That’s not just a continuum but also a timeline. Junior devs need to be encouraged to start thinking about why as much as what, and what more than how.
But they don’t teach you that in school. You get how and a bit of what. Maybe a little why in a sales pitch or an aside from a professor who hasn’t worked in industry except in research which is prototyping and we all know how people who only do prototypes are about rigor.
Wiles' proof of Fermat's Last Theorem was 129 pages and lean proof was 13 million lines
Interesting in this context because Alpöge has the lean proof but not the human readable variant while Alessandro Masullo proof has the human readable proof but not lean proof. This might support Alessandro's claim that it is not an AI proof, and it will be particularly exciting if they both solved the Reimann Hypothesis using different routes.
I imagine that's going to happen at month 6 when everyone with holding limitations can cash out and, ostensibly, OpenAI decides to go public shortly after.
I agree with you that nothing is certain, but their profitability is irrelevant if their business loses durability. Local models and many outstanding obligations loom over them. This kind of long-term thinking is irrelevant in the short-term markets, but I’m not going to buy and hold this any time soon.
local models will never be as good because of the prohibitive cost of running large models. unless there is some sort of a hardware breakthrough Anthropic business is solid and almost all of its revenue is from enterprise, not redditors doing role playing or generating cartoons on their 3090s
I work at an enterprise with one of these subscriptions. I see my direct cost usage every day. It’s not difficult to push it towards my daily pay, despite requiring me for it to be truly useful. They want to sell the idea that they can replace me or a junior, but the data coming in to that end doesn’t check out. Back on the business side, we’re now at nearly full SaaS margins (85%, in this case). Take API usage for example. The bull case is continued proliferation of models into direct process, like blindly inserting Fable 5.2 at high effort into your customer service ticket SaaS pipeline with a human still somewhere in the loop. You might say my human utility as a staff engineer is to look at that and question if it’s worth it. We just watched decision models dramatically cut costs there. I expect more and more of these kinds of optimizations, where models and usage are fit to size. People get promoted on optimization and cost-cutting competition: savings they can pass on to customers or increase their own profits. The market has many more niches to fill, basically. It’s also not difficult to make the case that I can only make Fable so smart about my business to mgmt. So it’s hard to keep juniors away forever, especially considering that I’m not getting any younger. They’re betting on the idea that they can make it five years without the majority of Fortune 500 companies shifting their mindsets very far towards my position at all. This isn’t hyperscaler cloud over again either, where they have a seemingly unlimited ability to keep offering unique services. It’s not that I think they’ll implode overnight. It’s only that I’m skeptical of the idea that a singularity or even the ability to run Fable on a laptop is required for their disruption. I could see a simple marketing blunder doing it.
In Synology's case a lot of their stuff is just legacy even though they don't describe it that way, I think most of their apps have not been worked on for years. At this point I wonder if they even have much more than a skeleton crew working on their software at all.
About 1.1T market cap will exist tomorrow prior to opening bell, out of a potential 1.5T that includes .4T stuck in options, convertible notes and such.
But it’ll likely open above 83 when trading actually begins. We may see upwards of 2T if it goes over 100 a share
Perhaps they were expecting to be more oversubscribed? Like they increased the number of shares offered to keep a target amount of money. Likely they see the rough few months ahead. Ramp.com AI index shows spend on Anthropic AI was in decline/flat July 31 - Oct 31
Likely they're best bet is to weather the continued devaluation of tokens and cheaper models until they hit on a better business model, which honestly is probably not that hard.
Thanks for sharing. Something seems off in that data. It says in Nov, Anthropic's share of token volume is 55.1% and share of token spend is 49.3%. Given Anthropic tokens are pricier than other vendors, one would expect share of spend to be higher than share of token volume.
That's a good point, likely there are other discounts / subsidies being realized on accounts? I think from Ramps point of view it might be hard to distinguish these.
That being said, I have also noticed that at times the X axis dates of the weeks changes. In the past, like a week ago, I noticed that the "Week ending Oct 23" changed to "Week ending Oct 25". Which also noticeably changed the shape of the charts.
I work with data a lot so cut them some slack, but I do with they had more grounded information in the report (a lot of % and not a lot of specific values).
I also wish they would somehow tie in to the massive OpenRouter Rankings. I think the two together would help paint a much more cohesive picture of how token budgets are spent and change week to week.
I wonder if early adopters are overrepresented in Ramp's customer base, so if they start learning how to get the best bang for buck in their AI spending at the same time the laggards are starting to spend real bucks in AI, Ramp's data may be bearish while the reality is bullish.
Oversubscription was kind of a self fulfilling prophecy. Large investors and institutions placed larger orders knowing the IPO would oversubscribe and they wouldn’t get their stated allocation.
It's a common pattern for IPOs, regardless of demand. Companies gauge the price range during the roadshow, make an initial announcement at a low price, then raise the target at the last minute going into the IPO to make it look like a hot offering.
The previous range was 70-78 and it priced above the target range, at $83.
So they are selling 20% more shares and selling them for 15% more than the midpoint of what they were targeting, because the demand is overwhelmingly high. They are going to make $20B more from the IPO than planned.
The rumor is they were oversubscribed by more than 6 times over. SPCX was reportedly only 3.5-4.
And RH allocated me 1 pre-ipo share against my request of 12 of them. so one in a billion shares. RH shows I'm the 260,000th retail investor of the ANT offering (at least within RH ecosystem[?]). Just observing, but pre-IPO share allocation does not appear outwardly to be an activity rooted in free-market fundamentals.
Today Anthropic, PBC priced its initial public offering of its Class A common stock at $83.00 per share. The offering consists of 980,000,000 shares of Class A common stock, 956,351,354 of which are being sold by Anthropic and 23,648,646 of which are being sold by certain selling stockholders. Anthropic will not receive any proceeds from the sale of shares by the selling stockholders.
That's the early investors cashing out. If my buddy starting a business selling dog toys, and I invested 10 grand in it to get it off the ground, in exchange I would receive some fraction of his company. If that dog toy business took off to the point that he's going public, I'd still own that fraction of the company, and I'd probably put some of my shares up for sale.
They will be the first and thus for now super-scarce frontier lab "AI pure play" stock and many people that are mouth watering for that will buy despite their fundamentals. But Value opportunity isn't really there right now. (It was there for early investors in private equity)
They also market themselves as a public benefit company and safety driven, which are not priorities for growth maximization. Their main page says "Anthropic is a public benefit corporation dedicated to securing (AI's) benefits and mitigating its risks."
Then there's cost pressure from the competition, like oAI's cost per task being quite cheaper with only marginally less intelligence than ant's models.
So there doesn't seem to be a clear path to long term upside. But short to mid term and before OAI's own IPO it's likely to attract more money than it should.
If you weren't doing it, you were doing it wrong.
reply