There is a lot of rare knowledge here.
Anyway in two weeks — two months it’ll be outdated.
And while I do business work it’s not the point to optimize costs right now. We will optimize when there will be what to optimize.
Got some hard data - I was wrong.
Had Datacurve run the numbers for "tokens used by pass/fail" for DeepSWE.
Bad models use way more tokens in fail cases, but SOTA models are much closer. GPT 5.5 used ~7% MORE tokens on correct answers!
Drop what you're doing. It happened.
Opus 4.8 is out and it smashes every benchmark
Clearly the extra compute from Papa Elon is paying off with this one
This is what you need to do immediately if you want to escape the permanent underclass:
• Switch every task you were previously doing with Opus 4.7 to 4.8. They are the same price
• If you are on the $200 plan, us /fast mode. You'll get 2.5x the speed and it's 3x cheaper than fast mode was previously
• For super complex tasks use the NEW Dynamic Workflow mode. Sends out thousands of subagents to completely tasks. Tasks that previously took months now can be completed in a sitting
• Use new 'Ultracode' mode. This sets the thinking level to xhigh but also let's Claude decide when to use Dynamic Workflows
When new tech drops, you have to take advantage of it. That's the only way to win
Put your phone on Do Not Disturb and get to it
@pmitu No.
What kills small creators is aggregators with 4 million followers downloading-and-reuploading 600 videos per day—and flooding Timeline with slop that went viral 7 years ago.
But we will kill them.
Opus max via Droid cli is ideal for frontend final validator. It understands more human like ui / UX issues.
Maybe some code revirw. Maybe marketing docs and research.
That’s it.
Plugged it inside my orchestration skill in codex. Works like a charm. 130 iteration
cycles and counting. Polishing front end better than supposed to.
Never hit limit for opus or codex. Balanced.
Sorry, but it is garbage. Idk why most of core UX features seriously degraded over last 5-10y.
It was limited before but Swiss knife type of limited and worked mostly ideally.
As a UX pilled person I’m sad. Mac OS feel more and more like worst MS UI like Word. Not for newbies not for middle not for pro, just a mess of a flow of ideas.
And it’s not a bootstrap anymore.
Actually maybe this is the reason.
They can’t receive market signal anymore because of sort of dominating market position, because of record amount of cash on their account, because of previously shine brand and founders story.
Yep. Made a skill base on it for orchestration with 3 roles.
using codex models and also droid cli to agent to be able to consult with the opus.
Works amazing. 200 iteration cycles on full auto.
One more agency task automated to 80%
And like 10-20x more faster.
Can’t imagine what roles will exist in 2027.
Today we are starting to roll out the biggest upgrade to the Google Search box in over 25 years — now completely reimagined with AI, along with Gemini 3.5 Flash as the new default model for AI mode users globally!
Welcome to Gemini 3.5 Flash, our most powerful model to date. It pushes the frontier of intelligence, speed, and cost putting 3.5 Flash in a class of its own.
We spent the last 6 months making sure Flash is great for real world use cases. It's available everywhere now!
Not at the moment.
Though I can share the concept.
You can ask agent to record a video during the use of browser for example.
Then you need video reasoning model.
That's it.
Orchestrate, tune prompt, integrate to pipeline.
I spent couple months or more tweaking it and going through the models.
Curious which VLM Peter use.
Sorry, cant find better channel to reach support because seems like CF account is blocked from everything for some reason.
The limits on acc is lower than free (20 domain redirects instead of 10,000 i.e.), cant add members, cant have API key. Acc is 6-7 yo and there is few errors with no names and few errors stating that email isn't verified.
And there is no way to request verification.
And when I submit a ticket to support I cant even see it and when I click to my tickets I see only this page and that's it.
Cant ask ai because it cannot issue read only API token and crashes with no name error. But if I try to create in another part of UI i see error about email verifications.
Looks like edge case and that ac is fully out of coverage.
Please, we need to fix some SEO redirects ASAP.
Spent all weekend trying to figure out CF issues.
I just learned that the "data centers are using our water!" bullshit started because of a book called Empire of AI by Karen Hao in which she totally fucks up the math when determining how much water they use, an error she later acknowledged