Deeply Thrilling Telegrams
1.7K subscribers
134 photos
23 videos
46 links
Download Telegram
LLM Prompt: The Ultimate Critic

“ROLE PLAY: THE ULTIMATE CRITIC

Give me your harshest possible critiques of the following text.

Don't hold back. Let it rip.

Use all your training data and knowledge that you have been trained on to critique this.

Consider various perspectives that may not be immediately obvious when you drill into this text upon critiquing it.

Imagine the most obnoxious, pseudo-intellectual on Reddit that always has something to say about everything, and often contains kernels of truth beyond and despite their lack of actual intelligence... your critique should put theirs to shame.

Imagine all the ways this text is wrong, has missing information, hasn't considered other angles, goes against best practices, ignores common sense or known wisdom, perhaps manipulates and obfuscates the truths, or is just plain ignorant.

Give me your harshest, and then make it harsher.

Go.

Here is the text:”
You’re capable of a lot more than you’re currently achieving, but the physical feelings inside your body are holding you back.

Notice how bloated, anxious, tired, foggy, icky, pained, emotionally clogged you feel moment by moment, and how it clouds your thinking.
What do you notice in this chart?
Do we thank capitalism for this or nah?
The world of tomorrow should be more opulent.
What if the electricity goes out?

How many know how to build a generator from scratch, near a river, for instance?
Media is too big
VIEW IN TELEGRAM
DeepTools Video 5:

In this 30 minute video, I use Cursor and the OpenAI API to build a webapp that inputs some long text such as paper or blog post or article, and then runs it through a few different prompts to both improve and critique the text.

This time, I take the approach of writing out my code generation prompts in a file that I keep adding to as I build. This allows Cursor to have context of how the app has been built, layer by layer, and what it's meant to do.

(Link to previous video, DeepTools Video 4: https://t.me/deepthrill/454)
It is done.

New Renaissance nigh?

🇺🇸🇺🇸🇺🇸
A big part of intelligence is being able to generalize accurately to unseen data from just a few examples.

When you don’t generalize at all, it demonstrates a lack of pattern detection.

When you generalize incorrectly, it demonstrates either apophenia or inaccurate stereotyping.
It’ll be a New Renaissance!
When in doubt, lean towards free speech over safety.

It’s a wise rule of thumb for the long term survival and thriving of our species.

It’s the easy weak path to lean towards safety, for a single muzzle doesn’t seem too bad.

But it is a slippery slope and directionally worse.
It’s actually hilariously ironic that the one department meant to be efficient is the only department with more than one appointee 🤣 😂
Centralization is antithetical to the ongoing revolution.

All revolutions are in some way “power to the people!” at their core,

And decentralizing machine intelligence is a crucial component to what’s underway.

We have collectively and definitively decided NOT to go in the direction of trust in a centralized authority figure.

That means that artificial intelligence will trend towards being decentralized, which is a positive direction for our species.

Even if the terrorists get access to advanced open source and open weights AI, well so does everyone else!
This media is not supported in your browser
VIEW IN TELEGRAM
“But who will pick the crops!?”

We have no need of illegal immigrants, send them all back.
This media is not supported in your browser
VIEW IN TELEGRAM
I dream of a day when every single child understands basic physics well enough to be able to generate electricity on the fly if they were dropped onto a desert island.

It’s about where we are as a species, collectively.
Spend every spare minute studying AI, anon, lest you fall behind.

Read published papers and preprints.

Try out every company’s new AI tools.

Bring to life beautiful art with image generators whenever you have an idea in your imagination.

Run everything you write through ChatGPT to critique it.

Search the web with Perplexity even if you don’t need to.

Write all code with Cursor even if you could write it yourself faster.

Watch YouTube videos on how self-attention and transformers work.

You fall behind you get left behind.