EFS Cache Now Auto‑evicts to S3, Cutting Costs
Actually this is sick. Files fall out of EFS cache back to S3 prices after configurable expiry. Previously mounting S3 directly was buggy af, and EFS cost-prohibitive.
Fake Post Offers Real Sonnet
I guess this is a fake? But they looked up real eval scores for sonnet/opus, making this a useful reference post 😆
Automated Tax Filing Finally Moves From Dream to Reality
Can't believe we're here. "Do my taxes" has been the impossible north star of agent automation for years.
Agent Systems Will Become the Software Backbone
The big switch on horizon is when agent systems *become* the software backbone, instead of just generating it.
AI Agents Cut Millions of CSS Hours to Weeks
Collectively hundreds of millions of hours spent on this class of CSS problems, solved in a few weeks with AI agents.
One‑Shot Prompt Powers Pi Day Computer Talk
Talked Computer at @PrimeIntellect Pi Day - here’s my slides, and the Computer prompt I used to 1-shot them ✨ https://t.co/QIXfNOwTOp https://t.co/tu8oLSuTwr
Agent Systems Are Just a Model Revision Away
I pretty much zone out when people pitch agent or memory or orchestration systems. These are all a model rev away from built-in.
AI Will Tackle Complex Group Dynamics in Business, Politics, Economy
After math, code, assistant competence AI will take on complex group dynamic problems like firms, politics, economy.
Big Models Still Struggle to Prompt Themselves
Why aren’t LLMs better at prompting? Big models should be able one-shot themselves.
AI Coding Is Here to Stay Forever
People haven't internalized that AI isn't going away. E.g. coding doesn't go back, it's AI coding from now to eternity.
Grok 4.1: The Underrated Go‑To Fast Model
Grok 4.1 is underrated as a goto fast model.
Alignment Comes From Prompting, Not Model Magic
Crazy that alignment is a prompt not any magic model hoodoo.
AI Rewrites Journalism in Inverted Pyramid, Saving Hours
I save hours by asking AI to rewrite independent journalist articles in inverted pyramid.
Comet Dominates All Browsers on Tough Test Queries
We test every browser & computer use model/api/app released on our difficult test queries and @comet beats all of them. It’s not even close. We have a lot to improve but damn people be shipping slop.
Copy Proven Methods to Amplify Compute Efficiency
The most Bitter Lesson-pilled strategy is to steal what works, thereby compounding overall compute 😱
