- The Searchlight
- Posts
- Prompt like a Meta Engineer
Prompt like a Meta Engineer
Meta engineers share 6 prompting tips to get the best results from Llama 2
Top of the Post 📥
Hello, entrepreneurs!👇️
Podcast: If you want to leave your 9-5, you MUST understand this…
Biz Hack: Prompting like a Meta engineer
Biz Essentials: Develop a commenting strategy for more love on your LinkedIn posts.
Save the Date: Richard Branson will answer all your career questions live
Weekly Catchups
News | Industry Insights | Trends
💼 For better or worse: Amazon is set to revoke its hybrid work policy and order its staff to come to the office five days a week. CEO Andy Jassy is concerned that the company which preserved the intensity of a start-up while growing out to be a tech-giant was seeing its work culture being ‘watered down’ due to flexible work hours and its layered bureaucracy.
🗣️For the better: Meta will now designate users under 18 as “Teen Accounts” by default, flipping on more restrictive settings on who can interact with them and what kind of content they’ll see. It’s a huge change from the platform’s previous attempts at protection.
💰️ $ 0.005: The increment that many stock quotes will come in, starting November 2025. SEC rules currently set the minimum price increments for most exchange-listed stocks at a penny. The SEC voted to approve half-cent quotes for stocks that frequently have a 1-cent difference between their buying and selling prices.
Podcast 🎙️
NEW: Earlier this week, I spoke to Kate Bernal, a seasoned financial freedom coach and online business strategist, helping 9-to-5 professionals break free from the rut and achieve financial freedom.
Whether you are aiming to retire early or build a thriving online business, Kate's expertise will help you leap forward. Are you game?
Biz Hacks
Strategies | Tools | Marketing Tips
Prompt Like a Pro: Six Tips To Make the Most of Llama 2
Your LinkedIn feeds are inundated with the forecasts of an impending doom.
Experts are convinced that AI will replace you in your full-time job.
But a part of you casts a shadow on this sort of wishful thinking. You are basing this on your experiments with AI. These have proven to be unhelpful.
You are not wrong. Perhaps you are a great marketeer or writer who cannot overlook the fluff in AI-generated responses.
In my experience, AI can never perform 100% of any job. But I’ll also admit that it can save you from much of the grunt work before D-day (read product launches or nerve-wracking investor pitches.)
That’s why we revisit six prompting tips Meta engineers shared to get the best results from Llama 2, its flagship open-source model.
Let’s dive in ⏬
#1 | Instruct Explicitly
It may seem a no-brainer, but we urge you to avoid lazy prompting.
Rule of thumb: Explicit instructions produce better results than open-ended prompts. Specificity is key. Use details on the length of output and character count.
Example
Generic: Summarise this document
Better: “As a startup founder preparing for an investor pitch, summarize the growth and revenue projections in this business plan in 300 words, highlighting key financial metrics and market expansion strategies."
Is the response still not up to the mark? Use formatting instructions (bullet points) or direct the model to adjust technical complexity.
#2 | Role Prompting
Rule of thumb: Assign the model a specific role for clear responses.
Generic: “What are the pros and cons of scaling a business?”
Better: "You are a growth strategist advising a founder looking to scale their SaaS company beyond the initial 1,000 customers. Outline the pros and cons of scaling with limited cash flow and expanding into new markets.”
#3 | Chain-of-Thought Prompting
Rule of thumb: Encourage the model to think step-by-step can improve its reasoning drastically. Chain-of-thought (CoT) prompting ensures processes information to churn out the best possible helpful answer.
Generic: “Compare the following: bootstrapping or raising venture capital.”
Better: "Which is better for scaling a B2B SaaS startup: bootstrapping or raising venture capital? Let’s break this down step-by-step, considering factors like equity, growth speed, and long-term control of the company."
#4 | Retrieval-Augmented Generation
Rule of thumb: Use external data sources to improve relevance.
Example: “Create a social media strategy for my startup by analysing engagement data [insert preferred timeline], including follower growth and top-performing posts. Incorporate competitor analysis and industry trends to suggest the best platforms and content types for the next quarter.”
#5 | Self-Consistency
Remember: Even while employing some of Meta’s prompting tricks, one SINGLE generation could produce incorrect results ‘self-consistency’ ensures the system regularly checks its own work to improve.
Example: "Analyze the most successful marketing channels for our product launch. Then identify which channel delivered the best ROI and explain why it performed better than the others."
This encourages the model to review multiple options and assess the most valuable one, similar to how founders analyze business decisions.
#6) Limiting Extraneous Tokens
Rule of thumb: To avoid irrelevant information, combine rules with role prompts and clear instructions.
Generic: “What is product-market fit?”
Better: “Provide a concise explanation of product-market fit, specifically focusing on the key indicators for a SaaS startup looking to scale."
For Those Who Seek Unbiased News.
Be informed with 1440! Join 3.5 million readers who enjoy our daily, factual news updates. We compile insights from over 100 sources, offering a comprehensive look at politics, global events, business, and culture in just 5 minutes. Free from bias and political spin, get your news straight.
Biz Essentials
Stay Ahead of the Curve
She was rejected by 100+ investors.
An outcast among founders.
But today, her business is worth $26 billion and has 200 Million monthly users.
Here’s her story:
— Hosun (@hosun_chung)
3:20 PM • Aug 26, 2024
Don’t Miss It!
Networking | Business Development | Entrepreneurship
Feedback, Please!
Your feedback is invaluable to us in shaping our future newsletters. Please take a few moments to share your thoughts.
How satisfied are you with this week's newsletter? |
Are there specific topics, questions, or areas you want us to explore further? Let us know, and we'll make it happen.