Hacker News Reader

Comment by bushbaba | Hacker News Reader

Welp that’s not a good sign for the anthropic IPO. Curious what happens now that both OpenAI and Anthropic will face budgetary pressure. I don’t think they can raise prices, as it’d just open ppl to switch to Open source models.

View on HN

It demonstrates why Anthropic has to IPO ASAP. Private investors are beginning to tap out. Anthropic is selling junior debt over 10% where it was about 8.5% last offer. If Anthropic wants to meet its compute purchase obligations (via actual business), it needs to start tapping public markets ASAP as its position is getting weaker. The size of the float and how much is primary vs secondary sale is the data we're missing right now. If the float is small and from secondary sales, the IPO is an exit and investors would be wise to stay away as Anthropic's runway stays where it is and therefore would require a re-acceleration of revenue growth to survive. If it is large and mostly from primary sales, it is a legit fund raise which extends the runway and revenue growth can be allowed to slow because of the pricing pressure you mentioned.

I believe Anthropic IPO is a pump and dump at this point.

Play the market accordingly.

Short the stock at the first available opportunity --- if there is an opportunity.

Shorting is more dangerous than buying long dated puts. At least with puts, if it's volatile there'll be opportunities to sell out of it if the direction isn't down in the long term.

It's so hard to time these things that short term bets are insanely risky. Yeah, thinks should go down, but they might or might not do it soon, and it might happen overnight because somethingsomething Yen Carry Trade or a surprise in the Middle East or some LatAm crash or an election result or God knows what.

Shorting is more dangerous than buying long dated puts.

Personally, I just buy inverse ETFs that target specific stocks.

The work going on to improve things on single card like a 3090 is nuts. HyperQwen, other things. I have my subscriptions, but always good to explore options.

You have to understand that Strata and HyperQwen are giving you good looking tokens per second but are not giving you anything close to the bf16 model’s capability (they quantize, worse they quantize KV caches, they approximate expert routing etc.)

The models are so impressive that getting 70-80% of their capability can be helpful, but beware, you are using a significantly dumber model.

I have subscriptions for necessary work and even pay for api calls.

The machine with the 3090 I built years ago and suddenly it’s worth 4x plus what I paid. It is for testing, but I use the model I run on it and the model I run on an mbp M1 Max with 64gig as workers for specifically scoped work.

I’d love to have a few dgx sparcs or the like to do heavy local stuff, but that money is best paying the pimps for now.

With luna's pricing, is the power consumed not likely to more than the cost of API calls?

At 50tps on a 3090 for qwen 28b or similar, that's about 200k tokens/hour. Which is what...10c for Luna (balanced input vs output). 1h of power at 30c/kwh is 30c? So you get a worse model, for more cost?

I pay about $0.14/kwh and don't need it all the time. API calls might be cheaper. I develop in a way where I break things out where well defined tasks can be handed to a local model.

Things like the hadamard rotations being applied before the quantisation mean that it has much less impact than before.

Thanks for the hardmard reference. It's pretty amazing what people can do with maths. I should have studied that instead of software.

If you look at the open weights official checkpoints they are either fp8 or even 4 (kimi). None is using bf16 for inference.

open weight, "how we trained it" is the open source of LLMs

real source is nice but boy have compile times gone up

Sure, but, in absence of money, where are the open models going to come from? They take a lot of effort and resources to train. The current ones have been made in hope the growth will lead to a payoff one way or another. With the two main giants struggling and the market cooling, that payoff becomes quite uncertain.

DOE, private labs, and universities are all working on building open models.

Here’s one example https://allenai.org/olmo

could be stone soup. a bunch of companies pitch in to train an open-weight model, a consortium or something. I think we need a return to something like the GPL for it to work, though. otherwise companies will take a good model and post-train it without contributing.

Could be state industry projects, maybe?

Claude may have captured a different mindshare of the past year who tried AI before and it didn't work.

The more people see AI is still just software (albeit a different kind), the more they will remember to instead start with where/how it works.

[flagged]

Selling the same product. It doesn't matter who is running the neural network, the utility derived from the neural network has nothing to do with the name of the company.

No, they sell different products in the same category. You are basically saying Google and Bing as the same product when Bing is superior to Google search in some non-critical metrics.

Folks certainly have their preferences, but I think most consider each company's roughly equivalent models to be of the same quality.

I would also say that Google and Bing are roughly the same thing fwiw.

It's more like a duopoly where the top two leaders have meaningful market share. Airbus and Boeing is more apt than Bing and Google.

Because they are? Google and Bing are the same product - search engines. Just like Ford and Ferrari both sell cars, whether you think one is "better" or not.

> No, they sell different products in the same category. You are basically saying Google and Bing as the same product when Bing is superior to Google search in some non-critical metrics.

In this case, it's pretty much the same product.

I mean, if you were to purchase electricity from two different providers (1x Nuclear plant and 1x Coal plant), would you actually care which one is being used to run your dishwasher on any given day?

I agree with this.

ChatGPT and Claude are differently operating and behaving products in the same space.

One is a little better/ different at considering the web, and the other is a juggernaut at sentences and reading in between / connecting the lines.

I'm also pretty proud, first flagged post ever for saying the obvious that they're two different companies.

They have different approaches to the tools they're providing. That fits people differently.

Like, no name of a company must or has to have anything to do with the product itself.

That's correct.

[flagged]

Please don't be snarky on HN. The guidelines make it clear we're trying for something better here. https://news.ycombinator.com/newsguidelines.html

I thought I had finished my sentence - the companies have different GTM, and maybe different trust/experience.

Reply on news.ycombinator.com