Is letting people spend $6,000 on tokens for $40 a sustainable business model?
>>108816518Anyone who believes the AI shit's about money has a dangerously low IQ understanding of the world.
They'll make it up in volume
Don't worry, they'll just write that off and claim it as profit.
>>108816518Steve "Sweaty Balls" Balmer, peace be upon him, did not build Azure for this shit.
>>108816531>the AI shit's about moneyI'm sorry?
>>108816531>the AI shit's about money
>>108816518um yes this is called the "bait and switch" business modelbut usually you're supposed to do this in a market where it's hard to switch doesn't seem like copilot has lock-in so who knows
>your totally made up number would've DEFINITELY costed this OTHER totally made up number, because, it just would've, OK?
>>108816635well it has a bit of believability when it's coming from the company that plans to bill you with the new method
>>108816619>>108816607>>108816590>Low IQ world understanding detected
>>108816518Crypto will fix this :)
>>108816698>>Low IQ world understanding
>>108816698>Low IQ world understanding detected
>>108816531This is grammatically correct, the replies to this are by ESL retards."Anyone who believes the AI shit is about money""shit's" is a perfectly grammatically acceptable alternative here.
>>108817007>"shit's" is a perfectly grammatically acceptable alternative here
I don't understand the focus on software really.I guess you could argue software was always the bottleneck since programmers would just shit all over the progress made on the hardware side.But now AI is just about shitting out slop as fast as possible and not about quality.Like for AI to get extremely good at software it would basically have to be AGI and at that point why would you care about software.Literally none of it makes sense
>>108816518what they lose per token they will make up in volume
>>108817178they know LLM is dead end for achieving AGI so they're looking for some usecase to sell their products for. Or at least to boost evaluations before IPO
>>108816531Tard fren, if it wasn't about money the DoE or DoD would be spearheading it directly instead of whoring it out to Anthropic or OpenGayI
>>108817178>why focus on softwareIf someone manages to make software that can improve itself autonomously or with little intervention, it should exponentially benefit all other areas eventually on its own since it can just self improve and/or creat its own adaptions to specific workloads.It's interesting to see people self feeding like this already. Using agentic software while developing said software. The next step is reducing the human in the middle until they're no longer necessary.Will this happen? idk but I think that's the reason it makes sense to target it.
>>108817595The models OpenAI And Anthropic have are just what the DoD/DoE had in the 1990s
>>108817751
>>108817636>software that can improve itself autonomouslyis there any example of an LLM improving software written by a human?
>>108817786
>>108817799>is there any example of an LLM improving software written by a human?Depends. Do you consider Indians to be human?
how do you know they aren't overcharging for api calls?
isn't that how tech products work at first?they release it at a loss when they're new, and then eventually it's $5,000/month with ads everywhere.
>>108817799I mean I've used them to minor benefit here and there, paired with all the wasted time.But obviously it's typically subjective and anecdotal what an improvement is. Although you can have objectivity with benchmarks, test coverage, and other hard metrics.Writing good code or improving an existing codebase does not seem to be too difficult to make these machines do. However, in my experience it heavily depends on the problem space. It's obviously very good at generating good code for already solved problems.Where it struggles is anything novel. Anything that isn't simple a known solution being referenced and reinterpreted for your requirements/codebase.When you need to write code for new things it tends to be questionable.I think it's inevitable but I have no idea how far away that is. Could be tomorrow or after our lifetime.
>>108816518>Is letting people spend $6,000 on tokens for $40 a sustainable business model?It's an investment. You make it cheap until it becomes absolutely indispensable, and once it is you can raise prices non stop. Look at google maps (used to be free, then required a login to use the API, then it's "free" for the first 1000 api calls and then starts billing...), or even just how much Netflix prices rose over the years.
>>108816518A phone plan can also give you the equivalent of thousands of dollars worth* of minutes for tens of dollars a month.* assuming the most predatory high-time-preference payday advance bullshit pricing schemeWhoever came up with that dashboard really earned their pay.
>>108817824Well they are by the definition of the word.
>>108817875the problem is that models are commoditiesyou can replace opus with gpt or kimi with a clickas soon as they stop subsidizing subscriptions, users will switch
