Anthropic to reach 100% global GDP in 21 months
>>108790352some chance of that when it's not even a stock with an IPO
every few weeks i ask claude and codex to rebuild half life and they refuse to do itif they can’t build a silly little game from the 90’s, how will they do all this other stuff you guys talk about?
>>108790375you're using it wrong
>>108790375How well does it do?
>>108790352>log scalelol
>>108790362>196 resultsCringe af
>>108790352why stop there? 420% of GDP in 24 months
>>108790352>be me>use claude opus to help me resort some isometric cubes so they render in the correct order>it adds a memory leakBravo Dario!
>>108790362Snailcat a cute.
>>108790352Fucking kek. You know they made this graph in complete seriousness too. God should nuke SF
HOLY FUCK AI WON.100% IS AI. I KNEEL
>>108790352Why would people assume Anthropic would be ahead in several years? I still remember when they hit a slump after Sonnet 3.5 during trying to get Opus/Sonnet 4 out and being behind everyone in reasoning. They released right after R1 from Deepseek even. You can't say anything about who is ahead vs behind in the race without knowing what is going on and what all the labs have. I remember still Google coming in from behind and beating everyone with Gemini 2.5 and people having to accept the reality there was a third Western frontier lab and Meta falling behind. Come back in a year or two and we'll see who is still there and on top.
>>108790362snailoids btfowhat will snailtards say when anthropic at 200%? they will lie down and just go to sleep like gay snails lol fucking idiots
>>108791095its so over for small peepee luddy fuddy squishy wishy snailz
>>108790785>slopusshould had used gpt
>>108791211should've'd*
/g/ is so shit with poos now trying to force their memes
>>108790352OpenAI is sweating
>>108790484Nigga, you're talking to a $current_year year equivalent of pic related. Just filter it.
>>108790469they just need to 1000+x their company bro, couldn't be easier for a company that gets 90% of its income from the video card-big tech-investment fund-AI hugbox funding
>>108790352>dat diagrammega kek
>>108791079I am unfortunately an AI engineer. It means I orchestrate api requests to sloppa endpoints from anthropic, openai, and google (although often through intermediaries like azure or aws bedrock).We use it for automated document review workflows and similar document-oriented automation.We use it in various capacities for these tasks, for example generating code snippets that run in sandboxes to execute functions, or to control a system where they operate in as part of an addin.Basically, in non-thinking mode, universally (i.e. any modality, any task), we found that 4o > 4.1 > literal garbage > opus 4.6 > opus 4.5 > haiku 4.6 > haiku 4.5 > random guessing > gemini-2.0-flash > gemini 3.0-flash (pre-cancelation) > gemini-2.5-flash > gpt-5 series (roughly whatever order you like, they all suck balls) > gemini 3.0-flash (post-cancelation) > gemini 3.1-flash (pre-GA).For document extraction (i.e. "read this document, extract data according to this json schema"), gemini-2.0-flash > gemini-3.0-flash (pre) > gemini-2.5-flash > gemini-3.0-flash (post) > gemini-3.1 > *. I'm not even listing the others because they literally can't do it even on small documents. They fuck up so terribly it's insane and I never thought there could be such a gap.We never test opus because costs are prohibitive even though we're b2b and pass costs on to our clients. Looking at like $10-20 to process one document when it's more like $0.1-$1 for other models.For thinking, though, anthropic models trump everything modern, but o1 > o3 > *.Outside our internal evaluation suites, I found that sonnet 4.7 works significantly worse than 4.6, and that 4.6 in chat mode (i.e. with all the orchestration that anthropic built, not the model quality itself) works heads and shoulders above anything else right now.Note that those evaluations are not merely a reflection of our internal ability to optimize for one model or another either: cs teams at providers concurred and other clients confirmed
>>108792895Slight correction: whenever I say opus it should say sonnet. We don't use opus as mentioned because of costs. In fact we avoid sonnet because outside of anthropic's built-in orchestration, the models themselves are actually pretty shitty. Not as bad as gemini but at these prices we would rather bash gemini into submission. Alternatively, it would have to be clearly better than other models and then we would use it at any price because investors will foot the bill instead of clients. But that's just not the reality of it.
>>108790783Beyond cringed and look the bosnian that keeps reply to yourself nooticed one of his spam got deleted
