bwfan123

> Hacker News is no longer a web site of an elite

hah, sorry, we are plebs out here.

Thanks for the input. Let me check it out.

> leveraged financial nukes inexperienced and FOMOed investors have assembled

A lot of debris waiting to happen in data-center debt for sure. The next several years are going to be interesting with more of these type of blowups. This is a canary in the ai-wreck coalmine.

> What could possibly go wrong

A barometer of the mania and one for the history books. A 23 year old wunder-kind publishes a 156 page prophesy with a catchy title which was lapped up by the silicon valley elite and funded to the tune of 10s of Billions. And, not sticking to his lane, the wunder-kind enters the finance world thinking his brilliance translates there too, and he was promptly taken to the woodshed by the wall street sharks.

> Are you buying from a provider that isn't listed on open router already

I am using openrouter with zdr guardrail which routes to any provider that supposedly doesnt train on user data. I also use fireworks (directly, not via openrouter) which is a provider promising zdr and has a bunch of open weights models. My issue is that these zdr providers dont transparently disclose caching/tokens etc and so they end up being far more expensive than directly using DS.

> They announced it already

I meant something I could download and run.

Deja-vu from dot-com. The tech-wreck had similar hedge-fund road-kill. Munder net-net fund comes to mind. A lot of hubris and leverage on a thesis that is not proven, and liquidity matters. Data-center debt will likely see similar debris in the next several years.

> Undercut the US AI providers and wait them out

You could have said the same of linux - that it extinguished proprietary OSs at least for server usecases.

> For the Code Agent tasks among the public benchmarks above, DeepSeek-V4-Flash-0731 is evaluated with the minimal mode of DeepSeek Harness (to be released) as the agent framework

So, are they planning to announce an optimized coding agent harness as well ? DSv4 flash is a fantastic model, and my daily driver. With reasonix or pi, I can code all day long and pay a few pennies for it. No token anxiety. Whereas the same model with fireworks/openrouter, with zdr thrown in, token costs ratchet up with no explanation. Likely that the model is subsidized for gathering usage data. I am waiting for the day I can run this locally.

I guess it gets bigger with each passing day. Tens of billions goes to Hundreds of billions.

> Not understanding to the core the software you wrote, is going to exploded in your face.

Short-termism at its peak. Code maintainers will learn the hard way how to set the boundaries between ai-generated code, and human maintainable code.

There needs to be wider reporting of this if it is true. Where are the journalists ? If this is true, it is scorched earth on the world's knowledge base. And the business models are not proven yet. I can now see why some ai companies paints a dire picture of the future - they are basically annihilating all knowledge sources at the altar of their ai gods. I can also see why there is growing ai-skepticism.

> Well its clearly a stunt. If it wasnt we would probably be up to our ears in technical detail.

LLMs are the script kiddies of the day.

Tax codes are a great example of something changing frequently, and something very nuanced. So, the training data doesnt exist for a future year's return while are rules are already changed.

So, instead of math/programming, tax codes approximate the real world better. Whereas math/programming/images/languages are frozen in time. I find this fascinating. Almost like the legal system, wherein rules are laid out in english but are almost mathematical in the sense that there is a legal/illegal judgement made for almost every action, but that judgement is not easily determined.

I was waiting for the announced customer of Meta's excess AI capacity. Lo and behold, it is Anthropic.

At this point, openai and anthropic have no problems announcing these deals because is good narrative, and helps them raise funding and IPO.

The same thing happened in dotcom. Deals and partnerships were announced because it helped raise money. I recall energy companies (enron/dynegy) doing deals with network vendors. Eventually, the deals were not worth the ink on paper but it helped raise funding.

Love the writing style, and I wish all math was written this way. These articles are accessible to anyone who has had some high school math background, and paints a historical backdrop on how the sausage was made (with links to the original papers). Some of these ideas took millenia to develop, and when presented in neat capsule, students are cheated of the mystery and struggle that went into the progression and development of the ideas. I would argue that all math education has to be re-invented to follow this style, ie, uncover the historical progression, wonder and mystery.

> There are open source alternatives that have >90% of the functionality, they'll even work w/ Excel files and generate them

Excel has network effects. If N people are using excel, the incremental N+1'th user is forced to use it or risk not being able to open documents from the N users. Excel is also sticky, in the sense that a user used to Excel UI is hesitant to switch to a new UI. Tokens dont have network effects. If N people use fable, the N+1'th user could use any other model, and have no impact.

> So, the whole brouhaha was so that Anthropic can write this blog post?

Remember mythos ? I was half-expecting them to scare us with a new danger: How AI can covert code into other languages, and confuse the hell out of all of us.

So, you have a vibe-coded TUI which happens to work, and then, as a workaround you vibe-translate its engine to make it more performant. Where does that leave you ? Basically, fully dependent on AI to fix whatever breaks. Workaround on a workaround is the way I see it, and it aligns with the AI design mentality in general. For a variety of usecases, this might still be a win in terms of overall cost. But, for software that is intended to be built to last, I dont see this approach working out.

> it usually is good enough for most tasks

The model is fantastic. And costs almost nothing. The only problem I see is that they will train on your data.

There are zero-data-retention providers of DeepSeek models, of which I have used openrouter (with zdr guardrails), and fireworks. But these are 3x to 5x more expensive than directly using DeepSeek, possibly due to poor caching. Thats the price to pay for zdr.

> I never truly understood what the intended business model around LLMs was

What appeared initially to be a huge innovation was later easily duplicated by many. There are no platform-lockins or network effects. Switching costs for users are zero, and there are low barriers to entry, with vast numbers of models to choose from and more appearing every day. As a business a token will be a commodity like an electron. Doesnt matter who produces it, or how (solar, wind, coal, nuclear etc) as long as it powers my toaster.

> who do you make money from?

When you own stock at a broker in a margin account, you may sign an agreement to allow the broker to lend out your stock to someone else. For lending your stock, you are entitled to a stock-borrow fee which usually is quite small say 0.25%, and paid by the borrower (short-seller). The borrower then sells the stock to someone else. At a later point, the short seller closes their position by buying it back, and returning it to you. This is roughly the mechanics of it. So, to answer your question, the short seller makes money from folks who buy high and sell low. In this specific example, the stock-borrow fee say was 5% because, the float is still low, and if the short seller borrowed at $165 after the IPO and sold it, and then bought it back at $135 and closed their position, they made money from folks who bought at $165 and sold at $135.

> but the whole gang of frontier labs, breastfed with a mix of VC money and pension funds from the average Joe, is being pressured by Chinese lab

Can you point to data that suggests rate of token revenue growth is slowing ? This announcement, and the price cuts from openai are hints in the direction of pressure from open weights models but not conclusive.

> AI is largely automating the most tractable parts of science rather than expanding its frontiers

By definition, creativity cannot be automated, and AI is a fantastic automation machine. It can explore thinking paths at a rate humans cannot match. But creativity is bringing the unthinkable into the thinkable, and that requires sensory experience [1]. Specifically, new definitions and symbols which never existed before. Imagine the concept vector space, and expanding that with new independent dimensions. Is that even possible ? When you look at history the answer is yes !. And each time there was an independent dimension added, it was an act of genius. It is an instructive exercise to name these moments in history where an independent dimension was added to human thought. Some examples in math would be the invention of a number, and in politics could be the idea of democracy. By contrast, LLMs are trapped in the vector space they are trained on, and they lack the feedback loop with sensory experience to be able to create and validate theories.

[1] https://philsci-archive.pitt.edu/28024/1/Scientific_Inventio...

> For almost everything else, the answer is no. No one else would pay the real costs to run them. It'll require the whole industry to shrink down massively compared to what we're seeing now - down to a profitable (and much smaller) core.

If there is any data to support this, please share.

Circular financing is a dead horse - dont beat it. Instead, what is more interesting could be: Is there a path to these builds becoming economically profitable ? Towards this, some metrics to watch are: 1) ROI per token per dollar 2) Enterprise token budgets. And at what point there is an overbuild relative to the token roi. Alternatively, pressure on token costs due to the open weights models etc.

> I'm below $1 writing code all day.

same, pi as a harness works best, better than claude code and open code.

> All these "rewritten in rust" projects only reinforce the idea that a significant part of the rust community consists of software talibans and not of engineers who must deliver something that works and is reliable over time.

Nailed it ! There are some folks who behave holier-than-thou just because they happen to use some language. Language missionaries if you will, and they are insufferable.

yea, I found LLMs poor at reasoning through system-wide behaviors. Good luck to the vibers when they try to rootcause a rare data-corruption problem. They will hand it off to fable which will attempt trial-and-error bug fixes which will eventually muck up the code. What happens then ?

Doesnt this prove that there is really no moat in proprietary models for coding usecases, or the gaps is narrowing ? Also, since GLM5.2 can be run equally on amd and nvda, I guess there is no hw moat either. Further, switching costs for users is minimal not only in agents but models too. So there is really no stickiness or user preference involved. For this usecase I think it is a race to the bottom for costs in a good way for developers.