatars look so real it’s actually remarkable. As famous and recognisable as Zuckerberg is though, I’m not sure having him illustrate various emotions was a good idea considering he rarely shows any human emotion, like a human anyway. Sorry Zuck, why don’t you leave the human stuff to us humans.

Does this mean the Metaverse is saved? I don’t think so, but it’s definitely a big leap from this selfie Zuck took just last year.

Elon doesn’t want his AI “woke”

In fact he hates “woke” AI so much he’s building his own one! Musk helped found OpenAI (creators of ChatGPT) back in 2015 with current CEO Sam Altman and a few others. Back then OpenAI was a non-profit company and since his departure in 2019 transitioned to a “capped” for-profit company. Capped meaning the profits are capped at only 100x any investment made. Chump change if you ask me.

Musk isn’t mucking about this like he is with Twitter either. Contrary to popular belief that his Twitter acquisition is going great, Twitter has lost over 500 of their top advertisers. This poses a problem considering Twitter makes over 90% of its money through ads. Their revenue is also down over 40% from this time last year, but who gives a shit when you can boost your account and have everyone on the platform see your tweets first because your Elon Musk and you own the platform.

With his OpenAI competitor, however, Musk has already made some big moves by hiring previous DeepMind (Google’s AI research lab) and OpenAI employees, two of them in fact. In the grand scheme of things, competition in the AI space is good for the general public. The more companies there are competing to build these models, the faster we’ll get to Artificial General Intelligence (AGI - AI that “thinks” like a human). But this poses a concerning problem. Lets talk about it.

The Problem with AI no one is talking about

Building an AI company isn’t particularly hard, you call a few API’s and suddenly you’re innovating an entire industry. But who’s building these API’s? Who’s driving the creation of these new models? Who’s training them on datasets large enough to fit the entire internets worth of text data? A select few companies (Microsoft, Google, Meta and a few others). It makes sense considering how much it costs to train a model like ChatGPT - anywhere between $2 - $12 Million. Don’t forget maintenance costs and the cost of keeping the models running if they’re made publicly available. It’s estimated ChatGPT costs at least $100k to run every single day. Being involved in the creation of this technology isn’t accessible to the vast majority of people or companies, either for financial reasons or a simple lack of knowledge and resources.

And here lies the problem. If you understand how models are trained you might have picked up on it. These models are trained using a method called RLHF (Reinforced Leaning from Human Feedback). Without going into detail, models like ChatGPT are trained by being given an input and generating a number of responses. These responses are then ranked by how “human-like” they are and this creates a feedback loop allowing the model to understand how to best sound human. But who’s ranking these responses? Who determines what sounds most human, what is most correct? Humans! We are. Not you and me obviously, but researchers and engineers working at these companies. Finally, we arrive at the problem.

It is no exaggeration to say that AI will drive the future of humanity. We’re all just here for the ride.

Thanks for reading ❤️

Written by a human named Nofil