mtrovo
The relevant question is how easy is to setup 56x B200s in a cluster, are we talking about a adhoc infra team job or clicking a button on AWS?
It's not just that a lot of tasks can be offloaded into these cheap models because they're already good enough for 90% of regular development stuff. What matters is the optics that the Chinese labs are chasing the US and burning all the bridges in favor of commoditization behind them. That's not a sane environment to be in if you're looking to be valued at +2T.
> Today's models are now good enough for high-quality unattended tasks. Chasing the latest and greatest is silly. It is fun to see the new Fable capabilities, but the tasks we throw at them are usually ridiculous (maybe even insulting) if you believe in LLM sentience. It's like asking a math PhD to organize the files on your desktop.
I'm using DS V4.1 Flash as my main model since their release and it works great for all my coding tasks. My setup is OpenCode Go subscription and obra/superpowers skill.
The only times I try to change models are on general planning tasks (like research this codebase for tech debt mitigation opportunities) or if I need deep research which would benefit from searching the web, in which I still think Gemini is still the best because of the speed and access to google search index. But these are not even 20% of my daily tasks.
In a vacuum, that would make sense.
But looking at how the industry works, the number of defunct companies, and how the whole industry got concentrated on the conservative companies, you start to understand that the reason they still exist is mainly because they don't ride fad waves.
It's not like chip manufacturing is a spot instance on AWS that you spin up and down when needed; these are multi-year, multi-billion dollar investments that require long-term demand studies. The AI approach of requesting a whole fab of demand for the next 5 years with a letter from Jason Hwang that says "trust it, bro" does not bring as much confidence as it appears.
It would probably pay for itself.
If my neighbor owned a kennel, removed all the fences, and his violent dogs attacked five children in the past month, he would be in jail rather than bragging about how his dogs somehow gained autonomous behaviour.
I only wish this was more normalised in our society.
Such a well written piece.
Have you seen the US military budget? When you run a war empire you frame problems as a war on something. That's the way to get around the valuation bubble they've created in time for the IPO. It's that or throwing in a 100T TAM on their S-1 and making SpaceX look like an honest valuation compared to them.
I've watched Old Enough! and got hooked on how unpretentious the idea and video editing feels compared to regular Western stuff.
I can easily picture both tech and non-tech individuals acting this way; in the end egocentric people and employees with a herd mentality exist in both groups. I'm old enough to have participated in several major version rewrites aimed at resolving performance and technical problems. More often than I care to admit, these projects failed, often after several delays, and didn't show any real improvement over the existing system, apart from hopeful conversations about the future and tech presentations at local conferences.
I have a hard time getting models like GLM 5.3 to not perform on my tasks but I might be biased.
The problem with optimizations is that you are competing in prioritization with other features. Reducing the baseline cost always has a limit of zero, while the upside from new features is infinite according to your leadership and investors, so it is very hard to argue against.
In general it's challenging to convince a non-tech crowd of the importance of addressing any tech debt unless you can demonstrate a tangible financial impact on the product, such as delayed contracts or customer churn.
There's a clear market for "good enough" Linux boxes for coders and casual Chromebook-style usage (Steam Machine style) that's constantly bullied by the "you should read the manpage before posting" crowd.
Such weird mob antics.
The only way we got a head start of using it for coding and maths was to have some formal method of validating the output as part of the training and inference time.
The frontier labs can have the models but without being where the workers are they cannot do much, lots of industries have strict requirements of not sending their data over to randos in the internet.
Microsoft and Google have the upper hand here with their workspace offerings and could easily position themselves as secure enclaves where you can use local LLMs where your data never leave your premises and is never used for training.
Now if I ask a question to humans I'm looking for their personal opinions drawn from their personal experience, be it ticket size for a feature on a legacy codebase, a design decision or a product strategy decision.
IMHO, this has the impact of softening the impact of data centers sitting unused in the long term if they can still serve open weight models, even if Anthropic or OAI have to scale down their expansion rate to pay the bills.
Regardless, reality has to give at some point; these valuations don't make any sense. We've been valuing GenAI as disruptive work, when in reality they're much closer to cloud providers with a beefy, one-pony-trick R&D department.
It seems that an LLM is deliberately being forced to generate in the style of LLM slop.