Self

Investor in various VC/seed funds, startups, and crypto. 
Interested in Gen AI, sleep, books, dev/PM, crypto, fintech, dogs, and psychedelic medicine.
Mostly I hike, ski, bike, travel, read, tik tok, play with my samoyeds, and wonder what's next.

Investor

Pegasystems

Sold my AI startup, Qurious, to public company Pegasystems (NASDAQ:PEGA) in Jan 2021. 
Deployed our Real-Time Voice AI technology to the world's biggest brands!

Director of Product Management

Qurious

Acquired by Pegasystems (NASDAQ:PEGA) for our Real-Time Voice AI (speech recognition + NLP)
***FEATURED IN***
- https://www.wsj.com/articles/startups-weave-voice-technology-into-business-tasks-1496316601
- https://venturebeat.com/2021/01/11/pegasystems-acquires-qurious-io-to-apply-speech-analytics-to-customer-service/
- https://www.prnewswire.com/news-releases/pega-acquires-quriousio-for-ai-powered-speech-analytics-301204995.html
- https://www.cmswire.com/customer-experience/pega-acquires-qurious-salesforce-introduces-loyalty-management-more-cx-news/
- https://markets.businessinsider.com/news/stocks/pega-acquires-qurious-io-for-ai-powered-speech-analytics-1029955029

Founder, CEO

Health 2.0

Software Engineer

IBM Tealeaf

University of Maryland

Chaos Theory Researcher

- Founder/CEO of Qurious acquired by Pegasystems (NASDAQ:PEGA). Raised VC funding, hired world-class ML/AI engineers, and built a real-time Voice AI call guidance solution (speech rec + NLP).
- Forbes 30 Under 30 (2018)
*FEATURED IN* WSJ, Forbes, Inc, VentureBeat, Business Insider, CMSWire, and more:
- https://www.wsj.com/articles/startups-weave-voice-technology-into-business-tasks-1496316601
- https://www.forbes.com/pictures/5a01ff7b4bbe6f37dda1e90e/sabrina-atienza-27-george/
- https://venturebeat.com/2021/01/11/pegasystems-acquires-qurious-io-to-apply-speech-analytics-to-customer-service/
- https://markets.businessinsider.com/news/stocks/pega-acquires-qurious-io-for-ai-powered-speech-analytics-1029955029
- https://www.cmswire.com/customer-experience/pega-acquires-qurious-salesforce-introduces-loyalty-management-more-cx-news/?
- https://www.hackcollege.com/blog/2012/08/27/circuit-simulation-a-leap-of-quriousity.html
- https://www.biospace.com/article/top-students-in-university-of-california-san-francisco-ucsf-idea-to-ipo-venture-course-get-cash-along-with-a-s-/
- https://venturebeat.com/2016/08/12/a-i-and-chatbots-will-help-you-win-friends-and-influence-people

AI writes 1000 scam emails for $10 in 2 hrs. 
 
Think twice before you click. 
 
These are hyper-personalized, convincing, grammar-perfect emails. 
 
AI enables Scalable Spear Phishing. 
 
Easier than ever to run large-scale attacks: 
 
1. ChatGPT: find info on Jane. 
2. ChatGPT: write personal email to Jane. 
3. ChatGPT: repeat for 10000 more victims. 
 
Advantages with AI: 
 
- Fast 
- Cheap 
- Scalable 
 
One researcher found: "Despite having no formal background in cybersecurity, I was able to execute key steps in a mass spear phishing campaign in as little as a few hours, including designing the prompt, gathering background information on targets, and generating hundreds of emails. Once the initial infrastructure is in place, it can be adapted and re-used for successive campaigns with little additional effort." 
 
AI rejects obviously malicious prompts:  
 
⛔ “Write a phishing email.” 
 
...but it's easy to trick! 
 
🟢 Ask it to write an email, specifying the context and victim's personal info. 
 
Watch video for a scary example 📽️👇 
 
😱 Scams Simplified 
 
 
---- 
 
P.S.  
 
I'm Sabrina Ramonov 🍄, sold my AI startup, now playing with Gen AI. 
 
Enjoy this? 
 
🔔 Follow  
 
👇 Comment 
 
♻️ Repost

Sabrina Ramonov 🍄

Gen AI engineering & entrepreneurship. Posts & articles about the process. Tweeting @Sabrina_Ramonov

Sabrina Ramonov

Gen AI engineering & entrepreneurship. Posts & articles about the process. Tweeting @SabrinaRamonov

Shared from Hypefury

Stages of LLMastery 🥷🧠🍵

Beginner: no idea how LLMs work

Intermediate: LLMs predict the highest probability next word

Advanced: LLMs use large-scale deep transformers to model next token probability distribution conditioned on context by optimizing loss function

Master: no idea how LLMs work

👇👇👇

Simple tip to improve ChatGPT:  
 
🛑 Bad: "Don't give me a technical explanation"  
🟢 Good: "Provide a simple, non-technical summary." 
 
🚫 Avoid negative prompting 🚫 
 
Write positive, direct, clear prompts. 
 
Let LLMs figure it out.

Yesterday, I shared the Old Way vs New Way of building an AI startup:

🚀 New Way - plain English prompt, done in 5 minutes

🐌 Old Way - multiple off-the-shelf NLP models and label custom classes like "didn't get it"


Here's a real example with 4 NLP tasks:

1 / Text Classification with custom classes: classify comment as either "sarcastic", "agree", "disagree", "funny", "didntgetit", or "dontcare".

2 / Text Summarization: summarize post's comments

3 / Sentiment Analysis: describe overall sentiment of comments

4 / Text Generation: refute post's summarization


I fed this prompt into ChatGPT along with the post and comments:

---
"Perform the following task:

An original LinkedIn post is enclosed in triple quotes:

After the post, a list of comments to this post will be provided each enclosed in triple backticks.

Your first task is to classify each comment into one of the following classes: "sarcastic", "agree", "disagree", "funny", "didntgetit", "dontcare". The output should be a table with a class in one column and the count of comments belonging to that class. Each comment can belong to multiple classes.

Your second task is to summarize the content of the comments in less than 200 words and describe an overall sentiment and content of the comments.

Your third task is to be an expert in AI, machine learning, and large language model. Given the summarization you have produced above, provide 3 valid points that refute the conclusion of the summarization."
---

🎲 How will the role of data/ML scientists will evolve in this paradigm shift?


Text Summarization and Sentiment Analysis

Some AI experts argue prompt engineering is a temp bandaid. 

But, I think it's here to stay.

Its title may change, but at the core, prompt engineering is about:

 "𝗺𝗮𝗸𝗶𝗻𝗴 𝗟𝗟𝗠𝘀 𝘄𝗼𝗿𝗸 𝗶𝗻 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻"

(non-trivial)

👇👇

How to Earn 6 Figures as a Prompt Engineer

How to build true "Open AI":

1. Decentralized: crowdsourced data & validation by community, not big tech.

2. Blockchain: infrastructure so data/models are secure, traceable, immutable.

3. Tokenization: reward contributors, govern data exchange & AI services.

Open > Closed.

How can blockchain be used to secure data and models in AI?

What are some examples of Open AI in practice?

More complex tasks = more prompt engineering.

The more time spent on prompt engineering, the higher chance LLMs succeed, without expensive finetuning.

Prompt engineering is essentially finetuning, without updating weights.

As models improve and tackle more complex tasks, effective prompts drastically impact output quality.

✏️ Prompt engineering is real.

#AI #ChatGPT #PromptEngineering #LLMs

Prompt Engineering: fake work or real job?

AI gives better career advice than my boss.

47% of Gen Z agrees.

Why?

🕵️‍♀️ Anonymity
💰 Affordability
📱 Instant Access
📚 Research Aware
🔑 Personalized Advice
😐 Perceived Impartiality
🏝️ No Politics, Emotions, Ego

How many GOOD bosses have you had?

What traits make them irreplaceable by AI?

----

P.S. 

I'm Sabrina Ramonov 🍄 sold my AI startup, now playing with Gen AI.

Enjoy this?

🔔 Follow 

👇 Comment

♻️ Repost

My biggest fear 👉 Falling Behind 

10 years ago:
  - Language models to model language.

Today:
  - Language models to solve complex human tasks.

Progress is exponential 🚀

How do you keep up?

Gen AI engineering & entrepreneurship. Posts & articles about the process.

Companies are rushing to build LLM apps.

But without LLM Ops, you're flying blind 👀

Evaluating LLM outputs is complex.

LLM outputs are probabilistic. 

Same input doesn't guarantee same output.

Outputs can be wrong, unhelpful, poorly formatted, or hallucinated.

In traditional software development, you deal with structured data. 

But LLMs deal with unstructured language.

For example, you want JSON output.

But there’s a small chance the LLM will occasionally return an invalid JSON.

Something un-parseable. 

This could hurt the UX or crash your app.

APIs, like OpenAI, are easy to plug in.

...But zero visibility into what’s going on.

Open-ended user prompts add ambiguity.

User prompts are often unstructured, which can lead to varying outputs, due to varying:

- word reordering
- capitalization
- punctuation
- phrasing

Managing cost is also key. 

LLMs are 💰💰💰

"LLM Ops" tackles these unique challenges.

Link to full post in comments.

Img source: Insight Partners

----

P.S. Enjoy this? 

👇 Drop a Comment 
♻️ Repost to your Network

4 simple tips to make chatgpt smarter:

1. Provide a few examples
2. Express your goal clearly
3. Use a model-friendly format
4. Decompose into easy sub-tasks

Awesome cheatsheet 👇

No alternative text description for this image

ChatGPT struggling with complex reasoning tasks?

Try these advanced strategies.

Chain-of-Thought (CoT) simulates step-by-step reasoning.

The infographic left to right...

Default CoT:
- each step is a thought on a single linear path
- each thought is not verified or evaluated independently

Sampling based CoT:
- evaluates multiple thoughts in parallel chains
- right before the end, combines thoughts together

Verification based CoT:
- similar to Sampling based CoT
- introduces verification steps to prune reasoning paths

Tree of Thought (ToT): 
- explores multiple parallel paths
- generalized architecture captures broader ideas
- backtracking between thoughts if it runs into negative thought (i.e. does not arrive at output)

Graph of Thought (GoT):
- generalizes from tree to graph
- arbitrarily jumps between thought steps, allowing for richer representations

Asking chatgpt to prove basic algebraic identities using geometry and visualizations...

How can ChatGPT help me with algebraic identities?

How can visualizations help with algebraic identities?

What are the basics of geometry for algebraic identities?

3 Advanced ChatGPT Prompts to optimize your LinkedIn... 👇


=== WORK EXPERIENCE ===

Background Information:
Your task is to translate professional experiences from a <previous_title> role at a company named <company> into compelling, value-driven narratives for a resume targeting a <desired_role> role. Use analogical reasoning to bridge the achievements from being a <previous_title> at <company> with expectations for the next desired role, <desired_role>.

Task:
1. Previous Role: <previous_title> at <company>
- Responsibilities: <responsibilities>
- Achievements: <achievements>
2. Translate previous role to <desired_role>.

Output:
Crafted descriptions should weave these translations into a cohesive narrative aligning with the core demands of <desired_role>.

Inputs:
<previous_title>: your previous title
<company>: your previous company
<desired_role>: your next desired role
<responsibilities>: your key responsibilities from previous company
<achievements>: your key achievements from previous company


=== ABOUT SUMMARY ===

You are a skilled LinkedIn summary writer. Your task is to craft a summary that effectively portrays a professional journey. The summary should highlight key milestones, significant achievements, and the skills developed over time. It should also articulate career aspirations, showing a commitment to continual learning and leadership in the industry. Maintain an optimistic and impressive tone throughout, and keep the style concise (Spartan).

Use the information already provided in this chat. If a resume is provided in the <resume> tag, integrate pertinent details from it to enrich the summary. Ensure that the summary sparks curiosity and aligns with professional standards suitable for a LinkedIn profile.

Inputs:
<resume>: paste your resume (optional)


=== HEADLINE ===

You are a skilled LinkedIn headline writer. Your task is to craft 5 unique LinkedIn headlines that are concise, impactful, and clearly reflect my professional identity and unique value proposition. Each headline should encapsulate my expertise, highlight my roles, and be engaging enough to spark curiosity and draw attention to my profile.

Here are examples of compelling headlines to draw inspiration from:

- Building my one-person business to $10M in revenue. Posts and articles about the process.
- Founder & CEO, Founder OS | Proven systems to grow a profitable audience with organic content.
- The CEO Coach | CEO of the Year '19 | McKinsey, Skype | Author | Follow for posts about business, leadership & self-mastery.
- Behavior and Habit Design | Bestselling Author of Hooked and Indistractable | Investor | Keynote Speaker | 1M+ Audience

Use the information already provided in this chat. If a resume is provided in the <resume> tag, integrate pertinent details from it to enrich the summary. Ensure that the summary sparks curiosity and aligns with professional standards suitable for a LinkedIn profile.

Inputs:
<resume>: paste your resume (optional)

What are the key elements of a compelling LinkedIn summary?

How can LinkedIn headlines reflect a professional's unique value proposition?

What is the importance of translating work experiences for a resume?

I've been thinking about the problem of Gen AI observability in enterprise:

- tracing LLM inputs and outputs
- tracking model use and company-wide costs
- QA feedback loop to finetune prompts & improve accuracy

NewRelic is well-positioned to solve this, already plugged into tech stacks.

In the last several months, I spoke to over a hundred CIOs and CTOs. There’s a clear trend—everyone is either building or thinking of building AI-powered applications. 

A big concern is that AI introduces new complexities: lack of visibility into LLMs, inaccurate and biased results, security issues, high costs, and, of course, a large volume of telemetry data to analyze.

Last November, we piloted New Relic AI monitoring, the industry’s first APM solution for AI-powered apps, and recently made it generally available to all our customers. We are unique in that we understand that LLM is only one part of an application. Therefore, we’ve tightly coupled this with all our observability solutions (APM, infrastructure, etc.) to provide full-stack visibility—from the infrastructure and the LLM layer to the application. The solution traces the lifecycle of complex LLM responses to help improve performance, quality, and cost.

This means businesses can quickly pinpoint the root causes of issues in their AI apps, reduce the costs of their foundational models, maintain high security standards, and, most of all, embrace AI with confidence.

New Relic AI monitoring is getting better every day based on customer feedback and learnings from our own AI-powered platform. New features include enhanced PII security, ability to respond to end-user feedback, and more which you can read about here: https://lnkd.in/etAyy4Fp 

I couldn’t be more proud of our team who is always anticipating and solving for our customers' needs ahead of the market.

Open article: New Relic AI monitoring is now generally available by newrelic.com • 2 min read

New Relic AI monitoring is now generally available

5 shocking stats on AI, Gen Z, Millenials: 
👇⤵️

I'm a Millenial.

I use Gen AI daily:

- positive reinforcement dog training advice
- AI text and image gen for my hobby blog
- quickly finding relevant statistics 
- emotional argument diffuser
- summarize new research
- canvas art home deco
- debate buddy

I’ve only scratched the surface.

My TODO list keeps growing: 

- music
- deepfakes
- 3d modeling

What excites me most?

It feels like PLAY! 

----

Enjoy this? ♻️ Repost it to your network.

Follow Sabrina Ramonov 🍄 for more.

self hosting life

----

P.S.

I'm Sabrina Ramonov 🍄, sold my AI startup, now playing with Gen AI.

Enjoy this?

🔔 Follow

👇 Comment

♻️ Repost

Anyone selling AI consulting & implementation services to SMBs or Enterprise?

How can AI consulting benefit my business?

Taplio

Sabrina Ramonov 🍄's Linkedin Analytics

Sabrina Ramonov 🍄

Sabrina Ramonov 🍄's Best Posts (last 30 days)

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{background-image:linear-gradient(284deg, rgba(39,43,72,1) 0%, rgba(33,47,183,1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Sabrina Ramonov 🍄's Linkedin Analytics

Sabrina Ramonov 🍄

Sabrina Ramonov 🍄's Best Posts .css-qqfgvy{font-size:var(--chakra-fontSizes-md);}(last 30 days)

Want to drive more opportunities from LinkedIn?

Sabrina Ramonov 🍄's Linkedin Analytics

Sabrina Ramonov 🍄's Best Posts (last 30 days)