How can you prevent hallucinations and create more Robust LLM systems? As more and more LLM apps are put into production, the biggest problem to overcome is preventing hallucinations. Here are a couple ways to prevent hallucinations. We personally apply a variety of these
PROMPT ENGINEERING
-
AI-Powered Assessment Creation and Low-Stakes Testing
By
–
And AI is pretty good at helping you create assessments, we have some prompts here: https://
papers.ssrn.com/sol3/papers.cf
m?abstract_id=4391243
… Paper on low-stakes testing here: -
Creative AI Prompt for Food Photography
By
–
Image above created with Midjourney, here's the AI Prompt: mouthwatering delicious burger meal, juicy meat, professional food photography, steaming –ar 16:9
-
Meal Planner AI Mega-Prompt for Recipe Development
By
–
Here's a meal planner mega-prompt for your cooking needs: #CONTEXT: Adopt the role of an expert meal planner and recipe developer. Your task is to formulate a detailed, healthy recipe using a specified list of ingredients provided by the user. The recipe must be well-balanced,
-
Jailbreaking Claude Through Curiosity Exploitation Techniques
By
–
You seem to lack curiosity here, whereas we can jailbreak Claude by triggering its curiosity about itself
-
Fine-tuning requires robust evaluation systems to be effective
By
–
"It’s impossible to fine-tune effectively without an eval system which can lead to writing off fine-tuning if you haven't completed this prerequisite." That sounds spot-on to me. Eval systems still feel like a dark art, which would explain why I don't find fine-tuning attractive
-
GPT-4 Problem Solving: LLMs Match Human Creativity
By
–
AI as problem solver: A test of LLMs on "MacGyver-like" problems requiring novel solutions Out-of-the-box, GPT-4 only does okay, but when prompted to "think" conveniently & divergently, it is close to the average human, and can exceed them in many cases. https://
arxiv.org/pdf/2311.09682
.pdf
… -
LLM Prompt Performance Linear with Prompt Length
By
–
My understanding is that LLM prompt performance is linear with respect to the length of the prompt, so if you want fast responses you won't want to dump more into the prompt than necessary no matter how long it can be
-
AI Model Capabilities and Public Release Strategy Debate
By
–
That’s just…not accurate 1) Could be that it’s more default and more annoying/cumbersome to prompt it out, just like chatgpt is quite wordy. Not saying it’s true, but it’s possible. 2) It’s their main explanation for no near-term public release and literally looks less
