Self-belief - helps you to do anything with confidence.
Believing in your team's abilities - helps you to delegate.
Encouraging & Increasing your team's self-belief - helps to scale much better.
At a different level, U need both go-getter ability and leadership qualities.
Stanford researchers built a new prompting technique!
By adding ~20 words to a prompt, it:
- boosts LLM's creativity by 1.6-2x
- raises human-rated diversity by 25.7%
- beats fine-tuned model without any retraining
- restores 66.8% of LLM's lost creativity after alignment
Let's understand why and how it works:
Post-training alignment methods like RLHF make LLMs helpful and safe, but they unintentionally cause mode collapse. This is where the model favors a narrow set of predictable responses.
This happens because of typicality bias in human preference data:
When annotators rate LLM responses, they naturally prefer answers that are familiar, easy to read, and predictable. The reward model then learns to boost these "safe" responses, aggressively sharpening the probability distribution and killing creative output.
But here's the interesting part:
The diverse, creative model isn't gone. After alignment, the LLM still has two personalities. The original pre-trained model with rich possibilities, and the safety-focused aligned model.
Verbalized Sampling (VS) is a training-free prompting strategy that recovers the diverse distribution learned during pre-training.
The idea is simple:
Instead of prompting "Tell me a joke" (which triggers the aligned personality), you prompt: "Generate 5 responses with their corresponding probabilities. Tell me a joke."
By asking for a distribution instead of a single instance, you force the model to tap into its full pre-trained knowledge rather than defaulting to the most reinforced answer.
Results show verbalized sampling enhances diversity by 1.6-2.1x over direct prompting while maintaining or improving quality.
Variants like VS-based Chain-of-Thought and VS-based Multi push diversity even further.
You can find the paper link in the next tweet.
π Over to you: What other methods can be used to improve LLM diversity?
@SwiggyCares Made an order via Swiggy genie - What we got is a threatening from the delivery person & his whole friends from all over india. Threatening to circulate the phone number all over his community.
Customer safety is not taken care well @SwiggyCares
@SwiggyCares Wondering the security and safety of customer's is not protected.
@SwiggyCares - In this case, the customer is a Women. It is more worst to see Swiggy Delivery Person's threatening like this.
@SwiggyCares - But - someone referring as Swiggy delivery person's brother calling WA, Sending Voice msgs continuoisely & threatening saying that he will be circulating the phone number to his community and make sure to take revenge.
@SwiggyCares - Ordered 2kgs Mutton (Bangalore).
- Delivery partner did not deliver but marked as deliverd.
- When contacted Swiggy support - They have given the delivery person phone number
- Contacted the delivery person, but no answer
- Later swiggy refunded the money. No issues till here.
Hand gesture control using @AnthropicAI Claude 3.5 single-shot python. Llama 3, via @GroqInc , significantly improved the physics( velocity, hysteresis, and deadband).
However, GPT-4o introduced several errors, lmao ππ .
Itβs not perfect yet; it still need to reduce shakiness.
Ever had a #YehMeraIdeaTha moment? π‘
We feel you, and we bring you a chance to make up for it. π
Weβre excited to partner with @sharktankindia to support a whole new era of budding entrepreneurs. Stay tuned. π¦ π²
I recommend everyone work at an early stage startup at some point in your career (preferably early).
Youβll learn more about business, what you enjoy doing and what you donβt enjoy doing.
The experience exposes you to so much. Youβll grow 10x more than a corporate job.