Yeah, I was waiting for an actual 9 headed hydra of crap, featuring:
* a whole page of sponsored-yet-indistinguishable-from-organic results
* a confidently incorrect AI snippet
* the below-the-fold top organic results being all content spam
* the first actually relevant result from a reputable source being then paywalled with a "you ran out of stories for this month" overlay on a website I have never browsed in a year
* the 2nd actually relevant result prompting a "login with google" overlay with prefilled identity that gets clicked by accident 20% of the time
* all of the above in a Chrome browser, requiring a quadruple opt-out before allowing you to use GMail without also starting a browser-wide session to keep track of your every keystroke
* or alternatively an app based mobile interface where links can't be copied and pasted to prevent loss of tracking
* something with AMP
This is not even close to everything that is wrong, if anything it should be called "barely the tip of the iceberg of everything that is wrong".
In their defense it does feature a bunch of other invisible things (lack of cached results, lack of direct links you can obtain with a right click, etc)
I can't imagine the hoops an accountant would have to go through to argue training cost is COGS. In the most obvious stick-figures-for-beginners interpretation, as in, "If I had to explain how a P&L statement works to an AI engineer", training is R&D cost and inference cost is COGS.
I wasn’t using COGS in a GAAP sense, but rather as a synonym for unspecified “costs.” My bad. I suppose you would classify training as development and ongoing datacenter and GPU costs as actual GAAP COGS. My point was, if all you focus on is revenue and ignore the costs of creating your business and keeping it running, it’s pretty easy for any business to be “profitable.”
It’s generally useful to consider unit economy separate from whole company. If your unit economy is negative thing are very bleak. If it’s positive, your chance are going up by a lot - scaling the business amortizes fixed (non-unit) costs, such as admin and R&D, and slightly improves unit margins as well.
However this does not work as well if your fixed (non-unit) cost is growing exponentially. You can’t get out of this unless your user base grows exponentially or the customer value (and price) per user grows exponentially.
I think this is what Altman is saying - this is an unusual situation: unit economy is positive but fixed costs are exploding faster than economy if scale can absorb it.
You can say it’s splitting hair, but insightful perspective often requires teasing things apart.
It’s splitting a hair, but a pretty important hair. Does anyone think that models won’t need continuous retraining? Does anyone think models won’t continue to try to scale? Personally, I think we’re reaching diminishing returns with scaling, which is probably good because we’ve basically run out of content to train on, and so perhaps that does stop or at least slow down drastically. But I don’t see a scenario where constant retraining isn’t the norm, even if the rough amount of content we’re using for it grows only slightly.
Well, models are definitely good enough for some things in their current state, without needing to be retrained (computer translation for example was a solved problem with GPT3)
That’s true but irrelevant. No AI company is stopping training and further model development. OpenAI didn’t stop with GPT3, and they won’t stop with GPT5. No company, AI company or not, stops innovating in their market segment. You need to keep innovating to stay competitive.
Got it, it's just an awfully specific term to use as a generic replacement for "cost" when the whole concept of COGS is essentially "not any cost, but specifically this kind" :)
Its possible your example is wrong and the confusion arises in the "twelves" (i.e. 1230 AM being 0:30 which makes no sense since anyone that didn't learn the quirk would expect the AM/PM bit to roll over at the same time as the hours, not one hour off the mark).
