If I could have one of these cards in my own computer do you think it would be possible to replace claude code?
1. Assume It's running a better model, even a dedicated coding model. High scoring but obviously not opus 4.5
2. Instead of the standard send-receive paradigm we set up a pipeline of agents, each of whom parses the output of the previous.
At 17k/tps running locally, you could effectively spin up tasks like "you are an agent who adds semicolons to the end of the line in javascript", with some sort of dedicated software in the style of claude code you could load an array of 20 agents each with a role to play in improving outpus.
take user input and gather context from codebase
-> rewrite what you think the human asked you in the form of an LLM-optimized instructional prompt
-> examine the prompt for uncertainties and gaps in your understanding or ability to execute
-> <assume more steps as relevant>
-> execute the work
Could you effectively set up something that is configurable to the individual developer - a folder of system prompts that every request loops through?
Do you really need the best model if you can pass your responses through a medium tier model that engages in rapid self improvement 30 times in a row before your claude server has returned its first shot response?
But at one point the model is sufficiently large enough to accomplish any task a human could specify. For software development, I think we're pretty much at that point with the latest Anthropic/Google/OpenAI models. We have no idea where the direction of token pricing is going to go in the future, but the consensus seems to be that it will only get more expensive. If Taalas can offer the same functionality that we have with frontier models today at a 1/10 of the cost and 10x the speed then they're going to take over a large part of the market.
I think so. The last few months have shown us that it isn't necessarily the models themselves that provide good results, but the tooling / harness around it. Codex, Opus, GLM 5, Kimi 2.5, etc. all each have their quirks. Use a harness like opencode and give the model the right amount of context, they'll all perform well and you'll get a correct answer every time.
So in my opinion, in a scenario like this where the token output is near instant but you're running a lower tier model, good tooling can overcome the differences between a frontier cloud model.
It's 2.5kW so it likely won't sit in your computer (quite beyond what a desktop could provide in power alone to a single card, let alone cool). It's 8.5cm^2 which is a beast of a single die.
Basically logistically it's going to need to be in a data centre.
It's ideal for small context high throughput. Perhaps parsing huge text piles like if you had the entire Epstein files as text.
I think Claude code benefits from larger context to keep your entire project in view and deep reasoning.
What this would certainly replace is when Claude dispatched to Haiku for manual NLP tasks.
> It's 2.5kW so it likely won't sit in your computer (quite beyond what a desktop could provide in power alone to a single card, let alone cool). It's 8.5cm^2 which is a beast of a single die.
I wonder how you cool a 3x3cm die that outputs 2.5 kW of heat. In the article they mention that the traditional setup requires water cooling, but surely this does as well, right?
According to google Tesla has shipped 8 million cars total since inception.
It is valued at 1.32 Trillion as of today. Which is roughly $165,000 per shipped vehicle.
Right, let's not forget that he's selling it to himself in an all stock deal. He could have priced it at eleventy kajillion dollars and it would have had the same meaning.
He's basically trading two cypto coins with himself and sending out a press release.
do it on mars then... that would have the added benefit of heating the planet so we could live on it. It seems so obvious if you think about it. Someone transfer eleventy trillion dollars to Elon Musk so we can get started.
the 4hww was such a damaging read for me. I wasted YEARS squeaking pennies out of shitty, poorly-realized side hustles trying to solo that shit when the real ROI would have been to focus on leetcode in my 20's and track down a role paying 10x the salary.
When I first read it, I was probably making about $60-65k and scrambling for $2.5k raises with a shitty meat-grinder employer while burning the midnight oil and acting as a "weekend warrior" trying to come up with something that would allow me the freedom that ferris spoke of.
Problem is, all of his techniques presumed you already had some sort of business that you could automate yourself out of.
If you're a wage-earner none of the shit he talks about is realistic. Sure, you can do the research to become a drop shipper or to place tiny ads in the back of soldier-of-fortune for brain pills.
For the typical computer programmer - your ROI will be 100x higher if you just get good enough at algorithms and system design to land a role at a company with Tier3 compensation.
when people do these complex CPU designs in minecraft are they laying the blocks individually in real time and in 3d space - or are they scripting some sort of algorithm that instantiates the system in one go?
It's impressive either way but the manual version seems ... impossible.
What an oddly editorialized version of this headline which does not appear to reflect the conclusion made by the article OR the actual headline of the linked article:
actual headline:
> Is Marvel leaving Georgia? Production shifts to UK spark industry shakeup
piece talking about reasons:
> shifting to the United Kingdom, where lower production costs, especially on wages and employee benefits, are giving studios more bang for their buck, according to the Daily Mail.
looks like differences in wages AND benefits. Lower overall salaries in the UK combined with public healthcare costs via the NHS is a much more nuanced take than "leaving to avoid paying health insurance."
I suppose the conclusion though is the same - that a public option would have prevented this flight of capital. Not necessarily dropping of mandates that employers cover health insurance to remain competitive (which I'm sure some will conclude).
I'd argue that Meta's income derives in no small part from their best in class ad targeting.
Being on the forefront of
(1) creating a personalized, per user data profile for ad-targeting is very much their core business. An LLM can do a very good job of synthesizing all the data they have on someone to try predicting things people will be interested in.
(2) by offering a free "ask me anything" service from meta.ai which is tied directly to their real-world human user account. They gather an even more robust user profile.
This isn't in-my-opinion simply throwing billions at a problem willy nilly. Figuring out how to apply this to their vast reams of existing customer data economically is going to directly impact their bottom line.
5 minutes on facebook being force-fed mesopotamian alien conspiracies is all you'll need to experience to fully understand just how BADLY they need some kind of intelligence for their content/advertising targeting, artificial or not...
You probably don't spend enough time on their sites to have a good ad targeting model of you developed. The closer you are to normal users, with hundreds of hours of usage and many ad clicks, the more accurate the ads will be for you.
Same terrible experience for me while I was on FB.
I was spending a lot of time there and I do shop a lot online. They couldn’t come with relevant ad targeting for me.
For my wife they started to show relevant ads AFTER she went to settings and manually selected areas she is interested in.
This is not an advanced technology everyone claim FB has.
People look at all the chaos in their AI lab but ignore the fact that they yet again beat on earnings substantially and directly cited better ad targeting as the reason for that. Building an LLM is nice for them, but applying AI to their core business is what really matters financially, and that seems like it's going just fine.
1. Assume It's running a better model, even a dedicated coding model. High scoring but obviously not opus 4.5 2. Instead of the standard send-receive paradigm we set up a pipeline of agents, each of whom parses the output of the previous.
At 17k/tps running locally, you could effectively spin up tasks like "you are an agent who adds semicolons to the end of the line in javascript", with some sort of dedicated software in the style of claude code you could load an array of 20 agents each with a role to play in improving outpus.
take user input and gather context from codebase -> rewrite what you think the human asked you in the form of an LLM-optimized instructional prompt -> examine the prompt for uncertainties and gaps in your understanding or ability to execute -> <assume more steps as relevant> -> execute the work
Could you effectively set up something that is configurable to the individual developer - a folder of system prompts that every request loops through?
Do you really need the best model if you can pass your responses through a medium tier model that engages in rapid self improvement 30 times in a row before your claude server has returned its first shot response?
reply