Artificial Intelligence thread

Phead128

Phead128

Major
Staff member
Moderator - World Affairs
If you are a decent senior dev who uses AI as assistant (human in the loop) instead of relying on vibe coding one-shot apps, the leading Chinese open source models are more than good enough as a daily driver due to immense token cost savings. You can always use frontier models for initial planning or stubborn debug, but for bulk of coding, the Chinese open source models is as competitive for immense cost savings.

The ones who disagree are probably vibe coders who need Fable to one-shot an app from scratch without errors, or one-shot a huge codebase factor which is rare. Or unless your company is paying for unlimited tokens, this doesn't apply. Ask Opus 4.8 for what the temperature is today in a loop every 15 mins, but even companies like Uber are reigning in LLM budgets.
 
I

iewgnem

Captain
Registered Member
Phead128 said:
If you are a decent senior dev who uses AI as assistant (human in the loop) instead of relying on vibe coding one-shot apps, the leading Chinese open source models are more than good enough as a daily driver due to immense token cost savings. You can always use frontier models for initial planning or stubborn debug, but for bulk of coding, the Chinese open source models is as competitive for immense cost savings.

The ones who disagree are probably vibe coders who need Fable to one-shot an app from scratch without errors, or one-shot a huge codebase factor which is rare. Or unless your company is paying for unlimited tokens, this doesn't apply. Ask Opus 4.8 for what the temperature is today in a loop every 15 mins, but even companies like Uber are reigning in LLM budgets.
Click to expand...
It's not about cost savings at this point, people and companies don't have infinite money, you literally can't do a lot of things if you can't use them above a certain level, which means it's not about being good enough or not, it's about if you can use them at all.

As an example with Kimi I can have 4-5 streams going at the same time trying out different ideas and do experiments in the background, and I can just abandon those ideas if I don't like it without thinking twice about how much money I spent. I can tell it to write entire applications just to debug one thing then throw it away, I can have it run build-monitor-debug-rebuild loop with real hardware overnight with massive logs and authorization to modify debug software as needed. You simply can't do any of that that if you're capped.
 
I

iewgnem

Captain
Registered Member
https://twitter.com/i/web/status/2065942612293365948
You have to feel sorry for Dario for having to explain AI models to people like Lutnick and Bessent.
If the whole thing isn't staged, their whole "we're so powerful we're dangerous" marketing stunt has got to be one of the biggest self own in tech history.
It's one thing to have a bad marketing strategy, it's another to have a marketing strategy that get you banned by people who believed your marketing.
 
9dashline

9dashline

Major
Registered Member
iewgnem said:
Also, just a reminder, lol
https://twitter.com/i/web/status/2065728537122775327

I don't think people are appreciating just how much Anthropic f*ed up with their marketing stunt, in a country that's notorious for it's education system and government official competency.

If this spreads outside of Anthropic it will literally insta kill the entire US AI industry.
Click to expand...
I'm calling it now this moment will be forever remembered as the downfall that lead to America permanently losing the AI race to China
 
9dashline

9dashline

Major
Registered Member
Phead128 said:
If you are a decent senior dev who uses AI as assistant (human in the loop) instead of relying on vibe coding one-shot apps, the leading Chinese open source models are more than good enough as a daily driver due to immense token cost savings. You can always use frontier models for initial planning or stubborn debug, but for bulk of coding, the Chinese open source models is as competitive for immense cost savings.

The ones who disagree are probably vibe coders who need Fable to one-shot an app from scratch without errors, or one-shot a huge codebase factor which is rare. Or unless your company is paying for unlimited tokens, this doesn't apply. Ask Opus 4.8 for what the temperature is today in a loop every 15 mins, but even companies like Uber are reigning in LLM budgets.
Click to expand...
Yup Boris from Anthropic didn't do them any favors by telling everyone he didn't even need to prompt anymore he just can write a loop for everything... And let the AI figure it out....
 
M

Matcher6130

New Member
Registered Member
iewgnem said:
Also, just a reminder, lol
https://twitter.com/i/web/status/2065728537122775327

I don't think people are appreciating just how much Anthropic f*ed up with their marketing stunt, in a country that's notorious for it's education system and government official competency.

If this spreads outside of Anthropic it will literally insta kill the entire US AI industry.
Click to expand...
Lets be honest, it's all but confirmed the "real" reason for this stunt is cover for the fact Anthropic is running critically low on compute capacity since at least April, likely much earlier.

Please, Log in or Register to view URLs content!
- WSJ

Please, Log in or Register to view URLs content!


Please, Log in or Register to view URLs content!


Quick aside, I hate WSJ. It's a Rupert Murdoch trash rag, but Murdoch also hates Tech Bros so I trust his willingness to take swipes over his willingness to lie.
 
Last edited:
9dashline

9dashline

Major
Registered Member
Matcher6130 said:
Lets be honest, it's all but confirmed the "real" reason for this stunt is cover for the fact Anthropic is running critically low on compute capacity since at least April, likely much earlier.
Quick aside, I hate WSJ. It's a Rupert Murdoch trash rag, but Murdoch also hates Tech Bros so I trust his willingness to take swipes over his willingness to lie.
Click to expand...
Initially that's why they did the Glasswing thing but then they got extra capacity by cutting a deal with SpaceX to use their collussus GPU clusters since no one was using Grok and Elon Musk had a massive datacenter idling

Fable was released to public to hype up Anthropic being top and surpassing Chatgpt 5.5 right as all three, Spacex, OpenAI and Anthropic race each other to IPO

It's not coincidence that it was the Amazon CEO (Amazon rocket blew up recently) snitching on Anthropic (Anthropic used to be exclusively AWS customer before going to/with SpaceX) that caused Fable to be pulled on the same day that SpaceX itself went IPO...

Reports were that OpenAI was going to have to drop it's prices to compete with Anthropic and it was going to be an unsustainable race to the bottom

But the issue is GLM 5.2 is Opus 4.8 tier, and Kimi K3 will be Fable/Mythos tier... So the US gov essentially just killed the entire US AI industry.... OpenAI and Anthropic can't compete on price, and now they also cannot compete on the high end since it's gated behind a US passport.
 
I

iewgnem

Captain
Registered Member
9dashline said:
Initially that's why they did the Glasswing thing but then they got extra capacity by cutting a deal with SpaceX to use their collussus GPU clusters since no one was using Grok and Elon Musk had a massive datacenter idling

Fable was released to public to hype up Anthropic being top and surpassing Chatgpt 5.5 right as all three, Spacex, OpenAI and Anthropic race each other to IPO

It's not coincidence that it was the Amazon CEO (Amazon rocket blew up recently) snitching on Anthropic (Anthropic used to be exclusively AWS customer before going to/with SpaceX) that caused Fable to be pulled on the same day that SpaceX itself went IPO...

Reports were that OpenAI was going to have to drop it's prices to compete with Anthropic and it was going to be an unsustainable race to the bottom

But the issue is GLM 5.2 is Opus 4.8 tier, and Kimi K3 will be Fable/Mythos tier... So the US gov essentially just killed the entire US AI industry.... OpenAI and Anthropic can't compete on price, and now they also cannot compete on the high end since it's gated behind a US passport.
Click to expand...
Now they're saying Fable is just GPT5.5 tier lol
https://twitter.com/i/web/status/2065835532018499808

I suspect USG won't read this as we should relax control on Anthropic, I suspect they'll do the most obvious thing
 
I

iewgnem

Captain
Registered Member
Matcher6130 said:
Lets be honest, it's all but confirmed the "real" reason for this stunt is cover for the fact Anthropic is running critically low on compute capacity since at least April, likely much earlier.

Please, Log in or Register to view URLs content!
- WSJ

Please, Log in or Register to view URLs content!


Please, Log in or Register to view URLs content!


Quick aside, I hate WSJ. It's a Rupert Murdoch trash rag, but Murdoch also hates Tech Bros so I trust his willingness to take swipes over his willingness to lie.
Click to expand...
Yeah it could all be a stunt, but it doesn't really matter: either they can't afford to run anything above Opus 4.8, or they're not allowed to run anything above Opus 4.8, in both scenarios they're f*ed, because China can both afford to run it and are giving them out for free to anyone in the world.

If it is a compute problem and US models are being capped due to compute while Chinese models aren't, that would transcend irony and enter poetic territory
 
You must log in or register to reply here.
Top