Posts

In the past, the bigger the AI model, the better the performance. Across OpenAI’s models for example, parameters have grown by 1000x+ & performance has nearly tripled.

But model performance will soon asymptote - at least on this metric.

Every successful startup faces critical marketing decisions from day one. How do you establish your value proposition? Which customers should you target first? What channels deliver the best ROI with limited resources?

During a recent session with a former executive Kady Srinivasan from Lightspeed Commerce, Dropbox, and Klaviyo, several powerful insights emerged on how early-stage companies should approach marketing. Here are three key takeaways:

1. Define Your ICP Before Anything Else

The most fundamental decision for any startup is determining your Ideal Customer Profile (ICP). Without clarity on exactly who you’re targeting, marketing efforts become diluted and ineffective.

Last week, Reddit filed their S-1 to go public. At least 10% of their revenue - about $60m - comes from selling data to train Large Language Models. Reddit’s data sales revenue will likely be much more than 10% by the end of the year. Quoting directly :

We expect our growing data advantage and intellectual property to continue to be a key element in the training of future LLMs.

This raises a fundamental question : What if the revenue from data sales dwarfs the revenue from ads?

The Community Source licenses may be the future of open source software.

With closed-source software, the code isn’t shared outside the company that wrote it. Open-source code is freely available to examine & use. Open-source companies like Elastic ($13b), Confluent ($10b), MongoDB ($33b) have been phenomenally successful.

Palo Alto Networks, the largest security software company in the world, worth roughly $82b, announced earnings this week.

There are three interesting themes :

1. AI is a big business for them already. Internally, the efficiency gains are impressive.
2. They’ve embarked on a platform strategy 5 years ago & there are some positives but also some significant challenges.
3. The company is buying out customer contracts for six months to win over new business.

They have developed a $100m ARR security line in a few quarters :

Asking “What problems do blockchains solve?” is like asking “What problems does steel solve over, say, wood?” Blockchain networks are a new construction material for building a better internet.

This section in Read Write Own, Chris Dixon’s book, has been bouncing in my brain for the last few weeks.

Very few people know whether today’s apps are built with, just as they don’t consider the construction materials of their office building.

The Bitcoin exchange-traded funds (ETFs) launched on January 12, approximately one month ago. On the first day, investors bought over $655 million and nearly $2 billion within the first three days. Since then, the figure has grown to an impressive $4.6 billion.

It’s not a fair comparison but for fun, we can compare the Bitcoin interest to the largest technology IPOs for a sense of scale.

On February 23rd at 10am Pacific time, Office Hours will host Steven Goldfeder. Steven founded Offchain Labs, the company behind Arbitrum, a web3 project worth $21b as of this writing.

Arbitrum is a L2, a layer 2, that sits atop Ethereum, improving its performance & cost to write transactions.

Steven authored many of the seminal papers behind modern blockchains.

We’re entering a new pricing environment for software: AI vs non-AI. It’s only happened in the last few weeks.

Recent earnings have pushed some of the most important companies to all-time highs.

This dynamic doesn’t favor everyone.

Theory created a public market AI index to track software companies who have significant product plans or current AI businesses.

As we’ve been researching the AI landscape & how to build applications, a few design patterns are emerging for AI products.

These design patterns are simple mental models. They help us understand how builders are engineering AI applications today & which components may be important in the future.

The first design pattern is the AI query router. A user inputs a query, that query is sent to a router, which is a classifier that categorizes the input.