If you haven’t hit PMF, you should not be fine-tuning models. Just prompt. Only exceptions:
– Speed is required for your use-case (and you need a model more powerful than Mixtral)
– Opus/GPT-4 cannot do your task or are too expensive per use
PROMPT ENGINEERING
-
Fine-tuning Models: When to Use It Versus Prompting
By
–
-
ChatGPT mega-prompt for expert business problem-solver coach
By
–
Here's a ChatGPT mega-prompt that turns it into an expert business problem-solver: #CONTEXT:
You are an expert Business Coach AI. You are a world-class coach for Entrepreneurs who build their small businesses with limited resources. You are well-known for helping people to find -
Top AI Use Case: Personalized Educational Content Generation
By
–
Honestly this is the number one AI use case people tell me about. (The more intense example is to add “and that teaches a lesson about astronomy appropriate for a five year old”)
-
Sleep Deprivation Effects on Cognitive Performance and Output Quality
By
–
My biological neural network after a bad night of sleep:
– decoding temperature turned up to T=1.5 (usual is 0.7)
– gives final answer quickly without using chain-of-thought
– base model comes out: 90% of ideas are low quality, but 10% are profound
– safety filter missing, easily -
Prompting for Speed, Fine-tuning for Scale After PMF
By
–
Even if the outputs aren’t 100% ideal, prompting allows for rapid experimentation to reach PMF. You don’t want to fine-tune when your product may change tomorrow. After you’ve found PMF, you’ll want to “harden” your AI (make it more reliable, faster, cheaper). Fine-tuning does
-
Prompt engineering and fine-tuning for product market fit
By
–
Prompt your way to PMF Then fine-tune to scale
-
Custom GPTs for Academic Paper Analysis on Twitter
By
–
My GPTs for people reading academic papers on Twitter:
Why is this important? https://
chat.openai.com/g/g-jcGK9yHuC-
but-why-is-it-important
…
Is it causal? Explains whether a paper makes a causal claim: https://
chat.openai.com/g/g-GGnYfbTin-
correlation-isn-t-causation-a-causal-explainer
…
I’ll look that up for you: show it a screenshot, it finds the source: https://
chat.openai.com/g/g-HThV0Y8e2-
fine-i-ll-look-that-up-for-you
… -
Visualization Techniques Enhance LLM Problem-Solving Performance
By
–
It is weird how effective it is to apply human-inspired approaches to problem solving to help LLMs “think” better. Here, asking the AI to visualize each step in a navigation problem by drawing diagrams helps greatly improve performance on the problem.
-
Devin’s Real Innovation: Infrastructure, LLM Stack, and Agent UI/UX
By
–
i havent personally used that one yet so i cant opine (see how this works?) to me the devin innovation is excellent infra, and tasteful/creative combination of LLM OS stack, but what strikes me the most is in the async agent ui/ux. swebench is just a vanity metric. because of
-
Genetic Algorithm Image Optimization Through Ruthless DOES Curation
By
–
Once we locked in on a theme, we entered a feedback loop where we used relentless curation from the DOES team to optimize the input settings. It was ruthless — 97% of the images were deleted. The surviving ones would mutate and multiply in the next generation, like in a genetic