Traster

There's two different things you're talking about here. There's the cost estimate- the amount you tell people up front a surgery is likely to cost, vs the actual cost to the hospital. You don't need to guess how many syringes of X you're going to need when you've already done the surgery. At that point you just addd up all the stuff you used, price it up and that's the BOM. Cuban is saying they don't even do that - they don't track their costs on a surgery by surgey basis at all. And because they don't track any of that it's impossible to reason about it.

It's impossible to answer "How much will it actually cost to insure person X" because you don't have any of the data on what the costs will be when person X needs a given surgery, all you have is the aggregate costs of the entire system - a lot of which is misleading because things are cross-subsidized because no one is really tracking costs. It may well be that whilst every surgery is billed equally in reality obese patience are responsible for 80% of the cost. Or it may even be that the hospital is making an average loss on hip surgeries because their negotiations with the insurer drove those prices down whilst brain surgeries give a nice profit margin.

And so you can't ask "How much would it cost for the government to fund service X" because you don't know how much it costs, all you know is the aggregate money spent across all medecine.

It's not difficult to define what curing cancer is, we've got tonnes of measures we can define. Let's define curing breast cancer - we have cured breast cancer if the 5 year survival rate is above 99% for all SEER stages, or less than 1 in 1000 women get diagnosed with TNM stage 1 or above. It's not that hard, it's just that deep down, they don't want to be judged on that basis.

(Details of the definition of these terms: https://www.cancer.org/cancer/types/breast-cancer/understand...)

This semi-analysis article reads a lot more like an OpenAI press release than a real analysis. And to be honest some of the statements seem like just straight up lies - they initially claim they were invited to benchmark it, and then half way down switch to claiming that OpenAI provided all the numbers. This really kind of sucks, because I want to read actual detailed nuanced and credible analysis of what's happening in the industry and it doesn't seem like you can trust this as far as you can throw it.

This is such a good example of a straight forward question that achieves the exact opposite of its purpose. What do we want? Highly ethical people that are mission focused and willing to give up on economic upside for the mission. What are we going to get? Sociopaths who are perfectly happy lying to your face about their intentions, or sycophants who are going to tell you what you want to hear.

Do you know who has the best answer to a moral quandry they've had to navigate during their career? Liars! That's a really difficult question to answer if you're honest, it's a really easy question to answer if you're dishonest.

If you really want to pursue this strategy rather than setting up bullshit interview questions you could do something much more straight forward. Agree some standard of living that's acceptable for your employees - let's say 250k in the bay area. Give them that as cash comp, and everything else goes into a "fund" which gets paid out when specific milestones are achieved, as judged by some panel. Cure cancer? You all get 10% of your deferred comp.

If you want your employees to be prioritizing things other than stock price you can literally just encode that into your compensation strategy if you like!

Obviously no one would do that, because they don't actually want to give up the comp.

Well clearly they choose, there's more demand for their GPUs than there is supply, so they have to decide somehow and the fact that the announcement of these deals frequently come with a plethora of weird funding/loan/guaranteed demand/exclusivity/bunlding, shows they're doing a lot more than just running a simple bidding process.

Can someone explain to me how Nvidia chooses who they sell chips to. As far as I can tell there are dozens of companies out there trying to build out these data centres; all frontier labs to some extent, all the hyperscalers, the neoclouds. The demand for Nvidia chips outstrips supply by a decent margin. There must be so much that goes into this - the commitment from musk to exclusively use Nvidia, the weird equity deals, etc. For example, is Google just completely unable to buy GPUs at this point because Nvidia sees the TPU as a threat?

All the conversation on twitter seems to be about how cheap this is. Are they just choosing to lose money on inference to gain market share or do they actually have inexplicably lower inference cost/more efficient models?

The shear hypocrisy of this is quite staggering. Ok, so we own the outputs, but you get to decide how we use them, and you don't trust us to use them. But the people whose books you stole to build your psychopath machine, they didn't trust you, did they. They didn't really want you to build a machine that can generate thousands of books to compete with their work, but you did do it anyway. Are you going to commit to auditing every input that was used to train your model and attain positive consent for their use for training?

It's also pretty wild to call this standard practice. It's not. I can grab any of the open weights models and train to my hearts content. So it's not standard is it. You'd like it to be standard because you don't want to compete.

And you don't trust us, but it is your company that's been going around telling us how excited you are that your model goes out onto the internet and hacking people.

This continual authoritarian bent from the least trustworthy people in the world is deeply problematic and the only saving grace is their absolute total and complete failure to enforce the restrictions they wish to place on us.

Your ability to build the God machine doesn't magically endow you with the moral authority or judgement to decide how it's used, and the fact that these people believe it does is a great indicator that they aren't to be trusted.

I really don't understand why they do trials for this. What would be the failure case for the trial? "We carried out the trial for 6 weeks over X period of time, we arrested Y people, at that rate we've decided this is not compatible with modern liberal democracy, we would've traded a free society for Y+2 arrests over X period but we just didn't quite get there"?

As suggested by other people - this is just going to be used to identify, infiltrate and disrupt the numerous peaceful protests that the police aren't keen on in London.

That is absolutely not true in this case. If what Apple alleges is true, the closest example we can look to is Waymo vs Uber. Where Uber ended up settling, they paid Google a quarter of a billion in stock, agreed not to use the IP they had stolen and fired the engineer involved. This could very easily just kill OpenAI's hardware product, they might literally not be able to release what they've built. At that point they have to decide whether they start again from scratch, with monitoring to ensure they aren't using any of the IP or they just give up on this expensive distraction from their core business in the run up to an IPO.

This looks really bad for OpenAI. Honestly, do not post text messages on the web, you look like a messy bitch. At this point I really don't see this company going public under Altman, they're going to have to bring someone in to bring some professionalism here. Altman seems to have badly mismanaged pretty much every corporate relationship they've got going. OpenAI is such a natural fit to be a partner with Apple, yet somehow they've managed to completely burn those bridges in the most public and messy way possible.

It makes me so angry that an entire branch of "science" gets to just take data and go "I dunno man, maybe kids are just dumber". There's so much of this in social science. Take some measurable thing, and then just wildly speculate about the cause of it in an entirely untestable way, and then test that theory in ways that absolutely do not justify the conclusions. At the very least, the whole field should get a lesson in how to communicate the certainty of their theories and conclusions. You've got some guys hanging around couching their uncertainty when they're banging sub atomic particles together at the speed of light and then you've got some social science bro being like "If you move that checkbox on that form to the bottom it'll be way more effective, trust me bro I've hacked how people think".

The "Why Bonsai?" I found really funny.

Let me re-write that section for you:

Why Bonsai?

At Jane Street we're super excited by Functional programming and by CAML in particular, so when we need low latency software, we use OCAML, when we need hardware, we write out own langauge - HardCAML, and when we need a Web UI, we build a Web UI framework in CAML. Because we fucking love CAML.

If you look at Apple's market cap on a graph, you would think Steve Jobs was holding Apple back...

Alternatively, if you looked Apple's market cap on a graph you would think covid absolutely exploded the stock market and it's never recovered since. Their P/E has basically gone from 15 to 35 since 2020 and never come back down.

Yes absolutely, famously American cities run crack dens using municipal taxes.

It is interesting to hear how openly some CEOs will just say outright stupid things. Like ok, so you re-do the voting system to give rich people more votes than poor people. Fine. It's done. What's the first order of business amongst the electorate at the next election. Well Elon Musk now has 10 times as many votes as Tobias Lütke, why would Musk not just vote to change the system again, handing more power to himself? And on the basis of what principle would you oppose him?

It's just fundamentally a stupid thing to say, and the fact he says it in public strongly indicates that he's lost contact with anyone who will challenge him on this stuff.

Having been to Gatwick plenty of times, the weakness with this vs the already existing valet parking is that the valet parking is just 1 story of the short stay car park.

You just drive up, drop your keys and go, your car is parked in a long line of cars that they can asynchronously drive off to long stay. The key thing is the location - you're a 30 second walk over a bridge into the terminal. Probably a hundred cars can be sitting waiting to be driven off to the long stay at the time, so they can handle the drop off/pick up of many cars at once. Whereas with this solution each "bay" will be occupied for a time while the autonomous robot drives off the previous vehicle, so you either need to over-provision - have many robots servicing a single bay to cope with peak times, or you'll need some local skid buffer.

From a customer point of view. This is on the outskirts of the airport so it's a much longer walk without a covered path (so you'll get wet because at some point it will start raining again in the UK). It's maybe slightly better than long stay, where you do need a bus journey (which tbh isn't that bad). But I'm not sure how much juice is in this application to squeeze.

Take a look at most successful journalism today. It's behind a paywall. You get paid by the people who are interested and value your work.

Can Google steal it and present it in an AI overview? Well kinda. Today Google is doing a trick - they're saying "You can refuse to consent to being fed into the slop machine, but if you do we won't crawl you for Google so you'll get no search traffic. But you're not going to get search traffic anyway! So you might as well opt out of being fed into the slop machine. And companies are starting to do that [1]

It's really interesting, because essentially what it means is Google is turning into a walled garden, but there's nothing growing inside it so they have to continually import new plants to live in their walled garden and they're going to have to pay to do that. So soon Google will be paying news sites for the right to plumb their feed into the slop machine.

[1]: https://www.wsj.com/business/media/google-search-publishers-...

I think this is a reasonable set of objections to how cut and paste works if you were designing it from scratch. I'm not saying I agree with the conclusions, cut AND paste, is 2 operations so it seems silly to say you can undo both with 1 undo. It's kind of a RISC/CISC thing, is your opcode CUTANDPASTE or CUT and PASTE?

I'm not certain the point of an interview is to ask you to write the exact lines of code that you would write during the job.

I think there's two elements to this. First - all the things you list are important too, and we'll get to that. But there's no point in getting to those things if the candidate can't write 5 lines of python and has no idea how the language operates.

Second, hiring at startups is massively different to hiring at big companies. At big companies you have a website and a linkedin and a recruiter and external recruiters and they're all funnelling CVs to you dozens at a time. You're glancing over a CV and if it looks roughly right you'll do a phone screen. People lie on their CVs. So you need a few questions like this just drop the 90% of candidates who have literally no chance at getting the job. Keep in mind, the external recruiters will interview the candidates you reject, find out your interview questions, find the answers and tip off future candidates.

We're also hiring for a different set of skills. At a start up everyone is doing everything. Does it matter how well you can talk to customers? Maybe! That could be useful. In a large company if you ever talk to a customer as an engineer you will have been at the company for years, you'll have a Sales guy, Technical Sales guy, an Account Manager, your manager, and maybe a Project Manager all in the room if you ever get to see a customer.

The praise for the Chinese labs is generally around how they deliver fantastic results with less resources. xAI is the opposite, they're struggling to keep up with frontier - massive churn, selling off compute because of lack of adoption, acquihiring to fix their deficits, whilst having the most resources of any AI Lab. It makes it difficult to evaluate things like "Is the API price competitive because they have an efficient model, or because they don't need to make money".

It's not a problem for you maybe, but if you run a business providing some niche specific software in some area, having the model suck up all your code and learn from it is pretty harmful because your competitors and customers will be able to vibe code your product after the next model release. You're training a competitor to your business. People are very worried about this which is why companies like Claude don't train on enterprise data, and for lots of companies this exact situation is a deal breaker.

>The Org Chart Survivor - Your company made you 'AI Lead' on top of your actual job. You use it, it helps sometimes, and you've stopped trying to explain to leadership what it can and can't do. You run little experiments and post the results. Just try things, you keep saying. Just try things.

This was scarily accurate, and I think the options for the answers were fairly representative of peoples' different views.

I think what's interesting is that on social media we're generally discussing one of those 15 questions, not all of them in context. Multiple things can be happening in parallel. I'm not convinced about it's use in medical imaging and I think the companies are going to probably mostly go bankrupt, but much like the internet or railroads something of value will emerge from the other side. It's not like crypto in the fact that there is actually something of value there beyond just crime.

Interesting. But if Sam really believes that the US public should share in the benefits of AI surely the number should be 50% not 5%.

The hardware will become obsolete but we may end up with an oversupply of cheap power - overbuild on things like Nuclear and solar. That would be nice.

I think we're going to look back on this as one of the things that really exacerbates this bubble. It's one thing for these companies to have high expectations of their future compute needs and therefore overbuild. But what the SpaceX/Anthropic deal showed was even if you fall short, you can profit by selling off your overcapacity. This means there is essentially no incentive to limit how much capacity you're building, anything you don't use you'll still profit from.

The problem is that only works if aggregate demand is higher than supply. At some point, all these new data centres are going to come on line, and the frothy "throw AI at everything with no regard to costs" is going to drawn down across the industry. And this supply will be much larger - because everyone thought the down side was limited.

At that point you're going to have all of these companies trying to dump their excess capacity on the market and it suddenly won't be true that you can just sell capacity to your competitors.

Obviously this won't bankrupt Meta - it'll just eat into their profits from their ads business. But it likely will drive a bunch of neo-clouds out of business very quickly, and the technology providers like Nvidia etc will suddenly come back to reasonable P/Es.

Let me blow your mind. The reason laundry doesn't take much of your time is because we built a robot to do it for you and it already handles 99.99% of the job and it costs a couple hundred bucks and it's called a washing machine. This machine costs $8000 and is promising to solve the other 0.01% of the time spent on this job, and it's promising to solve 0.01% of that 0.01%.

This is another industry that seems to me a lot like the AI glasses. It's that sweet spot of being extremely difficult to make work, whilst simultaneously offering almost nothing of value.

It's actually striking that what they're promising it can do is almost nothing, and that it still won't be able to do what they claim. "Folds and puts clothes away" - ok, can I see any video of it taking clothes out of a washing machine (it can't), folding them, identifying where the clothes should go and then putting them there, and for example, opening a wardrobe and putting the clothes into the right place - possibly underneath other stuff that's already in there.

It's being released on November 19th, you need a box so it can go on a shelf in walmart so your relative can buy it, wrap it up and put it under the christmas tree.