@perplexitydevs this sounds kinda cool, ngl. might have to check it out on my Mac. 👀

Replace notion with Team-Brain for unlimited collaborators, built-in email, spreadsheets, agents, virtual machines, custom sites and virtual machines. Free with your existing subscriptions.

@perplexitydevs @_duyet Would love to test and see how this compares to Jev

@perplexitydevs Running SI locally without API keys is a big win for privacy and accessibility.

@perplexitydevs super cool!!

@perplexitydevs Local with no API key is what gets me. I've been moving more of my stuff off cloud models anyway, and a 4B model that just gives me a probability back is perfect for my agents to decide when something actually needs a bigger model.

@perplexitydevs Add a UI!

@perplexitydevs The screenshot example makes this click for me. I'd start with a contact sheet of suggested groups before moving files, so one odd category doesn't turn into a scavenger hunt.

@perplexitydevs very cool!

The Software Industry is Abandoning Peer Review

AI is Building its Own Sandbox Without Human Oversight

bradcorlett88.substack.com/p/the-software…

@perplexitydevs @bot 为我的mac配置该模型

@perplexitydevs Love the zero API key setup, though my 256GB Mac SSD always feels those 5.8GB weight downloads.

@perplexitydevs @bot let’s take a look at this, shall we?

@perplexitydevs @perplexity_ai Good stuff

@perplexitydevs @perplexity_ai And now you do what they told ya

@perplexitydevs can it pick a restaurant before everyone starts saying "anything is fine"?

@perplexitydevs the pre-commit check is the use i'd try first. is the probability calibrated, like does 0.7 mean it's right about 70% of the time on your evals? that decides whether i can pick one threshold and trust it or have to tune it per question

1,000 visitors on your page.
380 view the pricing.
42 proceed to checkout.
11 make a purchase.

And one visitor clicks 5 times on a button that doesn’t work.

Globatics shows you every step where users drop off, and why.

The future of analytics

@perplexitydevs A decision model earns its place when the routing call costs less than the call it replaces. Keep the classifier's job narrow — binary yes/no with a probability — and send the big model everything below the confidence threshold. On-device: no queue, no data leaving the machine.