Natalie Ho

0 followers · 833 views

on the atlas — 43

highlights — 691

equity ask as conviction in the upside
How to negotiate your startup salary and equity offer
don’t say the word “join” or “sign” until you have the offer you want
How to negotiate your startup salary and equity offer
Some companies retain the right to buy back your options at the exercise price, meaning you don’t get any value out of those options. Another
Startup equity basics: What to ask before you accept | Human Interest
You need to file within 30 days of your grant
I didn't understand startup stock options - it cost me $300K
if the startup gets acquired, your unvested stock options / RSUs immediately vest
I didn't understand startup stock options - it cost me $300K
diversify its revenue
Clay's Valuation: A Closer Look Behind the Hype
sophisticated data brokerage with limited long-term defensibility
Clay's Valuation: A Closer Look Behind the Hype
Annual Recurring Revenue (ARR) may be pass-through revenue. This means that up to 50% of its reported revenue could be directly channeled to third-party data providers
Clay's Valuation: A Closer Look Behind the Hype
typically used to send metadata.
HTTP API integration overview | Documentation | Clay University
search or filter criteria.
HTTP API integration overview | Documentation | Clay University
their fragmented sales tech stack was causing inefficiencies across their rapidly scaling sales team.
How Vanta uses Clay to streamline RevOps and scale signal-based prospecting - The GTM with Clay Blog
rapid experimentation and self-service workflows
How Rippling uses Clay to scale growth experiments and email personalization - The GTM with Clay Blog
Scaling personalization and experimentation while maintaining accuracy
How Rippling uses Clay to scale growth experiments and email personalization - The GTM with Clay Blog
“If someone told me I had to get rid of Clay, I’d go on a two-week retreat to think about what my life had become
Partnering with Clay: On a Mission to Grow | Sequoia Capital
helping businesses grow
Partnering with Clay: On a Mission to Grow | Sequoia Capital
many users weren’t sure where to start
Partnering with Clay: On a Mission to Grow | Sequoia Capital
deeply understood the ever-changing ways developers used data—and they said he was thoughtful, trustworthy and wickedly smart.
Partnering with Clay: On a Mission to Grow | Sequoia Capital
platform was focused on workflow and data integration—an intuitive and powerful combination of spreadsheet interfaces, data ingestion, and functional programming that helped customers automate and streamline complex operations
Partnering with Clay: On a Mission to Grow | Sequoia Capital
phrasing was done in parts doing so
Gmail
Make Data-Driven Decisions: Utilize analytics and data visualization
Job Application for Strategic Projects Lead - New Grad at Scale AI
Build Complex Infrastructure
Job Application for Strategic Projects Lead - New Grad at Scale AI
strong work ethic, a passion for problem-solving, a versatile and entrepreneurial mindset, and a willingness to learn and adapt quickly
Job Application for Strategic Projects Lead - New Grad at Scale AI
wanted to find a way of cutting a pie that is “envy-free,” meaning that each person is at least as happy with his own piece as he would be with anyone else’s. They also wanted to make sure the division is “efficient,” so that no other way of dividing the pie would be better for one person without becoming worse for others.
Cutting a Pie Is No Piece of Cake
from voting, taxation, crime, climate change, or immigration in scenarios that foster numerical and data skills like statistical analysis or the assessment and weighing of evidence in the context of politics in our instruction.
The Missing Ingredient in Our Democracy: Math (Opinion)
My college students are frequently incredulous that they have not learned more about the math behind some key democratic processes like voting, the Electoral College, apportionment of legislative seats, and gerrymandering.
The Missing Ingredient in Our Democracy: Math (Opinion)
political quantitative literacy
The Missing Ingredient in Our Democracy: Math (Opinion)
With this procedure, a contingent of voters can force all the winners to be candidates of their choosing
Boston needs ranked choice voting - CommonWealth Beacon
ranked choice voting
Boston needs ranked choice voting - CommonWealth Beacon
Geoffrey Hinton, Reid Hoffman, Andrej Karpathy
Hello, World Labs
where rapid scientific progress has thinned the barrier between research and applications
Hello, World Labs
focus on generating 3D worlds without limits - creating and editing virtual spaces complete with physics, semantics, and control.
Hello, World Labs
Spatial intelligence also helps us create, and bring forth pictures in our mind's eye into the physical world. We use it to reason, move, and invent - to visualize and architect anything from humble sandcastles to towering cities.
Hello, World Labs
Large Language Models
Meet the AI agent engineer | Sierra
How can we optimize our model serving latency by 10x while keeping costs flat?" or "What's the best way to horizontally scale our generative AI pipeline to handle 100x more traffic?
What I learned as a forward-deployed engineer working at an AI startup
GPT APIs, building retrieval-augmented generation (RAG) systems, and creating small chatbots
What I learned as a forward-deployed engineer working at an AI startup
Need to optimize a model's architecture for performance? Develop a speech-to-text pipeline? Consult on a CI/CD pipeline for a novel AI system?
What I learned as a forward-deployed engineer working at an AI startup
Lots of people want to build things with machine learning, but they don't have the expertise to use it. It's not technology holding back adoption of machine learning, it's the fact that you need all this specialist knowledge to use it
Machine learning needs better tools – Replicate
makes it easy to package a machine learning model inside a container so that you can share it and deploy it to production.
Machine learning needs better tools – Replicate
It was hard to run open-source machine learning models. All these advances were locked up inside prose in PDFs, scraps of code on GitHub, weights on Google Drive
Machine learning needs better tools – Replicate
You shouldn’t have to understand GPUs to use machine learning, in the same way you don’t have to understand TCP/IP to build a website.
Machine learning needs better tools – Replicate
We argue that it is not fair that life-changing decisions are made with an error-prone system, without entitlement to a clear, verifiable, explanation.
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
it is easier to debate the fairness of a transparent model than a proprietary model.
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
A judge could easily memorize the models within these works, and compute the risk assessments without even a calculator
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
Without transparency, an incorrect understanding of a model (e.g., the form of its dependence on age) can go unchecked, leading to downstream consequences for independent analyses of a model.
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
137 variables (Northpointe, 2009) that are collected from a questionnaire
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
Correctional Offender Management Profiling for Alternative Sanctions)
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
Secret algorithms control important decisions about individuals, such as judicial bail, parole, sentencing, lending decisions, credit scoring, marketing, and access to social services.
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
Our analysis also suggests that COMPAS scores may often be miscomputed. These kinds of errors can lead to years of extra prison time, or the other extreme, dangerous individuals being released into society. These findings draw attention to how important transparency is in judicial decision making when using AI, machine learning, algorithms, or other types of statistical models.
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
n criminal justice, there has been substantial concern about due process, and whether these secret algorithms are unfair. One such secret algorithm—called COMPAS—is widely-used across the justice system, and has been the subject of high-profile lawsuits. But what do secret algorithms like COMPAS actually compute?
The Age of Secrecy and Unfairness in Recidivism Prediction · Issue 2.1, Winter 2020
GRPO optimizes the model by evaluating groups of responses relative to one another.
The Math Behind DeepSeek: A Deep Dive into Group Relative Policy Optimization (GRPO) | by Sahin Ahmed, Data Scientist | Jan, 2025 | Medium