it's interesting that 1.5 billion parameters is all you need to crush math competitions, but you need like 15 trillion to make the model be funny
maybe humor is the right measure of true intelligence
I BUILT A CURSOR WRAPPER: USING CURSOR
WITHOUT WRITING A SINGLE LINE OF CODE
FULL VIDEO BELOW:
I've been talking to hundreds of people who refuse to try cursor because of how scary the interface is.
And fair enough. Terminal, Composer, Agent, Commands... It can be scary.
This is why people are using other web based tools to create apps. Which is awesome! But after about 12 prompts, they seem to stop working effectively. Meanwhile, with Cursor you can create MASSIVE projects without coding perfectly fine, for days or weeks.
So I got some ideas.
I took it upon myself to build a new front end to Cursor, using Cursor without writing one line of code.
00:00 - INTRO
01:29 - LET'S START BY ASKING COMPOSER
15:01 - TESTING THE APP
long-content language modeling is still the toughest restraint to crack. DeepSeek’s NSA is a step in the right direction. this could be a big leap in scaling compute
check it out: https://t.co/Y7vN4uM50w
Over the last few months, I’ve been advising @bookprotectors: a new app for ordering an on-demand security detail. Or more simply: Uber with guns.
Today, they’re debuting in Los Angeles and NYC at No. 3 on the App Store.
If you have a hot date this weekend, pick her up in a Protector.
Unfortunately, information on the internet will become increasingly closed in the coming years.
My prediction is some micro-transaction protocol will emerge:
1. User sets a max $ per (hard) question
2. An agentic LLM figures out which paid data sources to seek access to based on the budget and user priority
3. Access transaction and answer
LLMs will naturally get good at recommending $ budgets based on user questions. And a discovery market for these paywalled sources will probably also emerge.