OpenAI in throes of executive exodus as three walk at once Three key OpenAI staff members – CTO Mira Murati, Chief Research Officer Bob McGrew, and Research VP Barret Zoph – are leaving the ChatGPT maker. Murati described the decision as "difficult," but wrote that "this moment feels right." She said she is leaving "because I want to create the time and space to do my own exploration …
Never has a quote been so aptly placed as this:
AI expert Dr Gary Marcus described the events at OpenAI as a "slow-motion train wreck."
I'd disagree that the motion is quite that slow, however. That said, given the nature of the bubbles, if OpenAI pops the "AI" industry is going to start looking shady pretty quickly, and anyone who's invested very heavily in it is going to be entering what's generally known as the "find out" phase.
When we asked OpenAI for comment, it had nothing to add about the departures specifically, with a spokesperson saying: "We remain focused on building AI that benefits everyone and as we've previously shared we're working with our board to ensure that we're best positioned to succeed in our mission.
Alternatively, it is just as easy to believe and realise that OpenAI is dedicated to remaining determinedly beneficial to all and not just to those few with deep pockets thinking to be able to yank novel virtual strings to get their future way because that is the plain simple way things are done in order to succeed and prosper ..... as the past will attest to and confirm ....... would be a good old-fashioned mission statement for a revolutionary pioneering enterprise which has the status quo terrified of the consequences and repercussions of their increasingly expanding and rapid rabid successes.
Vive la différence et la révolution‽ .
Although that could also be the reason why old trusted deckhands are jumping ship too, of course. Strange times indeed making for interesting news for situation publishing.
... AI is not going to go away entirely, but it's time for it to start hibernating again for most people. The hype is about over. Winter is upon us again.
Methinks the investors are starting to notice that the marketards have been crying WOLF! incessantly for a couple years, with nothing to show for it. Personally, I think the process started a couple years ago, but what with everyone sitting at home playing with themselves (because Covid), nobody noticed.
And now the rats are beginning to jump ship. Gittin' out while the gittin's good.
Perhaps that fancy processor you bought will help keep you warm this Winter.
People seem to forget that all of those expenses are for training AI models. Right now all companies are still trying to figure out the limits of this tech. It's expensive, yes, but the potential may be still worth it. These billions may seem like scary numbers for normal people but it's not that special considering multi-billion purchases IT giants do every once and then.
One you figure out by trial and error and a lot of research how to build a good AI, I am sure that maintaining its knowledge will be much cheaper and the first "land grabbing" companies to get there will earn it back and more.
We know what the limits are, they've already reached them! The emperor has no clothes, this is just a massive fraud in the same way that the metaverse and crypto were. They're dressing up existing technologies and calling it AI and claiming it's something new. There's nowhere for it to go, that's why the rats are fleeing the ship.
AWS had first-mover advantage in cloud hosting - and they're still both the biggest and the brand name in that business. OpenAI had first mover advantage in the chatbot business, and ChatGPT is very much the brand name in that business - even though its competitors are more or less as good (IMO Gemini Advanced is better). To be at the top of this business, very expensive systems are required: if you don't invest, you won't stay at the top for long. It looks as though Microsoft used to have a good reason to invest, even if OpenAI didn't make money - as a way to add intelligence to their own products. However, it looks as though Microsoft's long term plan is to move away from this strategy, and to build their own AI instead. So eventually OpenAI will run out of other businesses who will provide enough money for them to stay at the top of the game they're in - and given how much their systems cost to even just run, they would soon find themselves unable to pay their bills. So they need to make money, and they need to be "for profit": they have 2 ways to do this: (1) charge for access to more intelligent and capable models (2) charge other business to easily get intelligence by enabling them to do prompt engineering of OpenAI's models via an API. This way, they can earn good money for a few more years, until low cost chatbots become "good enough", and OpenAI can no longer charge enough people enough money to sustain the level of business expense they are currently incurring. The very long future: the brand name ChatGPT will live on for decades - but eventually it will just be a name on a cheap chatbot that OpenAI didn't build, and which is also available, at lower cost, under non-brand names.
My brain has mentally pigeonholed Sam Bankman-Fried and Sam Altman in the same slot so I keep getting them confused, which is ridiculous because one of them got incredibly rich by inflating a shady hype bubble that was self-evidently going to pop catastrophically for everyone who put money into it and the other is Sam Bankman-Fried.
If the generally dismissive tone and surprisingly earnest disbelief in AI’s ability to fundamentally and radically change current presentations for the future experience of a virtualised, SMARTR AIMachine-led existence for humans on Earth, as expressed so far by the clear majority of El Reg commentards posting on this thread, is truly indicative of the greater global whole, well ........ are y’all gonna be absolutely blown away by what the future has in store for you to worry yourselves about, and fight each other, possibly and therefore probably, to death over.
* Global Operating Devices
Hey, amfM ... Sit down, have a beer, and I'll try to keep it short ...
They've been saying the same thing since I was at SAIL in the '70s (for varying values of "they", usually marketing twats and SciFi authors).
And we'll all have flying cars, jetpacks, controlled fusion, and be living on the Moon and Mars any day now, too.
In all reality, useful AI is probably 20 years away. As it was 20 years ago. And as it was 30-odd years before that, when I was at SAIL ... I'm willing to bet that 20 years from now it'll still be 20 years away.
But first, we'll have to convince the GreatUnwashed that the current poster kids, the hallucinating LLMs, are not only not useful and not trustworthy, they are also not AI, and have been found to be dead-end, a complete waste of time and money now that the basic research has been done and found lacking.
You are absolutely right. In the past, I have actually learned things from the El Reg comments - but the readership's attitude to LLMs is truly astonishing. Maybe I am blessed to have seen part 1 of this film: In the 1980s, chess computers were regarded as toys by strong chess players (one or two did use them to help with analysis). In 1996, when Kasparov made Deep Blue look really poor, people assumed that computers would never overcome the horizon effect in game trees. A common response to the prospect of computers beating the world champion at tournament times was, "Not in my lifetime.". A year later, Deeper Blue won the rematch. Shortly afterwards, the prevailing doctrine became that the best chess was accomplished by humans and computers working together ("3-Hirn computer chess"). This is the current state of chatbots. In chess, though, this state was short lived. The current state is that free chess programs running on cheap hardware play chess far beyond the level of humans - with or without computer assistance. So my message to all the chatbot haters is this: I've seen this film before, and I know how it ends.
Except chess is a small game. 32 pieces, 64 squares. (Yes, I know the work of Claude Shannon, and Martin Gardner's "Mathematical Games" columns in Scientific American were published roughly during my formative years).
Even a very simple AI will be vastly more complex than any board game. IF, in fact, such a beast is even possible. Which I doubt. The artificial bit is easy ... the intelligence, not so much.
I thought all the cool kids were using Go as their example game these days?
I hear you buddy, and I don't understand the comments either.
LLMs are brainless - yes. BUT they allow you to create work that in your view is excellent in a very short time. Almost instantly.
Want to turn your brain farts into a great letter or document - they're perfect. Want it to write code (and tests) that you can scan through and ensure is correct - very near perfect, and in a tiny fraction of the time you can do it yourself.
So what if they have flaws... they're a frigging awesome tool for us to use. And like chess computers they're only going to get better.
this story over one Arse Technical was that Mira Murati couldn't face even the vaguest possibility of spending the better (worse) part of a decade sharing a facility with Elizabeth Holmes. For heaven's sake the prospect of tens years in the forced company of that decidely brain dead bunny boiler....
Theranos' greatest crime wasn't that it defrauded investors but rather that it deprived very powerful and, in their minds, important people of a minuscule part their certainly equally ill-gotten fortunes.
While OpenAI almost certainly hasn't broken any US laws that apply to normal people (so far), if the whole operation falls well short of the over spruiked benefits, as it very likely will, then the special offense of depriving the powerful and important will necessarily require fairly long terms of imprisionment for someone. Mira and the other refugees have decided they don't want to be that someone.
....if the whole operation [OpenAI and the Inexorable Unbelievably Stealthy Advance of Almighty SMARTR AIMachines/Global Operating Devices] falls well short of the over spruiked benefits, as it very likely will,... ..... Bebu
To imagine both the very near and distant future to be little changed from the present and the past as the above quoted snippet of text surely suggests, Bebu, is something which the amazingly rapid progress of Learned Large Language Learning Machines and invested establishment vested interest terror and reluctance to submit and surrender and cede them leadership command and control would certainly disagree with and thus is everything being so feared and dismissed more likely than not to be rendered both practical and physicalised, virtual and metadataphysical.
And you're either for the AI Change or against it and IT and a Hostile Enemy Combatant of the Future with all of the very likely directly personal ramifications worthy of support for those choices.
I'm with you on this one, Bebu is most likely a Hostile Enemy Combatant of the Grammatical Future Past Participle and Associated Conjugations in my book! ... and while we're in good terms, may I suggest: Learned Large Language Learning Machines Learned Large Language Learning Modern Machines -- or LLLLMM?
Rolls right off the thong ...
Whilst I might very well agree, AC, that 4L2M [LLLLMM/Learned Large Language Learning Modern Machines] is an even more accurate $L@M dunk, much greater cover and stealthier flexibility for delivery is provided and safeguarded by operands ensuring everything is not veering too far towards, and into specificities of AI, rather than the product remaining and exploring and displaying its generous general ubiquity and common equitable purpose in the multi-use, multi-disguise cloak that adorns and adores the simply complex moniker universally recognised just as AI.
AI is able to be like just a weapon ....... with the range in its arsenal fielding munitions mimicking the similarities to be found in the histories of conflict and wars employing developments ranging from a sharpened flint arrowhead and stout pointed stick right through a vast catalogue of goodies which currently might offer deadly nuclear explosive intercontinental ballistic missiles and anonymous catastrophically destructive exploits for Oday attack vulnerabilities known only to a certainly strange few in the thrall of ........ well, they would be those Rumsfeldian unknown unknowns — the ones we don't know we don't know. And if one looks throughout the history of our country and other free countries, it is the latter category that tends to be the difficult ones* ..... of Cyber Space Forces of Immaculate Source and Infinitely Resourceful Assets.
* ...... https://en.wikipedia.org/wiki/There_are_unknown_unknowns
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
