Jensen Huang says Nvidia engineers should use AI tokens worth half their annual salary every year to be fully productive
You're not productive if you don't use a lot of AI, says guy who makes all of his money selling AI hardware
https://www.tomshardware.com/tech-industry/artificial-intelligence/jensen-huang-says-nvidia-engineers-should-use-ai-tokens-worth-half-their-annual-salary-every-year-to-be-fully-productive-compares-not-using-ai-to-using-paper-and-pencil-for-designing-chipsOpen linkView original on piefed.social542
Comments124
They really want this to be an apt comparison and it's really not
Edit
They also really want to talk about tokens like they're some kind of currency
It's worse, they want to change the global economy to corporations paying corporations...
The total elimation of actual consumers, because none of us will be able to afford to consume enough.
AI companies need people to pay for AI to keep buying Nvidia chips. So Nvidia is making their employees pay for the AI so AI companies keep buying Nvidia chips.
It's not a sustainable system, it's just a money churn whose only purpose is to consolidate wealth.
I'm hungry. When do we eat?
That’s the neat part!
So ponzi scheme?
I didn't read it that way. I think he's saying "bosses: if you're paying a $100k salary to a dev and not also paying $50k for tokens, your dev isn't working hard enough". Which is better, but only just.
To refine that even further, he doesn't appear to imply that the dev isn't WORKING hard enough, only that they're not being OPTIMALLY PRODUCTIVE.
What he's trying to do, really is float and normalize the concept of baking tokens into HR math in terms of a "golden ratio"... which happens to be 2:1.
So, when a company goes all in on ai, and they cut thier workforce in half, they'll need to add in 50% for tokens. 50% of the original staff, at a new 150% cost, puts the company at 75% pre ai workforce cost. This is the "guidelines" they're trying to normalize.
It's funny how his calculation factors in the completely immaterial price of tokens variable instead the material one which is the number of tokens or better yet the productivity gain per token.
Huh, yeah. 50% cost in tokens could be 1 very expensive token or 1 million tokens. Who knows?
I think he's saying slaves should owe their soul to the company store.
Don't give this sack of shite the benefit of doubt.
he needs new jackets.
A straight white one
He didn't explicitly say it, but the language in use gels with the ones that have
With my trusty LLM, I follow the steps recommending that I try reaching inside the die press to look for any jammed parts that could have caused the machine to suddenly stop working. My coworker, who my boss sent to assist me based on instructions from her LLM, asks his LLM how to help me. My coworker's LLM recommends that he check if the emergency stop button has been pulled...
i think it is better than pizza fridays for them, because they probably can't barter pizza for free.
Just a few days ago they were indeed talking about giving tokens as bonus.
Is company chit still illegal if it’s a bonus?
That's one the most insanely stupid things I've ever heard. Tokens are a tool used to do your job, a business expense, not fucking compensation.
Are the kWh that you use to light and air condition an office part of your compensation? How about toner and paper in the office printer?
If those tokens are a bonus, they're yours, right? So you can burn half your annual salary worth of tokens translating Microsoft Encarta 95 into Klingon, and they will foot the bill?
The tokens as currency thing feels exactly like gaming microtransaction bullshit to me. Obscure the true cost of each purchase by selling in-game currency. Dark patterns lead to higher spending.
It's proof of work cryptocurrency. But at least this time the compute is aiming at producing something useful, not the lowest hash value.
Sure I don’t get any retirement or healthcare benefits, but look at all the company scrip I get for the sloppy autocomplete that is stealing all my groundwater no matter how many times someone says „closed loop cooling“.
This man went from high to low REAL fast.
The billionaire's way
The secret is he's always been low, nvidia has intermittently pulled tons of shady shit basically ever since the 3dfx days
What did they do?
Closed source linux drivers is not nice, but also not shady.
Idk when he was every regarded highly by Nvidia users/consumers. Only by corporate board and investors
ifs hes in the news this much like altman, they seem very DESPERATE like the bubbles about to pop.
It would take a great burden from my shoulders, if AI could wash my clothes, do the dishes, cook, and vacuum my apartment. Those are the things I don't want to do. I like my job. I can do it myself. Yeah, sure, there are occasionally things that AI can do for me in my daily work, but it's mostly stupid, repetitive shit. And then I have to review it, which sucks even harder.
So I found it kind of sucks to have those daily life tasks removed. I worked in a place early in my career that had on site catered food, on site laundry, on site massage, shopping services (before Amazon prime) and lodging for excessive overtime .
If it’s just leaving you more opportunity to work it sucks. I can’t be the only one who enjoys taking care of the daily life tasks Lê cleaning, shopping and cooking for myself. It’s what personalizes your life. When you do it for yourself, you decide what you eat and what you wear.
It seems cool but it’s offered so you never stop working and your life is never really your own.
I think the issue is that those things were tied to your job, not to some external source that just gives you free time to fuck around.
Yes.
Managers should send 100 emails per day to be fully productive
100 AI-written emails per day
And also have their AIs read another 100 emails for them.
and dictate to them with AI generated voice.
Huang wants employees to pay half of their salary back to the company ? There is no escaping the company store, eh ?
And CAD designers didn't have to pay half of their salary to use the CAD tools the company wanted them to use. How did that guy get to be in charge ?
You code 16 lines, what do you get?
Another day older and deeper in debt
St. Peter, don't you call me 'cause I can′t go
I owe my soul to the company store
I didn't read it that way. I think he's saying "bosses: if you're paying a $100k salary to a dev and not also paying $50k for tokens, your dev isn't working hard enough".
Which is still utterly absurd, because it implies that a harder working dev would be spending more time chatting to a bot.
oh absolutely, but generating tokens is how nvidia gets paid, and companies are terrified of being left behind if they're not 100% onboard with LLM workflows
This is clearly a situation where you are expected to spend 100% of a budget to keep your job.
Your didn't read the article did you?
Jenson gives then free tokens just in case you don't want to read.
Fuck reading.
That is not at all what the article says. It’s way more self serving than that.
"Dear AI coding agent, write for me a 10,000 page manifesto on the downsides of assigning performance metrics to employees unrelated to their actual work product. Populate it with generated images of Nvidia's CEO getting railed by a bunch of copyright lawyers in the style of a Studio Ghibli film. Please ensure every fifth sentence rhymes with orange. Continue to generate images and short videos of Jensen Huang licking shit off the floor of a 7-11 rest stop bathroom until you have used enough tokens to meet my salary target."
Beautifully prompted
I didn’t read this and just pasted this into a llm
CEO suggests raising employee costs by fifty percent and is immediately fired.
Sorry, we don't live in a sane world anymore.
They've all realized there's no point in even bothering to hide the grift anymore. People don't care.
Sam Altman floated paying employees in chatGPT tokens, that surely their landlord will accept. The main motive would be being broke. Nvidia is not broke, but it is weird as fuck to be comparing it to salary, if it doesn't segue into the OpenAI plan. The $500k salary example would seem to generate more tokens than you could have time to skim over in review.
He's sooo close to arriving at crypto tokens.
uh oh... if only GPUs could do proof of work mining, if the AI thing gets oversupplied.
BREAKING NEWS: Guy who sells AI hardware says his employees must use more AI.
Luigi knows the solution for this.
It's moments like these that make me think about the state of the world and my part in it. I may just be a random loser on the Internet, but I do know a lot more shit that some of the biggest multi-quad-spillion-dollar CEOs, apparently.
For example, it's an old fact that tech CEOs know jack shit about measuring productivity, even when they're obsessed with it. Yeah. One more example.
It’s amazing they can just make these claims without literally any evidence and no major “media” organization asks for it. They are just propaganda for these companies.
Something I don't understand - AI coding is mostly useful in common code, snippets, easy stuff. What Nvidia is doing (drivers, optimization, chip design, etc.) is something I imagine there is close to zero AI training, so what can they realistically even use it for so much?
Burn money on AI tokens so it looks like AI could be profitable some day so people keep investing in AI companies that can then buy Nvidia chips...
You're thinking of it like "how can AI make a better product"
They're looking at it as "how can we sell more chips"
Two very different questions with very different answers.
It's a house of cards and Nvidia can't afford to acknowledge no one wants AI or knows how to make it profitable.
I’m gonna take a guess that a big portion of it is infrastructure-as-code, the operations side and not product development itself. I work in the operations side of things and we never touch the product at all, but we deal with a lot of code due to how backend infrastructure is built and maintained now, especially if you're in the cloud.
Guess code review and troubleshooting. Not really sure, I have only really used it for code templates and ideas for troubleshooting to look into.
The most use I found is rewriting documents in a specific way. But only after I write it first. Then go back and forth. Just to make tone consistent.
There's plenty of driver code available. All of Linux and BSD, plus whatever internal stuff they have. Optimization is pretty generic.
Chip design maybe not, but I imagine you can train an AI on the principles and generate a bunch of candidates, then benchmark them in simulation.
I'd say not before the bosses and shareholders agree to take 50% of their compensation in the form of AI tokens.
These fucking leeches...
Oh this will be a fun. "Please spend more of our company money, or else we will think you are not doing your job"
it's so fucking easy to send bogus requests that use tons of processing time. This is the DUMBEST metric i've seen in a long, long time. If an employee of theirs uses 500k worth of processing time now, are they better employees? What if someone manages to spend 10 million worth of tokens? 100 million? Performance metrics should be outcome based instead of "how much money did you spend" lol
It's not a metric it's a marketing stunt.
I'd be amazed if they internally do not have objectives to implement the kool aid.
But yeah, of course it's propaganda. If this bite lands in the right ear of an exec, they can implement this as a metric in their review process.
Kinda similar to musks LOC requirement. But at least with NVIDIA this idea directly correlates with more business if it spreads beyond the company
Can't wait to to see AI that designs chips driving Nvidia out of business
They're allegedly already vibe coding drivers so it's bound to happen eventually
You can't vibe code IC design. Look at the number of humans in the world that design advanced ICs. It's a really low number for a highly specialized skill. Layout requires spacial awareness at a high level, which an LLM does not have.
Noone talks about LLM here. I'm talking about building new specialized model for that
Let me translate this for you, "My bonus depends on you showing our massive investment wasn't a waste so I'm holding your jobs hostage until you make up busy work to pretend it was worthwhile."
This is even dumber than tracking how many lines of code each dev commits.
The AI bubble went from a trillion dollar worth of circlejerk investments, to a self-sucking circlejerk trillion dollar ponzi scheme. Jensen Huang is right the moment an Invidia's employee stop sucking their own dick or stop jerking off their coworkers and bosses simultaneously, the whole economy collapse.
Highly paid asshole says shit and instead of ignoring it gets all the attention.
Jesus that's a lot of tokens.
Even if I was trying to do everything in my power to make burn tokens for real work I'd be hard fucking pressed to burn more than a couple thousand a month and that's just being wasteful.
you just need to use more context injection and more agents working in parallel on git worktrees or whatever and then get really depressed because you own an absolute fuckload of code now and you are less familiar with it
can I tell an AI to burn exactly that on Jan 1 doing absolutely nothing, and then never touch the shit for the rest of the year?
Openclaw got you covered!
Wouldn't an AI researcher naturally find generative AI disadvantageous because they are attempting to develop novel tools which could not exist in the training set in the first place?
Even novel solutions are usually built out of smaller common building blocks. E.g. many novel solutions surely use a database. You can make the LLM help you set up and use the database, that your novel solution uses.
“No we don’t need databases anymore, only blockchains.” —Nvidia CEO a few years ago
And AI can help you migrate your database solutions to blockchain, utilizing 3000W worth of Nvidia co-processing power to validate your blockchain database that used to work on a 0.3W ARM processor.
The database was an arbitrary example. A more relevant example would be tenserflow layers in a neural network. As I understand it, you can in some cases get a novel solution to a problem just by choosing a smart enough combination, with the right data.
ChatGPT absolutely knows how to help doing the grunt work setting up the tenserflow configuration, following your directions.
Smart, lucky, who can tell the difference?
If used by an expect developer, then the combinations are not just random "lucky" choices.
Or, if you take the machine learning approach, you just try all the combinations and use the one(s) that perform the best.
The world is not that simple. There are too many combinations to try. And you risk hitting local maxima, even if doing the gradient thing.
If you are capable of giving good directions...
I'm probably not arguing with you, and I'm not trying to regardless. You seem like you have tried this, watched it happen, go "huh, neet!" And then get it to take the next step in whatever you were doing in the first place only to find out you didn't provide adequate requirements for your config.
Well, yes, that is a central point.
I am a senior programmer. LLMs are amazing - I know exactly what I want, and I can ask for it and review it. My productivity has gone up at least 3-fold, with no decrease in quality, by using LLMs responsibly.
But it seems to me that some people on social media just can't imagine using LLMs in this way. They just imagine that all LLM usage is vibe coding, using the output without understanding or review. Obviously you are very unlikely to create any fundamentally new solutions if you only use LLMs that way.
Senior programmer. I know exactly what I want. My requirement communicated to the LLM are precise and adequate.
What I find LLMs doing for my software development is filling in the gaps. Thorough documented requirements coverage, unit test coverage, traceability, oh you want a step by step test procedure covering every requirement? No problem. Installer scripts and instructions. Especially the stuff we NEVER did back in the late 1980s/early 1990s LLMs are really good at all of that.
Nothing they produce seems 100% good to go on the first pass. It always benefits from / usually requires multiple refinements which are a combination of filling in missing specifications, clarifying specifications which have been misunderstood, and occasionally instructing it in precisely how something is expected to be done.
A year ago, I was frustrated by having to repeat these specific refinement instructions on every new phase of a project - the LLM coding systems have significantly improved since then, much better "MEMORY.md" and similar capturing the important things so they don't need to be repeated ALL THE TIME.
On the other hand, they still have their limits and in a larger recent project I have had to constantly redirect the agents to stop hardcoding every solution and make the solution data driven from a database.
I were simply unable to convince Codex to split a patch into separate git commits in a meaningful way. There are things that just doesn't work.
Still useful for lots of stuff. Just don't use it blind.
Yes, this is why I point it out. I agree with you, but no part of this is actually common sense. It just feels like it.
Every software development project, ever.
Review your requirements before starting development. Review them again after each phase of development. Address inadequacies, conflicts, ambiguities whenever you find them.
AI is actually helpful in this process - not so much knowing what to choose to do, but pointing out the gaps and contradictions it can be helpful.
That’s fair. I guess it could be no different than a scientist with some grand scheme handing his plans off to others to implement.
I think I was assuming that cutting edge AI research involves more math/theory than just… bootstrapping existing tech stacks and tweaking configs.
The beauty is when those companies run out of human training data and start training on AI slop, just to generate even more AI slop.
This is probably already happening though.
It is, intentionally. Some of the training data is synthetic
Just have an LLM ask another LLM questions for a couple hours a day. That should hit that target.
This sounds stupid (and it absolutely is!), but if you implemented it, Facebook would hire you. They hired the last guy that did this.
I remember back when Nvidias PR team used to push this humble rags to riches story about Jensen back in the day, I guess even they would have a tough time doing that now that he's gone mask off
I thought AI would save us money? If it adds 50% onto the cost of a salary and bynall studies does not improve productivity output, then it's not great.
Oh, he left out their plan to fire half of them and drive wages down by a third.
This may sound weird, but I think anecdotal evidence might be more informative than the productivity stats for now, until the industry settles on a new equilibrium.
Some engineers are more productive with AI, and some (maybe even most, still) are less productive. People are still putting in the effort to learn how to use it more effectively/productively (there is a learning curve), and some of the less productive are getting laid off.
It sucks, but that's just how it is now.
Also, AI tooling is still evolving very rapidly. A lot of information and stats are only valid for maybe a few months.
Wow, that doesn't sound like a pyramid scheme at all. At all.
No shit, CEO of company producing AI hardware wants everyone to use their hardware more. 🤷♂️
because they know the AI is about to go bust, and its not generating any actual value.
Jensen Huang should suck my dick.
Only if he has the tokens to do so.
you first, mr huang
What a fucking clown
If these AI engineers were fully committed to AI, then they would be happy to be paid in tokens. The fact that they still demand to be paid in human money proves that they are just philistines and sycophants.
What do you have against philistines?
Tech people are all circle jerking dipshits
Mostly corporate tech people, or really suits in general.
Jensen in particular has some serious craniorectal impaction asking with autofecal anosmia.
I want him to explain where the half salary number comes from, specifically and in detail backed up with analysis from real life data. Also I will still bust out paper and pen for graphs, layouts and design ideas because that's how I figure out what works and what really doesn't work before I waste a bunch of time with a PC. Incidentally I use zero AI in my job, no matter how much they want us to.
Would Nvidia lose a bunch of money if the ai bubble collapsed? Could that be why he is saying absurd things like this? I bet if any employee did drop that much on ai they would be in trouble.
"And to not get mysteriously laid off"
I mean... if you take into account fair price of utilities (water and electric) and skyrocketing costs of RAM... I don't fully disagree.
Ballparking, but 50% of an engineer's job being "boilerplate" and "charlie work" sounds reasonable to me. Maybe even 75%. That IS what generative AI is really good at. It is the kind of work that you can have a particularly ambitious student intern do.
What makes a good engineer is the ability to verify that work. In software this is "code review". And the other aspect is actual innovation. Solving particularly complex problems or breaking a problem down into manageable and verifiable tasks.
And... guess how you maintain your skills to be able to do that? That's right. Charlie Work.
Which is the problem. Managers (and wannabe managers) just see short term gains. So they want EVERYTHING to be done with "AI" because they want to bulk fire people and reduce operating costs. And they don't at all care that they are causing a massive brain drain because that is next quarter's problem.
But yeah. Use generative AI to accelerate your workflow. But also understand when it is very much worth taking your time. Either to keep those skills fresh or just because you can do it in a "fun" way. And if you feel that ALL of your workload can be done by chatgpt... maybe think about how much cheaper a claude subscription is compared to your salary and benefits.
Aww yea bitches digital scrip
Nothing like a wage slave
I think that's far too low tbh