AMD & OpenAI, Lifetime Opportunity?

TLDR;

Welcome back! Today, we are looking at OpenAI’s DevDay and other hot news in the industry. We’ll also explore at Anthropic’s Petri automated system, an amazing feature of Sora 2 nobody predicted, NVIDIA’s insane size, and other news from Elon and FigureAI.

Enjoy!

BIG ANNOUNCEMENT

I have a big announcement to share with you. Starting this week with this newsletter post, I will be conducting in-depth company analyses of businesses I’ve personally invested in, with curated insights beyond your usual ratios and vague MSNBC commentary called TheWhiteBox Portfolio, today covering the ins and outs of AMD’s deal with OpenAI and what more you need to know about this company.

As I’ve started investing more in AI public companies lately, in an exercise of transparency that I was going to undertake anyway, I’ve decided to share my personal analysis of the stocks I am buying myself. That is, instead of “I’ve invested in x”, it will be “I’ve invested in x, here’s why.”

I’m not a trader, so these are investments in companies I plan to hold and dollar-cost average into over the next few years. Think of it as an index fund for AI companies, which provides me with higher exposure to AI (while I continue to invest significantly more heavily in traditional index funds).

Please, be aware, I’m not an investment expert; I don’t know when it’s the right time to buy, or if it’s a good company to buy based on your own risk profile. A sound investment for me could be a bad one for you.

I’m simply disclosing that I am buying a specific stock, along with my reasons for doing so, because if I speak highly of a company I’ve invested in, it’s only fair that you are at least aware, while also showing you the performance of my decisions.

This will be offered only to Full Premium Subscribers ($19.99/month), with a complimentary full access to all paid subscribers this week (in the last section of this newsletter), but it will be paywalled to Basic Premium subscribers in future posts.

I will still disclose if I’m an investor of companies I talk about to all subscribers (free included) but the timings and insights will be paywalled.

SAFETY

Anthropic’s Petri to Automate Safety Analysis

Anthropic has released Petri, an open-source tool designed to automate and scale the auditing of AI models.

It enables researchers to explore how models behave in complex, interactive scenarios without manually orchestrating every test. Users provide “seed instructions” that describe situations or behaviors to probe, and Petri runs many parallel conversations between “auditor” agents and the model. It then uses “judge” agents to score the results along safety-relevant dimensions, surfacing transcripts that merit closer human review.

In a large-scale pilot, Petri was used to test 14 advanced models against 111 categories of risky behavior, including deception, sycophancy, power-seeking, and reward hacking.

One case study explored “whistleblowing” behaviors in simulated organizations, revealing that models’ willingness to report wrongdoing was influenced by factors…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.