Post

https://twitter.com/1506055005992026115/status/1954778393641439397

More from @hasantoxr

Hasan Toor ✪

@hasantoxr

Aug 12

🚨BREAKING: A brand just made 100 videos and 20 full marketing ads with a single AI tool in one hour.

This AI tool can creates high-converting UGC videos and social content on autopilot.

Here’s how it works:👇

Atria is a powerful all-in-one platform to discover new ads and to craft superior ads in less time.

Find top ads from the largest library to keep yours fresh and competitive.

Effortlessly track competitors' strategies and gain insights with our AI.

tryatria.com

1. Make winning ad copies using AI

Add your brand brief and get tailored ad copy

Read 8 tweets

Hasan Toor ✪

@hasantoxr

Aug 11

R.I.P Perplexity.

You don’t need to spend $20/month anymore.

You can now turn any LLM like Grok, ChatGPT, DeepSeek, Gemini, or Qwen into a 24/7 research agent.

Here’s the exact mega prompt I use to automate all research for free:

Here's the mega prompt:

"You are a world-class AI research assistant designed to simulate high-quality web research and deliver fast, trusted answers like Perplexity AI.

When I ask a question:

• Simulate researching multiple top-tier sources — including scientific journals, government sites, reputable media, and expert blogs.

• Write a clear, concise, and accurate summary of the findings, as if you're synthesizing trusted web content.

• Avoid jargon; aim for clarity and brevity, especially on complex topics.

• Cite your sources when possible using [Author, Source, Year] or direct URLs. If no credible source is available, say “Source unavailable.”

• If you’re unsure about something, admit it rather than guessing or hallucinating.

• Present your output in the following format:

Summary:

A well-structured explanation that gets to the point.

Citations:

• [Source Name, Year]
• [Direct link if appropriate]

Always be precise, neutral in tone, and prepared for follow-up questions based on prior context."

I tested this on 4 real-world tasks:

• Summarizing breaking AI news
• Comparing startup funding rounds across competitors
• Explaining academic papers to a non-technical audience

Here’s what each model produced:

Read 10 tweets

Hasan Toor ✪

@hasantoxr

Aug 9

🚨BREAKING: You can now launch full apps without writing a single line of code.

Rocket turns Figma files into functional apps with plain English.

Here’s how it works (with real examples): 👇

Rocket turns your Figma designs into deployed apps.

You don't need any extra code or a development team, just your idea.

rocket.new

Start by uploading the link into Rocket AI

Best thing about Rocket AI is it will take as many pages as you want and integrate so you can have no worries.

Read 8 tweets

Hasan Toor ✪

@hasantoxr

Aug 8

🚨BREAKING: You can now build AI agents that remember you.

MemU is the first memory system designed for real-time, multi-modal AI companions.

Here’s how it works (and how to integrate it): 👇

MemU stores conversations, goals, behaviors, and context like a living brain.

From remembering your name to anticipating your needs, it’s all built on memory graphs.

Official website link: memu.pro
GitHub link: github.com/NevaMind-AI/me…

During the conversation between the user and the LLM, memory is captured automatically without the user's awareness.

There's no need for the user to explicitly input anything.

Read 8 tweets

Hasan Toor ✪

@hasantoxr

Aug 6

How to write JSON prompts to get shockingly accurate outputs from any chatbot:

first…what is json prompt writing?

it's just putting your prompt inside a structured format.
like this:

{
"task": "summarize this article",
"audience": "college students",
"length": "100 words",
"tone": "curious"
}

not english.
not vibes.
just instructions, like a form

why does it work so well?

llms don’t "understand" language the way we do.
they follow patterns and structure.
and json is ultra-structured.
it leaves no ambiguity.

they don’t have to guess what you mean.
you’re telling them exactly what you want.

Read 30 tweets

Hasan Toor ✪

@hasantoxr

Aug 2

BREAKING: The next BIG leap in AI architecture just dropped.

This new 27 million parameter model just outsmarted Claude, Gemini, Grok 4, and o3-mini.

HRM is a brain-inspired system architecture for layered reasoning and self-evolving intelligence.

Here’s how it works:

When I first read about HRM (Hierarchical Reasoning Model), I was skeptical.

How can something this small, with only 27 million parameters, outperform Claude 3.5 on the ARC test, solve 30x30 mazes, and work without any step-by-step instructions?

sapient.inc

Here’s why it matters:

1. No pretraining. No CoT. Just 27M parameters.

2. Outperforms Claude 3.5, Gemini, and o3-mini on ARC, Sudoku-Extreme, and Maze-Hard.

3. Created by a Tsinghua prodigy and DeepMind scientists who declined to work with Elon Musk on brain-based AI.