i tagged theo on x with this:<\/p>\n
> I think @theo was right in one of his old videos, we are coming to the end of the golden age of AI usage. @AnthropicAI cutting the boost and @OpenAI now reintegrating the 5 hour limit. adding chinese plans + gemini just to keep my finances in line!<\/p>\n
his reply:<\/p>\n
> Yes and no. It’s nuanced. I think this video covers it better<\/p>\n
the video: [Theo’s “Prime is (mostly) right about AI”](https:\/\/www.youtube.com\/watch?v=VDPMXSAxiWk). worth the watch, told him so in the thread:<\/p>\n
> totaly get the compute and the actual constraint, but with OpenAi annoucing reducing tokens on terra and luna and also there own chips coming out, felt a bit harsh to push the 5 hour limit right now. (and thanks for the video, had missed that one \ud83d\ude42 made my evening)<\/p>\n
then i went off on a tangent about google:<\/p>\n
> and also, the microsoft making money part… WTF is google doing !!
\n>
\n> they had so much to get the first and they are so behind. hoping that they will jump back masivly as they do have the best visual models (and I still goto them for anything graphic or visual cleanup)<\/p>\n
so what does it all add up to, for me as a daily user? a few things.<\/p>\n
the compute story is real. OpenAI dropped their own custom chip with Broadcom in june, gpt-5.6 luna and terra got 80% and 20% price cuts in july, serving efficiency finally catching up. microsoft is still on the hook as OpenAI’s biggest investor and Azure host, so OpenAI has runway to ride out the squeeze. none of it is wrong.<\/p>\n
the timing on the 5-hour limit is weird tho. if you know your cost basis is about to drop 40% in six months because your own chips are landing, why squeeze users right now? “we’re rationing because compute is tight” doesn’t survive “we just announced chips that fix compute” three months later. felt harsh, as i said in the thread.<\/p>\n
the google thing is what bugs me most. they had everything, first mover on transformers, deepmind, tpus that are genuinely better at training than buying nvidia. they shipped antigravity with gemini + anthropic models baked in, smart move but also an admission they need help on the model side. for visual work they still win, gemini’s image gen is what i reach for whenever i need graphic cleanup, nothing else gets close. so my routing-around ends up like this, google for visual, chinese ai plans for bulk text, frontier models when i actually need them.<\/p>\n
and yeah, theo’s “yes and no” boils down to this. yes, the golden age of cheap ai is ending, Anthropic cut the boost, OpenAI reinstituted the 5-hour limit. no, this doesn’t mean ai is worse, it’s getting better, serving economics catching up, OpenAI has chips coming, Google still wins visual. and the thing you do is just route around. multi-model isn’t a clever strategy, it’s just where the tools actually are.<\/p>\n
original exchange: https:\/\/x.com\/theo\/status\/2094523070472192301<\/p>\n","protected":false},"excerpt":{"rendered":"
i tagged theo on x with this: > I think @theo was right in one of his old videos, we are coming to the end of the golden age of AI usage. @AnthropicAI cutting the boost and @OpenAI now reintegrating the 5 hour limit. adding chinese plans + gemini just to keep my finances in…<\/p>\n","protected":false},"author":1,"featured_media":50,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_kad_post_transparent":"","_kad_post_title":"","_kad_post_layout":"","_kad_post_sidebar_id":"","_kad_post_content_style":"","_kad_post_vertical_padding":"","_kad_post_feature":"","_kad_post_feature_position":"","_kad_post_header":false,"_kad_post_footer":false,"_kad_post_classname":"","footnotes":""},"categories":[3],"tags":[],"class_list":["post-42","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai"],"yoast_head":"\n