Lets be clear here. Zuck is betting on his own bias that he will succeed at this, that society will accept it and that it will be beneficial to enough people that they will pay for it. His track record says otherwise, general public opinion is shifting now it is seen as putting people out of work and there isn't a working financial model yet based on the ROI and retention figures.
----
Prediction
Fact: The ROI from investment on this is not showing up at all and behind the scenes the private equity firms bankrolling it have started moving capital out to make positions for the hype investors to buy the risk out. They are out.
Agree for small non-frontier model companies, they are never getting off the ground and staying there long-term unless someone acquires them. ROI doesn’t matter to half of the large players and China is already putting out conservatively 300% more research than the west which will continue to drive demand for even older hardware for years.
Opinion: What I expect will really happen is there will be a capital crisis within the next year. The general direction of things at the moment is pretty awful and that makes people hold on to money rather than make risky investments. When it's individual investors bankrolling the markets and they are bearish due to political instability, worries about work and war etc, there will be no more capital available unless you're a long term stable low-risk investment. Targets will be missed, investor confidence will decline and it'll snowball into a tech stock crash. Because all the major ETF constituents are tech stocks it will wipe everyone's "secure" personal investments out at the same time which will damage market confidence further.
There’s too much cash on hand for this to materially matter that soon. Everyone always tends to predict doom for the near-term and while it occasionally happens it is transient. Elon and Zuck will fund AI at any cost for a couple years at least, they don’t care about share price since they have controlling interest in their companies, e.g. no one can force them to change their mind.
FB / Microsoft / OpenAI / Anthrophic / Google / xAI will be dead with no capital to fund further research and model Normality will resume by 2027.training. NVidia will suffer a huge demand drop and have to crawl to TSMC to go back on previous capacity demands. GP
Meta and xAI are beholden to no one. Zuckerberg has independent majority control and xAI doesn’t even have public shareholders. They are going to chase this down at least until world models are proven or disproven which will take 5ish years.
Google will pivot to ads eventually, OpenAI is going to launch a browser. Google has half a BILLION users of Gemini as of this summer. It has absolutely crossed the threshold to be something that will exist for a long time and ironically a lot of the most tech-savvy people are the ones who are the most willfully ignorant about this fact. It took me a while to come around, too.
Normality will resume by 2027.
…2026/2027 crash.
In 20 years everyone will wonder what the hell the CoPilot button was for on PC keyboards they find in the trash while complaining that they Microsoft rewrote their photos app 5 times in the last couple of decades and it still doesn't work.
I’lll take that 2027 bet.
Will send you a dollar if you’re right, feel free to @ me in January 2027 if the AI trend completely collapses.
I think there is a fairly reasonable chance that within 2 years a single frontier model will be useful for ~3x longer than they last currently and it will financially pencil out then because the cost is in heavily front-loaded in training. They will create even more expensive tiers for cutting-edge and keep regular people on the old version for longer periods of time once the arms race settles down a little.
The market is probably going to tighten but it won’t collapse. There is undeniably utility in these tools as they are today, and like I said above many technologists who used ChatGPT a year ago are woefully out of date with current capability and utility. Just opening up internet search dramatically improved things since they can source other data, and the models connected to realtime data (e.g. Grok 3) have benefits in a day-to-day way that no model a year ago could.
Anyone who is grouping this all together as one “mass” of ill-fated technology is ignorant, whether willfully or not. It doesn’t mean I endorse the existence of these things but they are here to stay and have achieved critical mass. The internet did a lot of harm (and good) for society. We will see how truly next generation technology pans out.
I’m not a blind AI booster, I use every major model regularly along with local ones occasionally that I alter parameters to check on output strength and coherence. The field is evolving rapidly, and a lot of the takes about this technology are emotionally driven, not rationally.
I do personally think Anthropic should find a buyer at their current valuation if possible, and whoever gets them will have made a pretty wise investment, if they can be competently managed which is a whole different hurdle (e.g. copilot).
