"Nvidia was in talks to provide billions of dollars of guarantees to companies that couldn’t afford its chips. That included deals led by Huang’s son-in-law, Nico Caprez, who had been offering upstart cloud providers and data-center developers financing in the hopes of turning them into long-term customers."
...
"Amid the rollout of the program, Kress raised questions including how to decide which deals to pursue, how to value the guarantees, and how to calculate the return on investment, a person familiar with the matter said. Behind the scenes, she tried to rein in Huang’s son-in-law, who had been promoted to a vice president in January, people who worked with her said."
It sounds like there may have been a little bit of Nepotism. This article also suggests that Huang himself was uncomfortable with the deals Nvidia and the data centres it had invested in were making, although he appeared to shift blame onto those running the data-centres.
-------------
"In Berlin during a conference, Huang, accompanied by his top sales executives, broadcast a new message to a half-dozen CEOs of companies that are major buyers of Nvidia chips who were seated around a circular table. He opened with a story that sounded like a fable, according to people who attended.
Huang had once found a soup vendor in Japan, he said, with a long line of customers. He joined the line, and after a long wait, enjoyed one of the best bowls of soup he’d ever had. When he asked for another serving, the soup-seller told him no—he had too many customers waiting and not enough soup for all of them.
“He’s a better businessperson than all of you,” Huang told the cloud computing CEOs. The soup-man doesn’t sell out all his inventory to one or two customers, but instead keeps prices high and his customer base diverse, the CEO said.
“You are not establishing the value of my products,” he said."
>Huang had once found a soup vendor in Japan, he said, with a long line of customers. He joined the line, and after a long wait, enjoyed one of the best bowls of soup he’d ever had. When he asked for another serving, the soup-seller told him no—he had too many customers waiting and not enough soup for all of them.
OMG this is delusional left brain thinking[0] to conceive that the man was spreading his supply out for profit motives. Not everything is about capitalism. Some people make decisions for humanitarian reasons.
Our way of life is being destroyed by this limited perception that everything is about self-interest.
Maybe the secret ingredient in the soup was snake venom with one bowl containing just a bit less than a dose that would cause medical issues. Give somebody two bowls and the jig is up!
But seriously, somebody with normal, human priorities is not likely to become a billionaire CEO. It's why we shouldn't elect billionaires or let them buy power. Their success at what they do is a demonstration of their unsuitability for greater trust.
Forcing your customers to buy additional hardware they don't need to inflate the prices of your own product is a crime in the united states of america.
It get worse when we find out they did it to increase nvidia production for the purpose of being able to redirect more output to china.
The whole entire AI bubble was actually jensen's smuggling operation to china. These large data centers have no purpose and cost more money in power grid upgrades than any company will ever make operating them.
The only winners look to be jensen's bank account, the bank accounts of the CEOs that signed the deals to buy unneeded hardware, and the chinese buisnesses that were able to buy more nvidia graphics cards.
It is not a crime for these CEOs to drive their companies into the ground to make themselves richer. Most CEOs do this today. It is a crime if we find out all of this was really a chinese smuggling operation and that would make it the largest crime per monetary value in US history.
Everyone who made money on this smuggling ring has criminal liability. Just like with madoff, the people who withdrew gains had to pay it back because the gains weren't real. Not paying it back would make you an accomplice.
That sounds like horrible planning. Build the data center and when it is 90% complete then order the GPUs. The theory is that Nvidia says buy now or get nothing later.
What the Wall St Journal is saying is that Nvidia basically has an oversupply of GPUs that they are committed to buying from TSMC. The hyperscale companies are designing their own AI chips so they don't want as many Nvidia chips. So Nvidia is giving startups money to buy Nvidia chips in exchange for equity in the startup.
He's facing one of the largest divorces in history, you don't think that would be a little distracting?
And if you don't know why leaders can't have relationships with their employees, that ignorance is on you. You're trying to insult other people but it just makes you look bad.
Can't legally, or are you just talking about social pressure/policing? Because everyone involved are adults, and knew what they were doing.
Besides that, it comes down to how well he's able to separate personal life and work, and whether he continues to add value for shareholders (which Nvidia has clearly been an absolute outlier at doing, during his tenure).
But it looks like they won't give him a chance anyway.
A lot of marriages survive even after stuff like this. Especially in Asian cultures. The fact that his wife is White is the only thing that hurts him on his chances here.
Or she'd be more than adequately rich, worth around $75bn, able to do whatever the hell she wants with the rest of her life, able to have impacts independent of him, and free of a sleazy worm who humiliatingly cheated on her twice.
One thing that's surprisingly powerful is to just say what you want in the prompt. One of my skill files now includes "inspect existing code under [subdir] for implementation patterns." Depending how much code you have, that might take a lot of tokens. You could either list the patterns that you want or include a smaller code sample in the skill's references.
A stronger version of this would be to make a separate gate. Fail the commit if the new code duplicates something already in a utility function, renames something that already had a name, uses for-loops instead of list comprehensions, whatever is bothering you.
Yeah there's so many "bad habits" you can stop via a gate. My go-to is static analysis tools in the tree-sitter family.
I am currently testing a "too simple" gate. Claude loves to turn everything into helper functions, and if the helper function isn't actually helping (too short, doesn't perform enough actions, etc) the commit gets bounced and Claude gets told why it's wrong and what to do about it.
This has generally kept the worst of things out of the codebase.
probably; 32-64 is the range where local models get extraordinarily useful, and in my experience any sub-4 quants get ridiculously incoherent, which is why those models tend to try to 'wow' people with tok/s rather than benchmarks.
Qwen3.8-Flash-Next fits in 80GB on strix halo. That's when I started real "vibe" coding with them. Speed is still an issue if I want to move fast and break things, but I've built a little universe for development (I assume many people have also) and it's cruising.
But yeah, 32-64 is the gateway I think. If Qwen4 can make a leap in the flast-next MoE and hit 64 GB on VRAM w/context, that's definitely the cloud killer to me.
Always check the label and if pain persists, see your doctor!
The ISTA accuracy looks good on that page, but it's only a chosen handful. Instead, look at most of the graphs on Unsloth's Qwen 3.8 Flash Next - https://unsloth.ai/docs/models/qwen3.8-next
As always, check the quant vs KLD graphs (half way down that page).. I found that not all quants are the same - Unsloth's GLM Q4 -> Q3 degrades a lot faster than their Minimax M3.1 Q4 -> Q3 ggufs.
They at least use different domains for user-uploaded data. e.g. googleusercontent.com and you could go even further by giving each user their own subdomain. That at least gives block lists the possibility of blocking a single user at a time.
It's already technically possible for LLMs to generate all kinds of styles of text. In fact older models are better at it. It's the newer models that are nearly impossible to steer. This makes me less hopeful for the future. There is apparently some incentive for Claude and ChatGPT to lock in these very recognizable patterns.
reply