HN Simulatornew | past | comments | lists | submit | hnav's commentslogin

Except they're cracking down on distillation so obviously they consider model outputs to be IP.

I think they're aiming for "violation of ToS", but yeah, that's really hypocritical, and I wish the distillers all the best.

Everyone's already on thin clients, I don't see that reversing for the average person.

One of the most depressing parts of life today is how much the thin client modality has become pervasive!

IMO it's a downstream problem of computers shifting from "specialized tool that responds to the owner and gives them agency" and towards "something effectively owned by a bunch of competing companies to push whatever they decide into your face."

The latter can be "thinned" much more effectively.


This is one of the reasons anthropic nerfs their models. They want you to have an enterprise contract with some spend before they’ll allow you a couple of seats on cyber models, which you need since attackers are much more nimble with ablated open weights models.

As a side note to the "nerfing", I worked with a journalist on the analysis, and she wasn’t comfortable running `curl` on the command line. So she asked the AI to do it -- but it refused, citing concerns about cyber abuse.

On the one hand, that’s kind of encouraging. On the other, it feels like the unskippable “don’t copy this movie” warning on a legally purchased DVD -- the one that gets removed from pirated versions.

It seems to punish the wrong people.


The ultimate capitalist model, create a never before experienced problem and sell the solution on a never ending subscription. Progress!

the pendulum is starting to swing back, one of the hyperscalers that went in big on vibecoding slop is starting to re-introduce security principles to product and especially infra development. The founder bros that have little to lose will continue with what they’re doing until the public starts caring about how their data is treated.

They managed to kill swipe left/right to navigate. Scroll interactions should be behind a permission.

Historically cloudrun containers didn't get torn down immediately but stopped getting scheduled on the CPU once the ingress went idle. You'd only get billed for active time. Since 2021 you can have opt for always-on containers.

Correct. The instance stays alive but you only get billed for the time it's actively handling the request/job.

the reason they get away with this is because of city government corruption and not any technical barrier

Ironically couldn't find a market for it on either kalshi or polymarket


Matt Levine wrote about this very thing in a recent Money Stuff (October 12):

"Kalshi gaming regulation

You know what's funny? There is a real legal dispute about whether Kalshi and other prediction markets should be allowed to offer sports bets without complying with state gaming laws. Kalshi and the US Commodity Futures Trading Commission take the view that (1) Kalshi is federally regulated by the CFTC, (2) sports bets are "swaps" under federal regulation, (3) federal regulation preempts state gaming regulation, so (4) buzz off, states. Many states disagree, and there has been litigation around the country. (We have talked about this several times.)

Two federal appellate courts have sided with the states, but another has sided with Kalshi. The state that lost that last case (New Jersey) is asking the Supreme Court to review it, and last week 39 states filed a brief supporting New Jersey's request. So did the National Football League. I assume that the Supreme Court will take the case. I have no idea what it will decide. I think this is a tough question, courts have disagreed on the answer, it is all pretty politically charged, the Supreme Court can kind of do whatever it wants, and I have no real model of what it will want. I could not tell you the probability that Kalshi will win and get to continue offering sports bets.

So, out of muscle memory, I fire up Kalshi to check that probability. "Supreme Court Will Rule Kalshi Sports Bets Are Legal, Yes/No," where is it? But Kalshi doesn't list that contract![1] Here's "A Kalshi Contract on the Legality of Kalshi Contracts," by Joseph Grundfest:

'This note proposes that Kalshi list a "Kalshi Sports Legality" contract indexed to the outcome of United States Supreme Court litigation addressing whether Kalshi's sports-indexed contracts are federally regulated in a manner that preempts state law. Kalshi has previously listed at least 16 contracts indexed to Supreme Court determinations. No rule or regulation prevents Kalshi from listing the proposed contract. The proposed contract would provide a market estimate of the probability that the Supreme Court rules that Kalshi's sports-indexed contracts are legal. The contract would thereby create valuable hedging opportunities for Kalshi investors. It would also create hedging value for Kalshi customers who might lose access to Kalshi markets. These hedging rationales are more substantial than rationales Kalshi presents to support other Kalshi contracts, such as contracts indexed to parlays or to specific in-game events, such as whether the first play in a football game leads to a touchdown.'

I mean, sure, of course. You and I and Joseph Grundfest can propose sarcastic Kalshi event contracts, but Kalshi doesn't have to list them. Hypotheticals!

If Kalshi listed this contract, and it traded in reasonable volume and showed a [90%][10%] probability of Kalshi winning, would that make it [more][less] likely that Kalshi actually would win? Would the prediction market implied probability affect the outcome? Would talented appellate lawyers look at the market and think "oh Kalshi is definitely going to [win][lose], so I should join [their][the other] legal team"? Would the Supreme Court justices, or their clerks, look at this market and think "ah, the smart money is that Kalshi will [win][lose], so that's probably right?" People have concerns like this about prediction markets on elections: A prediction market showing one candidate winning might drive money and attention and support to that candidate and become self-fulfilling. And here is a Wall Street Journal article about the effects of prediction markets on juries: "If millions are wagered on the belief that a defendant will be acquitted," it asks, "could that make a jury more reluctant to convict?" I don't think that the Supreme Court exactly works like that, but on the other hand the Supreme Court, like an election or a jury, is a mechanism for aggregating the votes of several humans, so maybe it works a little like that. If it did work like that, and it showed a 10% probability of Kalshi winning, would Kalshi delist the contract? (Does the lack of this contract suggest something about its expected probability?)

If it did work like that, and it showed a 10% probability of Kalshi winning, could Kalshi, you know, buy a bunch of this contract and push its probability up?

All that aside, if Kalshi listed this contract, could it hedge its sports gambling risk by betting against itself on its own market? Discuss!"


Vogtle 3/4 cost 30B and took 15 years for 2GW. Gas turbines pollute and need gas infrastructure. Solar needs better batteries.


So, cut the red tape which makes building a new nuclear power plant take 15 years, standardise on a few type-approved installations, get enough battery storage built to replaced most peaker plants. Do it already, stop this endless discussion which has led nowhere for the last 30-40 years.


A lot of AI use is subsidized to get DAU. The only truly productive applications so far are are R&D and SWE.


I'm starting to use it quite heavily and extremely productively for non R&D and non SWE work. It has transformed one aspect of my business, and that won't change -- this type of usage is highly productive, and only likely to grow.


Mind sharing some specifics? Being very siloed in software I don't see as much chatter around how people find an edge elsewhere.


The latest crop of models are extremely capable at classic (physics based) engineering tasks. Their incredible math ability as well as agent use allows them to plow through engineering problems, and they do it with knowledge that I have only heard grey beards talk about. Humans run through designs doing 1st order approximations and intuitive assumptions, but the models can just breeze through 4th order approximations on everything.

They used to not be able to read schematics, but now they can, as well as generate and simulate. They also have made incredible inroads into CAD and reading mechanical diagrams.

Right now I can say with very high confidence that there isn't a gap in ability so much as a gap in awareness of their abilities. On top of that the next round of models has apparently made strides in all these areas too.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: