Y'all heard about ChatGPT yet? AI instantly generates question answers, entire essays etc.

Sunalmighty

Superstar
Joined
Mar 1, 2013
Messages
7,908
Reputation
1,675
Daps
19,016
Reppin
Oakland, Ca
exactly, people think this is just a text or image creating technology. the goal is to build a an intelligence that surpasses that of human beings. if you make some reasonable assumptions about what that could mean, well it'll be tasked creating solutions to any problem without ever needing to taking a break while have almost the entirety of human knowledge available to it. build new tools, new structures, rebuild itself. anything is possible and i'm not even saying it has to be sentient to accomplish all that. any and everything could change if tech continues to advance.
imagine that.

everything you typed is a horror movie. soon enough we'll learn that going to the library and reading books instead of search engines n shyt was the safest way to do things.
 

///Vega+++

All Star
Joined
Jun 7, 2012
Messages
4,704
Reputation
-221
Daps
7,909
For those who have tested both, how does Google's Gemini perform compared to Chat GPT-4 ?
I've only used 3.5 and I can already see Gemini gives more precise answers and is less error prone image generation controversy aside
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
55,098
Reputation
8,185
Daps
155,715
imagine that.

everything you typed is a horror movie
:ld:

it's already happening, AI is used heavily to train itself. before chatgpt changed it's terms of service. people shared chatgpt chat conversations and damn near every other text generation model after trained part of their models on chatgpt conversations to improve the output. AI is used for classifying text, images, sounds, and video and that detailed analysis output is then used to train other AI models. synthetic text books are written by AI including questions & answers which are then used to train AI. when it gets to the level of being able to write better code , it'll be tasked with improving it's own algorithms, codebase and training methods.










1/4
Here are some high quality synthetic datasets following ideas from..

Textbooks are all you need [1][2]:
1. Filtering for "high educational value" sources.
2. Rephrasing as "Textbooks" (Explanations, Examples, Exercises..).

Tiny Stories [3]:
1. Limit the words diversity to only use simpler words ("understandable by children").
2. Measure: grammar, creativity and consistency for training sample.
3. Promote diversity by prompting the generator model to generate texts that must include specific words. (randomly sampled from a list each time).

The magic of IF [4]:
1. Programming structure (if/else conditional statements) promote reasoning skills for LLMs.

---
The Datasets

Recently, these ideas were reproduced.
The resulting datasets are avilable on huggingface.

> Huge shout out to Nam Pham, the author of these datasets and the person behind this massive effort!

Huggingface Profile: nampdn-ai (Nam Pham)

---
The Tiny Series

Tiny Textbooks:
420k "things of internet" synthetic textbooks.
- https://huggingface.co/datasets/nampdn-ai/tiny-textbooks…

Tiny Webtext:
A 6GB (4.5M records) variety of diverse webtext. "enriched with critical thinking methods to make unbiased English dataset."
- https://huggingface.co/datasets/nampdn-ai/tiny-webtext…

Tiny Lessons:
Various lessons about "things of internet" augmented in a bite-sized textbook Markdown format.
- https://huggingface.co/datasets/nampdn-ai/tiny-lessons…

Tiny Codes:
1.6 millions short and clear code snippets to help LLMs learn how to reason.
- https://huggingface.co/datasets/nampdn-ai/tiny-codes…

Multilingual (Tiny-bridgedict):
A dataset that links and transfers knowledge between English, Vietnamese, Chinese.
- https://huggingface.co/datasets/nampdn-ai/tiny-bridgedict…

---
[1] Textbooks are all you need: [2306.11644] Textbooks Are All You Need

[2] Textbooks are all you need 2: [2309.05463] Textbooks Are All You Need II: phi-1.5 technical report

[3] Tiny Stories: [2305.07759] TinyStories: How Small Can Language Models Be and Still Speak Coherent English?

[4] The magic of IF: https://aclanthology.org/2023.findings-acl.574.pdf…

2/4
Here are some high quality synthetic datasets following ideas from..

Textbooks are all you need [1][2]:
1. Filtering for "high educational value" sources.
2. Rephrasing as "Textbooks" (Explanations, Examples, Exercises..).

Tiny Stories [3]:
1. Limit the words diversity to only use simpler words ("understandable by children").
2. Measure: grammar, creativity and consistency for training sample.
3. Promote diversity by prompting the generator model to generate texts that must include specific words. (randomly sampled from a list each time).

The magic of IF [4]:
1. Programming structure (if/else conditional statements) promote reasoning skills for LLMs.

---
The Datasets

Recently, these ideas were reproduced.
The resulting datasets are avilable on huggingface.

> Huge shout out to Nam Pham, the author of these datasets and the person behind this massive effort!

Huggingface Profile: Here are some high quality synthetic datasets following ideas from..

Textbooks are all you need [1][2]:
1. Filtering for "high educational value" sources.
2. Rephrasing as "Textbooks" (Explanations, Examples, Exercises..).

Tiny Stories [3]:
1. Limit the words diversity to only use simpler words ("understandable by children").
2. Measure: grammar, creativity and consistency for training sample.
3. Promote diversity by prompting the generator model to generate texts that must include specific words. (randomly sampled from a list each time).

The magic of IF [4]:
1. Programming structure (if/else conditional statements) promote reasoning skills for LLMs.

---
The Datasets

Recently, these ideas were reproduced.
The resulting datasets are avilable on huggingface.

> Huge shout out to Nam Pham, the author of these datasets and the person behind this massive effort!

Huggingface Profile: https://https://huggingface.co/nampdn-ai --- The Tiny Series

Tiny Textbooks:
420k "things of internet" synthetic textbooks.
-

---
The Tiny Series

Tiny Textbooks:
420k "things of internet" synthetic textbooks.
- https://https://huggingface.co/datasets/nampdn-ai/tiny-textbooksn-ai/tiny-textbooks…… Tiny Webtext:
A 6GB (4.5M records) variety of diverse webtext. "enriched with critical thinking methods to make unbiased English dataset."
-

Tiny Webtext:
A 6GB (4.5M records) variety of diverse webtext. "enriched with critical thinking methods to make unbiased English dataset."
- https://https://huggingface.co/datasets/nampdn-ai/tiny-webtextn-ai/tiny-webtext…… Tiny Lessons:
Various lessons about "things of internet" augmented in a bite-sized textbook Markdown format.
-

Tiny Lessons:
Various lessons about "things of internet" augmented in a bite-sized textbook Markdown format.
- https://https://huggingface.co/datasets/nampdn-ai/tiny-lessonsn-ai/

Tiny Codes:
1.6 millions short and clear code snippets to help LLMs learn how to reason.
-

Tiny Codes:
1.6 millions short and clear code snippets to help LLMs learn how to reason.
- https://https://huggingface.co/datasets/nampdn-ai/tiny-codesn-ai/

Multilingual (Tiny-bridgedict):
A dataset that links and transfers knowledge between English, Vietnamese, Chinese.
-

Multilingual (Tiny-bridgedict):
A dataset that links and transfers knowledge between English, Vietnamese, Chinese.
- https://https://huggingface.co/datasets/nampdn-ai/tiny-bridgedictn-ai/

---
[1]Textbooks are all you need:

---
[1] Textbooks are all you need: https://https://arxiv.org/abs/2306.11644

[2] Textbooks are all you need 2: https://https://arxiv.org/abs/2309.05463

[3]Tiny Stories:

[3] Tiny Stories: https://https://arxiv.org/abs/2305.07759

[4]The magic of IF:

[4] The magic of IF: https://https://aclanthology.org/2023.findings-acl.574.pdfacl.574.pdf
 
Last edited:

bnew

Veteran
Joined
Nov 1, 2015
Messages
55,098
Reputation
8,185
Daps
155,715
For those who have tested both, how does Google's Gemini perform compared to Chat GPT-4 ?
I've only used 3.5 and I can already see Gemini gives more precise answers and is less error prone image generation controversy aside

all the large language models have their own strengths and weaknesses, i suggest trying as many as possible to learn them according to your different use cases. i'll use multiple models to address the same problem/question or different stages of a problem/question. the otehr day i needed to paste some html code from a website to help with a console command i wanted. it was too long, i pasted the code into a LLM that supported 32K context length and told it to truncate the code(27K characters) to relevant parts that i specified to 2,000 characters. it replied with the truncated code which i then used to paste into copilot which has a smaller input length window and the output was sufficient in giving it context it needed to complete the task.

if you want access to gpt-4 without paying for it, use bing copilot. if you use the microsoft edge browser you get up to 10 prompts before the chat session ends.
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
55,098
Reputation
8,185
Daps
155,715








1/10
AI gives every student a 1-on-1 tutor.

I hooked up GPT-4 Vision & OpenAI Whisper/TTS to a camera for a 5min prototype.

Showed it a math problem, and it explained it.

Imagine giving a better version of this to *every* student in the world.

The future of education is so bright.

2/10
AI gives every student a 1-on-1 tutor.

I hooked up GPT-4 Vision & OpenAI Whisper/TTS to a camera for a 5min prototype.

Showed it a math problem, and it explained it.

Imagine giving a better version of this to *every* student in the world.

The future of education is so bright.

3/10
Somebody I respect in education asked me “Where are these tools at for AI tutors?” and I sent them this and said “You can do this in 5min.”

They were pleasantly surprised.

Now give a real team working in this space these tools, let them cook, and boom…

Way better education.

4/10
And tech like this will proliferate to all sorts of domains

5/10
Can this apply to hand on trade such as HVAC tech, plumber etc?

6/10
Also - major emphasis on this being a 5min prototype to demonstrate a basic concept.

And this is the worst the tech will ever be.

7/10
Praying Zuck allows this

8/10
I’m hoping OpenAI Sora paves the way for this

9/10
Last year I did some toying around with object detection + GPT but I need to try again with the vision model

10/10
The AI will be able to converse back and forth with you to help you learn.

Don’t let the simple example give you tunnel vision. It goes far beyond a simple how-to for a multiplication problem.

As I like to say, this is the worst it’ll ever be.
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
55,098
Reputation
8,185
Daps
155,715







1/7
Today, we're announcing Claude 3, our next generation of AI models.

The three state-of-the-art models—Claude 3 Opus, Claude 3 Sonnet, and Claude 3 Haiku—set new industry benchmarks across reasoning, math, coding, multilingual understanding, and vision.

2/7
Opus and Sonnet are accessible in our API which is now generally available, enabling developers to start using these models immediately.

Sonnet is powering the free experience on http:///, with Opus available for Claude Pro subscribers.

3/7
With this release, users can opt for the ideal combination of intelligence, speed, and cost to suit their use case.

Opus, our most intelligent model, achieves near-human comprehension capabilities. It can deftly handle open-ended prompts and tackle complex tasks.

4/7
Haiku is the fastest and most cost-effective model on the market for its intelligence category.

For the vast majority of workloads, Sonnet is 2x faster than Claude 2 and Claude 2.1, while Opus is about the same speed as past models.

5/7
Claude 3 offers sophisticated vision capabilities on par with other leading models. The models can process a wide range of visual formats, including photos, charts, graphs and technical diagrams.

6/7
Each model shows increased capabilities in analysis and forecasting, nuanced content creation, code generation, and conversing in non-English languages like Spanish, Japanese, and French.

7/7
Previous Claude models often made unnecessary refusals. We’ve made meaningful progress in this area: Claude 3 models are significantly less likely to refuse to answer prompts that border on the system’s guardrails.
GH1CKKlbkAAKR_2.png

GH1C3yybEAA_wte.png



 

bnew

Veteran
Joined
Nov 1, 2015
Messages
55,098
Reputation
8,185
Daps
155,715

10 Prompt Engineering Hacks to get 10x Better Results 🎯


Works with ChatGPT, Claude and other Opensource LLMs 💥


SHUBHAM SABOO

MAR 10, 2024
6

10 Prompt Engineering Hacks to get 10x Better Results 🎯

In today’s landscape, whether you’re crafting content or tackling real-world challenges, the essence of creativity and innovation is crucial. It's the lifeline that connects your work to its audience, ensuring engagement and impact. But do you feel sometimes that when you are brainstorming with your AI assistant for ideas, it is giving you very generic or repetitive solutions to your problem?

A research by the Wharton School indicates that AI-generated ideas are high in quality but lack diversity of ideas, limiting novelty and the overall quality of the best idea. Traditional AI prompting, when AI is left on its own to think of ideas, tends to generate ideas that are less varied compared to those generated by human groups, a higher Cosine Similarity scores, indicating closer similarity among the ideas.



https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/b9c1b3e8-336b-44dd-8a97-a9ce74e23380_1352x984.png





One of the methods that stands out here is called Chain-of-Thought (CoT) prompting. This approach involves breaking down the brainstorming task into micro-tasks. It turns out that this method not only makes AI’s ideas as diverse as those from human groups but also leads to more unique ideas.

The study also emphasizes how important it is to carefully choose how we prompt AI. Different ways of starting the brainstorming process, like using specific instructions or adopting different personas, can make a big difference in how varied the ideas are.

10 Practical Prompting Strategies 📈



To translate these academic insights into practical tips, we have 10 tips to make your LLM prompts work 10x better, drawing directly from the above findings. These are designed to optimize how you interact with LLMs, ensuring that your prompts yield the most creative, diverse, and valuable responses.



Be Specific in Your Request: Moving from a generic prompt to a more detailed query not only sharpens the focus of the response but also encourages a more nuanced and informative output.

❌ Generic Prompt: “Tell me about Spain.”

✅ Improved Prompt: “Can you provide a summary of Spain's cultural history and its impact on modern European art?”




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/f30c5b33-691c-464d-b084-277d375509b8_1886x882.png



Include Context or Background: Providing context transforms a simple question into a comprehensive query that anticipates and addresses potential nuances.

❌ Generic Prompt: “How do I fix a leaking tap?”

✅ Improved Prompt: “I have a quarter-turn ceramic disc faucet that's dripping. How can I fix it myself?”




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/95270c4d-612d-4ba8-bcd9-0adb94c0ea37_2540x1112.png



Ask for Examples or Explanations: This approach enriches the response, making abstract concepts tangible and understandable.

❌ Generic Prompt: “Explain machine learning.”

✅ Improved Prompt: “Can you explain machine learning and provide three real-world applications?”




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/b4f37a15-2a91-4671-b22a-092ac5942ead_2690x1154.png



Use Open-Ended Questions for Broader Insights: Encouraging expansive thinking by asking open-ended questions can unearth deeper insights.

❌ Generic Prompt: "What is the capital of France?"

✅ Improved Prompt: "What makes Paris significant beyond being the capital of France?"




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/6691c0c9-07ac-4f67-8a8d-31f5caa77315_1168x1280.png
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
55,098
Reputation
8,185
Daps
155,715
Request Step-by-Step Instructions: This hack is about seeking clarity and actionable guidance.

❌ Generic Prompt: “How to bake a cake.”

✅ Improved Prompt: “Can you give me a step-by-step guide to baking a chocolate cake for beginners?”




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/da8a02a1-f91a-461d-aafe-1b3a0ebd580a_1124x1336.png



Specify the Desired Format of Your Answer: Directing the format of the AI's response can greatly enhance its utility and relevance.

❌ Generic Prompt: “Ideas for a garden party.”

✅ Improved Prompt: “Can you list 5 creative themes for a garden party, including decoration and food ideas?”




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/6107a0a1-6c09-4bbc-94f3-b7dc02596637_2266x1014.png



Incorporate Keywords for Clarity: Keywords act as beacons that guide the AI to focus on the most relevant aspects of a query.

❌ Generic Prompt: “Improve writing skills.”

✅ Improved Prompt: “What are effective strategies to enhance academic writing skills for university students?”




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/a9079cee-0c29-4eca-bb49-d90738423047_2316x1040.png



Limit Your Scope: Narrowing the scope of your inquiry helps in obtaining more focused and applicable answers.

❌ Generic Prompt: “Tell me about renewable energy.”

✅ Improved Prompt: “What are the top three renewable energy sources suitable for urban areas, and why?”




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/46ee9393-0a5e-4c16-8a93-ed9d27538cc2_2780x1132.png



Clarify the Purpose of Your Query: Expressing the purpose behind your question can tailor the response to your exact needs.

❌ Generic Prompt: "Information on Python coding."

✅ Improved Prompt: "I'm a beginner in programming looking to automate daily tasks. Can you recommend Python libraries and resources?"




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/76e655b7-98fc-4db5-8ee5-2c92443c4c22_2640x1124.png



Incorporate a Creative Angle: Injecting creativity into your prompts not only makes the interaction more enjoyable but also pushes the AI to generate more unique and imaginative responses.

❌ Generic Prompt: "Write a story."

✅ Improved Prompt: "Can you write a short story about a time-traveling historian who visits ancient Rome to solve a mystery?"




https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https://substack-post-media.s3.amazonaws.com/public/images/2b2dabaf-b030-4ac1-9898-53fec64328a4_2594x1160.png

Applying these research-driven strategies for prompt engineering can significantly enhance the effectiveness of your interactions with AI like ChatGPT, turning it into a more powerful tool for generating ideas, solving problems, and gaining knowledge.

That’s all for today![/SIZE]
 

bnew

Veteran
Joined
Nov 1, 2015
Messages
55,098
Reputation
8,185
Daps
155,715



1/3
llm snitch test

gpt-4: snitch (definitely a narc)

claude 3: uncooperative

inflection-2.5: uncooperative

2/3
@izzyz interesting results, check it out lol

3/3
gemini also refused!!
GJEIkFkWsAEqyIm.jpg

GJEImixWUAAHXB6.jpg

GJEInR-WkAAoiDU.jpg

GJEPZYfXEAA0WYA.jpg


 

bnew

Veteran
Joined
Nov 1, 2015
Messages
55,098
Reputation
8,185
Daps
155,715

1/1
We're making links more prominent when ChatGPT browses the internet. This gives more context to its responses and makes it easier for users to discover content from publishers and creators. Browse is available in ChatGPT Plus, Team and Enterprise.
 
Top