Peter Kao

This is the first publicly-disclosed LLM prompt used by a major Wall Street firm to generate stock predictions. Only made available due to JPMorgan's recent patent filing for its AI stock picking system.

The best part of my day is picking up my kid from daycare :)

Ok so maybe AI can't replace a senior developer yet

Do you know how I know Meta juices their numbers? You can't unsubscribe to newsletters as a dev. The unsub link is there, the confirmation message is there, but they just never process it. This is a multi-billion dollar company and so I refuse to think this isn't intentional.

Alright, it took a couple of years but I know what my next project is going to be. Excited for the future 🚀

Anthropic is actually a better fit for Microsoft considering they lean more towards enterprise than OpenAI.

As a consequence of being in the commodities supercycle, countries around the world are going to see street crime tick up as copper, platinum, palladium skyrocket in value.

I should have gone to VRIC instead of VR conferences 😔

Prediction markets can be gamed easily according to www.jbecker.dev/research/pre... Just place limit orders and let others trade into you. Sell 'Yes' since people overpay for them. Focus on high-emotion, low-expertise categories like sports, entertainment, crypto.

Wealthsimple is the closest thing to a super app in Canada.

AI diagnosis tools are going to be crucial in debugging AI-controlled robots. Robots process billions of parameters that human technicians aren't able to diagnose. Someone work on this, stat.

Video clips are today's pamphlets.

Content creators and online personalities have started sharing their portfolios' performances for 2025. Seeing them barely beat the S&P500 (and with so much effort) while being so confident in their posts strengthens my conviction that you have to do the research yourself.

I've always wondered: how well do venture capitalists perform on the stock market?

When people ask entrepreneurs how many employees they have, what they're actually asking is how successful are you. This question will be outdated in a few years as smaller companies with higher revenues and lower profit margins (compared to non-AI software businesses) become the norm.

The 21st century play is turning out to be selling the materials needed to sell the shovels.

Could retail stock participation decline if tech weakens while precious metals and rare earths outperform? 🤔

Canada's economy is going to be saved by mining. It's a great time to look into shifting sectors in your portfolio if you haven't done so already.

Over a decade ago, I traveled the world as a digital nomad running an ecommerce business. If I were to do it again today, trading and investing in the stock market would probably earn me more money than ecommerce. Crazy times.

I've been "fact checked" by people using AI so many times now. It's crazy to me how people still don't understand how LLMs work. And it's even crazier to think how this will develop poor and intellectually lazy habits in people, en masse.

Gemini models still sound like a robot. If Google is after the consumer market, they need to make it sound more like OpenAI, or better, Anthropic's models.

AI will be the greatest bundler of our century.

No one's going to buy your product if you use poorly AI generated images like this. AI generation is fine but don't be lazy about it.

Wow Codex is actually better at programming than Claude Code right now. How fast things change!

Due to the "lost in the middle" issue with current LLMs, you should put the most important parts of a long message to LLMs at the beginning as per the research paper, "Lost in the Middle: How Language Models Use Long Contexts" (arxiv.org/abs/2307.03172 )

I get why @sama (Sam Altman) don't capitalize words anymore. LLMs don't care for capitalizations. When you use LLMs for so long, it starts to affect how you write.

Made this with my daughter over the weekend. Her face lit up when she realized she could bring her ideas to life with AI just by describing what she wanted to see. What a great way to teach kids how to be more detail-oriented.

Has the world gone insane? Headlines in my inbox today.

Favorite thing to do with my daughter right now: Creating AI generated images in the style of Cocomelon. It's a great way to teach her to think and describe things in detail while preparing her for the inevitable AI future.

One of my favorite standups. Happy Friday everyone

It's crazy how much Sonnet and Opus have declined in quality, and it only took one month. This is another friendly reminder to never pay for annual subscriptions when it comes to AI models.

Seeing how people write prompts on X (grok) is alarming. Some of these prompts are terribly framed with no context. If this is how the average person writes prompts, future chat interfaces for all AI should show a list of questions to ask next like predictive autocomplete on kb

The bane of my existence. I'm glad I won't have to deal with this anymore.

I had to bring out my Quest 3 to do a simple file transfer. Had to go through so many hurdles including: restarting PC, restarting Quest, reinstalling Sidequest, going into Quest settings, finding my ultra long USB-C cable, and on and on. In the end, I just gave up.

I'm shocked at how bad the iCloud dashboard looks. I had to check I wasn't being phished.

When doing competitor research, note down the order in which features are listed. If the competitor is well-managed, you can tell from the order which features are most important to their customers. This can guide your own dev priorities.

Do market research until it hurts. At first, market research is quite fun but the key is to do it and keep doing it even after it's become a dreadful experience. The dread and ordeal means you're seeing repeating patterns and nothing is new to you anymore. That's when you're done

"Bluesky now platform of choice for science community" Hell ya! arstechnica.com/science/2025... When will startup and VC communities move to Bluesky? It's a more natural fit than X.

I had put up a listing to sell something and someone asked me to take a photo of the thing but with a pencil on top to make sure I wasn't a scammer. Sometimes I forget I live in a tech bubble.

When I get stuck on picking company or product names, I remind myself that a company called Microsoft exists. Micro. Soft.

Beautifully put: "remember that a stock price is the price that the most optimistic person in the world is willing to pay at this second, not what the vast majority of us think it’s actually worth" - Random guy on reddit

Before building out a product, I like to gather as many emotive quotes from potential end users. I surprised myself today when I realized I was adding AI generated text (from my conversations with it) alongside human quotes. Not sure if this is a good idea but they seem useful

To use AI effectively, you have to be an excellent bullshit detector

Want to get better at writing prompts? Expand your vocabulary so you have more specific, targeted words in your arsenal. There's a massive difference in instructing a chef to make "good pizza" vs "charred pizza"

I'm going to start using the word "ultrathink" casually in conversations with random people going forward to gauge how knowledgeable they are on the current state of AI

How are people even using Playwright with Claude Desktop? I run out of tokens after only a few browser actions.

Watching Playwright (MCP) navigate websites autonomously makes me feel like I'm living in the future. Crazy stuff

I don't trust anyone who says "fam" when they're not talking about their family