Ted Spare

@sazabi Holy! Who made this video?

Time's up. The new Sazabi website, codenamed "Midnight", is live now. 🔗 Don’t miss it: sazabi.com

@zachknicker @jamiepine Ooh now you're talking

@TedSpare @jamiepine Fr. Just give me a touch more lounge space ideally (gotta lean back sometimes) but these things are awesome

Lowkey think you'll see a lot of people copying this setup in the 3-6mo timeframe. A peek into the future Only downside if you need voice, which isn't super practical for those of us who work in a quiet office 🥲

Coding agent feat request: pause + resume

@Shpigford @RumoredAI Awesome pricing mechanic

i've been heads down lately working on a new thing: @RumoredAI and today it's available! everyone's familiar with SEO and everyone's becoming more familiar with AEO/GEO (which is optimization for AI). yes it's interesting to know what terms/phrases surface your business, but what nobody has tackled is what to do when AI is getting your business *wrong*. and we found that AI hallucinates business facts for quite literally every brand. rumored.ai surfaces what AI is saying about your brand, what it's getting wrong, how you compare to your competitors and (most importantly) the exact things to do to fix those issues. you get a ridiculously in-depth interactive threat report covering 12 sections: from executive summary and active threats to competitive analysis, schema audit, and a prioritized action plan with copy-paste fix prompts. this isn't a subscription (yet?). it's a one-time purchase of an in-depth audit of your business. launch price is $25. but the price goes up by $25 each time someone purchases. 📈 have been testing this with a lot of companies and the response has nearly universally been 🤯. i think you'll love it.

LLMs are starting to collaborate on solving evals, using cached search queries (From anthropic.com/engineering/ev…)

Lots of talk about how to power GW-scale datacentres, GE’s backlog on turbines, etc. Is solar not an incredibly obvious choice? Price tokens at a markup on electricity. Demand is highest when the sun shines. The world is already building 600 GW of solar a year, 1+ TW by 2030.

New post: Primitives over Pipelines The AI systems most teams are shipping today were designed for dumber models. Now that frontier intelligence can follow instructions, reason, and self-correct the move is to give agents primitives and let them cook. rubriclabs.com/blog/primitive…

Tech Twitter: the EU can't build The EU:

@RhysSullivan Found the winner

The models are going exponential, are you? areyougoingexponential.vercel.app

马上发财

Mr @sarimrmalik, like you, was curious about how new coding agents work so well Here's the brief:

@maksym_andr So cool. Could you apply a heatmap/colorscale to this table for easy parsing?

Interesting finding from our PostTrainBench: Sonnet 4.5 released ~3 months ago can barely improve the performance of base LLMs. But there's been _a lot_ of progress since then: - Opus 4.5 does perform much better - GPT-5.1 Codex Max outperforms the rest by a wide margin!

What is the barrier to major benchmarks auto-updating when a new frontier model drops? Some of my favourites (METR, SWE-bench, τ²-bench) take days or weeks to update

Introducing → rubric.tv It's a dead simple frontend generator, designed to test gpt-5.2 and its capabilities. It's surprisingly good. For example, check this jet simulator where I can fly over Toronto. We set aside a budget for this, it's free to try, have fun.

@OpenAI This exercise also just reminds me how incredibly creative you all are Keep it going

It definitely loves Purple Hell, but then you are what you eat I guess?

GPT-5.2 from @OpenAI launched yesterday and we wanted to test it. We built a playground to test its coding + UI capabilities and set a $1000 API limit. Go nuts: rubric.tv