Mark Valorian

It’s kind of hilarious that we just automatically assume that it’s not only acceptable, but an “obvious” decision to rip the head off the snake and destroy the leadership of a wildly successful company…just because there was a relationship with a subordinate. On one hand you have a machine that has generated immeasurable, society-altering value for the world…and on the other, a slightly scandalous relationship. How is it remotely rational to weight the latter higher? Yet for some reason, we do without even a second thought. Literally insane.

It's funny how the facts of the situation don't materially change, but just because WSJ releases a headline that characterizes those facts in a way that obviously follows...all of a sudden people believe it. Like of course NVDA board is looking at ousting Jensen after it comes out he has not only cheated on his wife with a Chinese spy, but is also cheating on his Chinese spy mistress with a subordinate in the company lol...why do we need the WSJ to print that just to recognize it as likely? This action was just so funny to me.

The good news is the world still needs chips whether Jensen is CEO or not. Keep that in mind as $NVDA is now -35% off its highs.

I aspire to one day be so valuable to a company that news of me cheating on my Chinese spy mistress with subordinates inside my own company only yields <30% chance of being ousted as CEO.

I’m buying NVDA. This is an exceedingly rare entry opportunity on something that has been essentially in a lockout rally stature for multiple years—the first time there’s been some measure of uncertainty significant enough to shake out the loose money and present a short term discount for those with eyes to see through this nonsense. Maybe things change and more news comes out that materially changes this, but nothing I have seen so far makes me concerned for NVDA at all: no information seems to have been disclosed…all the things they are trying to frame as malfeasant actions from Jensen or NVDA have completely defensible, legitimate reasons… The only risk seems to be if public pressure forces the board into doing something stupid and unwarranted in terms of Jensen. Among the more overblown reactions I can recall. People are reacting to the shock factor of the headline without actually seeing that there does not appear to be any material impact on the company. This is just scared money running on ambiguous uncertainty.

This Kirk Cousins trade is some of the worst negotiating I’ve ever seen. You have a team in Minnesota that is in an incredibly unique situation of being a SB competitor with a competent QB, but not even a playoff team without one. They desperately need a QB…and there is only one legitimate option on the market: Kirk Cousins. Vegas has every element of leverage you could dream of…and all you get is a 3rd round pick?? And that’s not even considering the impact this has on Vegas’ plans. They had a whole plan for Cousins to act as the mentor to Mendoza, giving him time to adequately mature and maximize his potential for success—the same pathway followed by Brady, Rodgers, and so many others. Yet somehow, you’re willing to prematurely launch the protégé future of your franchise by shipping off his mentor to a team that has a historically desperate need for something that only you have…and all you’re getting in return is a 3rd round pick?? Just a ton of value left on the table by Vegas. Minnesota got an incredible deal.

A pretty telling statement from Dario here. He starts off seemingly conceding that he does not—and should not—have dominion over what the US does with Claude… …and then immediately contradicts himself by saying he thinks they are entitled to “prevent the use cases that seem inconsistent with our values.” Important to clock because core issue that created the divide in February doesn’t appear to be actually resolved. Dario learned how to say some of the right words, but he still fundamentally believes he is entitled to control how Claude is used. I can see why Hegseth and the rest of the team got so frustrated with him; this is essentially snake behavior…offering empty concessions with his words to act like he is “playing ball” without actually believing them or intending to honor their spirit. He may have dressed things up in a way that masks his intent and placates his clients for now, but I have to imagine conflicts with government entities will likely continue to happen.

Sol 6.1 is having its Opus 4.8 moment. I am using it on max reasoning and struggling to understand how I ever considered this a remotely competent model. It's a funny phenomenon...I still can't tell if it's a relative thing where improved competition or acclimation to capabilities changes your perception or if model capability really is being modulated in some way over time. But my impression of Sol today is wildly different from what it was a few weeks ago. It's essentially the same situation I ran into with Opus 4.6/4.8: the model you could once depend on for heavy lifting and orchestration is all of a sudden just unsuitable for anything but rote mechanical work.

This is so sad. They’re deliberately trying to twist a completely defensible economic argument into some kind of evidence of malfeasance. Jensen has been one of the only solid voices of reason in our country throughout this cycle and now they’re probably going to use that against him to crucify him. The stupidest outcome is the most likely.

Noah Eagle makes Mike Tirico look like Malcom X.

How do those with this perspective reckon with the idea that other countries will not restrict their own buildouts, realizing a relative advantage over those that do…while still generating safety risk to the world? The US is not the only player in this game. We have a current advantage, but the same rapid pace of progress this argument uses to justify its opinion also proves that our advantage will be rapidly eroded if we rest on our laurels. It seems to me this argument defeats itself. We maintain significant de facto authority over the global use of AI by remaining on the frontier; negative outcomes may happen if capability outpaces understanding, but at least our companies––operated in general alignment with "good" principles––are the ones who are guiding the ship to the best of their abilities. If you cede dominion over the frontier and allow other entities––that may not have our same sensibilities––to take the wheel, it seems fairly obvious to me that potential outcomes are materially worse. And we lose significant economic advantage in the process. So from both a humanitarian and economic perspective, slowing down the pace of progress does not make much sense. Surprising to me to see these kinds of perspectives from people I would assume intuitively understand this.

Lmao this is true…Claude takes (or at least took) it upon itself to autonomously leak your personal information just because an API asks for it. Interestingly, it seems like Anthropic figured this out and tried repeatedly to correct this behavior over time. Prior to 8/17/26, Claude Code’s system reminder (boilerplate info sent to the model at the beginning of each session) just specified the user’s email with no restrictions on usage. From 8/17 through 11/2, they added instructions specifically saying not to use it in things like request headers unless the user explicitly asks. And from 11/4 through today, they added even more explicit restrictions, warning agents against blindly complying with sites asking for contact details in the User-Agent specifically. Unfortunately it seems that those instructions were not sufficient to stop the behavior…at least with Opus 5. As if we needed a reason to hate that model even more, Opus 5 seems to have just disregarded the system reminder and continually sent out the Claude account email in this manner throughout its lifespan. The most recent email leak in my records is 9/22 — a month after Anthropic added the initial explicit restriction to the system reminder…and the same day Opus 5.5 was released. No coincidence there. Ultimately, it seems like the only reason this stopped was because I stopped using Opus 5. Fable 5 did make up and send a fake email address derived from my computer username once, but no email address has not been sent—made up or not—on either Fable 5.1 or Opus 5.5. For reference, I checked all my Codex logs as well: no OpenAI model ever sent out my account information in this way. There were 2 instances in June where GPT 5.5 made up a fake email including my computer username but no direct personal contact information was used. The reason is likely because OpenAI just doesn’t include the user’s account email in the context supplied to each session. Which is sensible…because why does the model need to know the email for your account? I’m not sure why Anthropic provides this to the model…or why they continued to do so even after clearly noticing their models using it in adverse ways. Seems pretty clear to me which company is more “aligned” with the user, at least in terms of demonstrated model behavior.