Goodbye, Twitter. Hello X, xAI
Twitter HQ on the evening of 23rd July 2023 - Source: Twitter.com
After 17 years of existence, as of 23rd July 2023, Twitter is officially gone and has been rebranded as X. The transition into 'The Everything App' is well underway as planned and Tweets as they've been known form just one pin in the soon-to-be-filled cap of the new umbrella company. The idea worked well, but needs revision and much has changed since the takeover; their length limit has been increased, videos have been added along with subscriptions to your favourite creators and a share in advertising revenue. Not to mention the significant decline in bots and negative content, it’s actually never been better!
Spaces
In fact more features have been added in the last six months than the past six years under the previous leadership, a prime example being Spaces. These are hugely popular voice chat rooms that can be immediately created to host meetings and discussions of any sort, between any users. They've already been used in official capacities for big announcements such as Ford and GM CEOs sharing their news of adopting the Tesla charging standard and having supercharger access for their vehicles. Even Ron DeSantis hosted a Q&A and announced his US Presidential candidacy using Spaces, becoming one of the largest online chatrooms almost immediately, reaching well over a million participants in just a few minutes.
Proof that these kind of important, long-form debates are completely viable in this medium and the new X platform should really be seen as the world's digital town square, where any discussion be it political, ethical, or especially entertainment can thrive. A place where you are encouraged to be your true, authentic self. This is just the beginning, we can definitely expect payment services to appear, possibly micro-transactions to allow traditional media to transition to a viable business model now that advertising is becoming irrelevant (if viewed by bots/ChatGPT instead of you). Who knows what else is on the horizon and in the mind of Elon Musk!
xAI
Source: x.AI
Coincidentally and as if not busy enough, he has also launched yet another new company - xAI, focusing on developing a truth-seeking artificial general intelligence and dedicated to understanding the true nature of the universe.
Announced on the 12th July 2023 in honour of Musk's hero, Douglas Adams, since 12+7+23=42. Fitting since the machine will be pondering life's greatest questions in an effort to decrypt the answers that humans cannot. As in The Hitchhiker's Guide To The Galaxy, the question itself is clearly the most important decision to make and this is exactly what the team, comprised of a seemingly random collection of exception individuals, aim to determine:
Elon Musk
Igor Babuschkin
Manuel Kroiss
Yuhuai (Tony) Wu
Christian Szegedy
Jimmy Ba
Toby Pohlen
Ross Nordeen
Kyle Kosic
Greg Yang
Guodong Zhang
Zihang Dai
Not all of whom have a background in AI, but each have impressive histories and successes. Hear directly from the brilliant minds themselves and get to know their interests and motivations by catching the recording of the announcement here: Twitter Spaces recording.
Elon admitted that he initially approached these people to join Tesla, however, they were not interested in joining a large corporation, instead preferring to work in a small start-up environment - hence the need for the company. It's certainly not the case that Elon wanted more on his plate, but it's not a bad thing to have an official entity working on these problems that's not beholden to Tesla's shareholders, yet still technically has access to all of the same resources.
It's clear that anyone working under Musk's umbrella is afforded the opportunity to traverse multiple fields of work and research, perfect to keep the most brilliant minds engaged and constantly challenged whilst also providing massive benefits for all involved, being able to share discoveries, guidance and ideas between such diverse teams and industries. Material science doesn't care if it's being employed in a passenger land vehicle or a space-bound starship, it's a huge benefit for the innovation in all of their ventures to be able to tap into such vast pools of knowledge in every area, for any problem!
So far we only know the individuals involved and their mission, but we can expect further updates over the coming weeks and months to share more details of their work. To keep up to date, you may wish to follow them on X (formerly Twitter), where they recently requested input for the best ideas to begin working on.
AI Competency
With predictive language models like OpenAI's ChatGPT, Google’s Bard and others causing such a storm, AI has been the buzzword of 2023, now everyone is scrambling to put it into their products or develop the next big breakthrough. It's clear that the initial shock was more hype than substance, whilst these tools might be miracles to people from the 20th century, at this point their limitations have been fairly well tested and their 'intelligence' clearly proven to be a sham.
What seemed originally unbelievable without concept of how it works is now clear after time to study and digest; what we've built so far is machine learning to operate (ironically almost too human-like) in the direction of least possible resistance, to say whatever it believes we want to hear with the least amount of work. Quite the opposite of what is going to free us from ignorance or redundancy, at best we've created our own yes-man companion with no interest in understanding, only pleasing.
What is referred to as 'AI hallucination', in practice blatantly fabricating lies to support the answer, is simply unrealistic expectations from a transformer-based Large Language Model. The AI is never looking to tell you the facts, it's trying to tell you the answer you'd be most happy with. In this way it s oblivious to the meaning behind anything, if trained on the right material it can definitely provide correct answers as they were the most likely from the source, but this allows it to fail spectacularly and is especially noticeable in it's ignorance of maths. For example, it knows that when prompted with such a question that a number is the most probable response, but has no notion of there being a right one, or how to come up with it. Without that it confidently shares anything it's seen before and happily agrees with you should you point out a mistake, but has no concept of what that mistake is.
The correct method?
That kind of training model certainly has a place after refinement through future iterations, but it may never be capable of discovering anything that we haven’t already fed into it. What we really need is a mind much more in line with our scientific method, one that desires to constantly question, to always be skeptical and double-check it's work. To logically process a problem step by step, breaking it down into layers of tasks that must all be completed and accepted before an answer is provided, to ensure that it has the highest possible probability of being true, not that we will agree or necessarily even want to hear it.
This is the kind of tool being created by Wolfram with their GPT4 plugin, there’s an element of realtime computation under the hood, but even that is more of a data analyst than theoretical physicist. More plugins are sure to enhance and adorn additional hats onto the friendly chatbot, expanding it’s practical applications massively, but this is a brute-force, bolt-on attitude compared to simply growing it correctly from the start.
xAI’s mission certainly seems like a most noble of endeavours and one of the best applications imaginable for an entity which can think infinitely faster than all humans collectively. The dangers are far less ominous with this kind of approach, with no expectation of the machine to take any action, purely speculative discovery seems like the safest option. Of all the known AI systems currently in development, if correctly nurtured, this could be the least likely to cause any direct harm to humanity.
It remains to be seen if Tesla will afford the use of their Dojo super compute cluster towards any of this training, or whether the xAI team will need to build their own, either way the greatest challenge will simply be procuring enough hardware. Demand for Nvidia GPUs reached insane levels after the Bitcoin mining rush, now AI training cores are the most valuable chip commodity. AMD are slowly dipping their toes into this water, but the supply is limiting progress globally. Tesla of course having the ability to design and build their own custom silicon is a huge advantage, but consumes a significant amount of additional time and capital so it will likely be a similar story as with their battery cells; they take as much as they can get from outside suppliers whilst also running full speed production internally.
Overall this is great news for humanity, the need for this kind of AI is never greater, something that will always be impartial and that everyone can agree on - whether or not they choose to is a different question…
The discoveries that can only be dreamed of might be right around the corner, unlocking not just wild sci-fi technology by truly understanding the physical world and laws within it, but also healthcare and wellness to be able to live our best, most fulfilling lives.
This announcement will be one that is looked back on in years to come as a defining moment in our evolution. I have a firm belief that if any AI is going to change the world for the better, this will be the one, doing so in such spectacular style that no-one can predict. It will permanently alter humanity and it's only a matter of time.