Discover more from Value Investing Substack
🇺🇸 Satya Nadella for President 2024 (OpenAI)
The Game Theory behind last week’s drama at OpenAI
In this article, we explore the Game Theory behind last week’s drama at OpenAI. I’m not sure how many people realize how incredible this outcome actually is — there were so many ways things could have unravelled way, way worse. If OpenAI were a video game, this outcome could perhaps be described as the Perfect Good ending.
Over the past several days, most of you will have caught up with the boardroom tussle at OpenAI. The short version is that there was a management disagreement over the future of OpenAI (in complete good faith territory, I might add) — and boiling tempers got the better of good sense when Chief Scientist Ilya Sutskever orchestrated a hostile takeover by kicking Sam Altman out as CEO. Sam has notably been the face of OpenAI thus far, and his widespread publicity and subsequent defenestration has been likened to Steve Job’s ouster from Apple in 1985.
Without reinventing the wheel, I’d invite readers who have been living under a rock for the past week to Google other sources to catch up with all the drama llama. But the latest news to come out of this saga is Microsoft CEO Satya Nadella’s announcement that Sam had been invited back into the Microsoft fold under a (presumably) newly created business unit — where he will given influential oversight over some of MSFT 0.00%↑’s most prized assets such as GitHub, Mojang and LinkedIn. Oh, and probably a million-dollar signing bonus too.
Essentially, how all this got started was the equivalent of an emotional tantrum thrown by a kid whose dad didn’t let him get what he wanted in the supermarket. Except this kid had already raised between $11.3B and $13B of funding from external stakeholders (much of it from MSFT); and there was an imminent planned sale of shares by employees valuing OpenAI at $86B. Off the top of my head, the possible financial outcomes that naturally came to mind when all this broke out were:
OpenAI Investors get spooked and demand to exit their investments, sparking a crisis of confidence and a “bank run” of equity investments;
OpenAI employees see their years of hard work evaporate overnight due to an emotional tantrum, lose confidence in leadership, get demotivated and resign en masse in protest; (this actually almost happened)
The trillion-dollar business potential of OpenAI as a business unit of MSFT vaporizes overnight; not to mention the $10B they had already put in it.
Keep in mind the context of this spat. The argument between Sam and Ilya seemed to have been over whether OpenAI would take a for-profit or non-profit direction going forward. These were ridiculously tiny stakes relative to the usual bad faith contests that happen in Big Business with billions at stake. It would have been inconceivable by Nadella that MSFT could potentially lose its trillion-dollar golden AI goose over the equivalent of a nerd fight.
Perhaps also because of that, the leadership spat at OpenAI unravelled in possibly the most chaotic, irredeemable and public way possible — since no one was actually trying to control the narrative. It happened suddenly, with zero time for any of the involved actors to prepare, and with every minute detail being studied in real-time by the global press (Twitter). This was further exacerbated by the fact that it happened in Big Tech, where everyone was fluent in communicating online and information literally travelled at light speed.
On top of that, many of these voluntary journalists were also users of OpenAI — some very heavy users of them which had their workflow impacted by the recent outages at ChatGPT. There was a very real and present risk that OpenAI would lose its headstart in AI to other Big Tech competitors like Google’s Bard over bad optics — this was a purely marketing problem that superior technology couldn’t fix. Other large companies had been taken down by such marketing tragedies before (e.g. Chipotle), and MSFT was not immune to it either.
So not only did Nadella had to contend with managing the internal politics at OpenAI, he also had to deal with the court of public opinion that had been lighted up with OpenAI board’s suggestion that Sam had been fired for “not being consistently candid in his communications”. Pandora’s box had not only been opened up, but was experiencing violent diarrhea — and Nadella had to mop it all up.
To better appreciate the diplomacy supremacy that Nadella actually pulled off, let’s flesh out these two factors a little to provide further context to the aforementioned financial stakes. On the internal politics front, it bears reminding that neither Sam nor Ilya were dispensable at OpenAI in an operational context. Sam was the face of OpenAI and ran day-to-day operations; while Ilya was the brain at OpenAI, as he oversaw the development of its AI models. So Nadella had to somehow get them to kiss and make up despite the cavernous ideological schism that had opened up between them — in the most public fashion and with utter accountability for their public actions, too.
Secondly, neither Sam or Ilya were dispensable in the court of public opinion either. If Sam left OpenAI to another Big Tech competitor (or started his own), there was a good chance that ChatGPT would lose its commercial momentum as it shed an unacceptable number of customers to a competitor. If Ilya were publicly flagellated instead (for legitimate reasons, I might add), it would make OpenAI look weak to customers and stakeholders — perhaps into perpetuity (i.e. debranding). If OpenAI couldn’t even handle their own internal spats, how could they be trusted with the entire world’s private and sensitive information?
There were other orifices for the crisis to further unravel out of as well. A potential crisis in leadership confidence could have sparked an exodus of key staff members, as each member incrementally felt forced to choose a tribe with the weight of global accountability upon them (interim CEO Mira’s response to Sam’s tweet above was dissected by the Twittersphere into oblivion). This was on top of the major sense of loss from the employee share sale — which would have made each of them extremely rich, and which they definitely would have felt was unjustifiably rugpulled from under them, with no control over the outcome. In this timeline, MSFT would be scrambling to replace longtime OpenAI staff members with newbies, and could have lost its technical advantage as well.
On top of that, it seems like California doesn’t recognize non-competes. This meant that Sam and his generals could have just up and left to a MSFT competitor like Google and start working on Bard, providing the competitor with key insight into how OpenAI works. And let’s not forget about how the other OpenAI investors might have felt with millions of dollars each on the line. I’m sure Nadella must have spent a lot of time on the phone putting out fires over the past few days.
The whole of this meant that Nadella was on the clock to get the two founders to at least appear to kiss and make up. Basically reverse everything, sweep it under the rug, and pretend that nothing happened. But how could this possibly be done, when things had already unravelled so publicly and so out of control? Even if he somehow got Sam and Ilya to stop shooting at each other, they both had strapped directly contrasting bombs of public accountability to their chests — and one of them had to lose face in the court of public opinion. There was a structural irreconcilability between their two fates, perhaps by design.
And the way Nadella actually threaded the needle can only be described as magical. Officially, Sam was invited to oversee an entirely new business unit entirely separate from OpenAI and contribute towards some of MSFT’s other prized assets. As explained in the Semianalysis article linked below, this is just a cover for some OpenAI assets to be transferred to the new business unit where Sam can continue his previous OpenAI work with his own generals. In this way, he can also be insulated by any further machinations by Ilya — and Ilya can continue to seem like he’s captaining OpenAI. Meanwhile, Nadella can apply a stern hand to both of them and provide ‘adult supervision’. :
More importantly, this big brain move by Nadella comprehensively addresses OpenAI’s marketing problem resulting from this outburst in the court of public opinion. None of the underlying work that made ChatGPT so great in the first place has to skip a beat. Staff under both of Sam’s and Ilya’s business units don’t actually have any beef with each other, and they can continue to work on OpenAI together since MSFT has legal rights to all of the GPT model weights.
Honestly, Nadella’s ability to commandeer this ship and wrest control back from Poseidon’s wrath (the court of public opinion) took my breath away. There were so many other possible ways this situation could have spun out of control and shipwrecked OpenAI for good — and of all the paths it could taken in the multiverse, the events that occurred in our current timeline is borderline fiction (both how it started and how it was resolved). The way Nadella managed to thread the needle of the OpenAI schism is nothing short of diplomacy supremacy.
Value Investing Substack is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.