EthanHeilman

I've found it to be useful for prototyping. You can throw a game mechanic up, have it build some AI players, tell you any balance issues it runs into, play against the AI yourself to see if the mechanic feels fun. Then iterate.

A few times I've made a tweak to a mechanic and the LLM balance tests immediately found a major issue. Sometimes the AI players it creates just suck at the game. You still have to think, but just having balance tests and then having them cheap to update when I experiment, it extremely valuable. I tested this on a game I had mocked up with cardboard and forced my friends to play. It found all sorts of stuff I missed in those gameplay tests. It didn't replace those sorts of tests with humans, but it let find issues faster so I got more out of human interaction tests.

Agreed, that was what I was trying to say. You can build a lego structure that you can not infer by just looking at from the top, bottom and sides.

Build an object out of legos and then reconstruct the shape of what you built by only looking at it from the sides and above.

That's what 95% is intended to capture. That some level of expertise in an area is should be captured by 95% sample of the population, you could push it to 99% of 99.9%, but you want it to be quantified by a number to avoid arguments that something is not AGI because obscure field or expert exists that AI can not do.

95% better at 95% of the population is already approaching ASI. One could even argue that AGI is 50% better than 50% of the population.

All that said, this over rotation on benchmarking misses something critical. What is general intelligence? We assume that humans have it and we assume it is captured by benchmarks on "intellectual tasks", but it is probably the case that general intelligence is based displayed by judgement on uncertain outcomes. Benchmarks by their very nature have certain outcomes, they have a wrong and right answer.

Test AIs on questions we don't have the answers to and there is no clear right answer, but there will be at some point in the future. What will the economy do? Which US senators will be be re-elected that polling correctly suggests will not be re-elected. Which US senators, currently not in office, will actually pass bills representing the wishes of their voting base? What published papers will be seen are groundbreaking in 5, 10 15 years? What approach to unifying physics should be taken?

Does RSI actually mean anything specific anymore? RSI, AGI, at this point seem like buzzwords. Sure AGI has definition that are measurable, say "better than 95% of humans on 95% of intellectual tasks" but if we used that definition we already have AGI and almost no one thinks we have achieved AGI. We use AIs to train AIs which we use to train AIs, why is that not RSI? How much human intervention means that is not RSI?

Oh nice, thanks for sharing

I believe jrflo was saying that OpenAI watches the chats of everyone from Anthropic because watching what Anthropic employees type into their personal ChatGPT accounts is a critical source of intelligence on is happening inside of Anthropic.

I would be surprised if OpenAI isn't doing that. OpenAI will take any advantage they can get. If an employee at their primary adversary is typing useful intelligence into OpenAIs website, a website that does not promise privacy from OpenAI, the only reason they wouldn't weaponize that information against Anthropic is ethics or fair play.

A careful reading of "we cannot rule out that de-identified data derived from their usage of our products helped improve our models" could be saying that yes they trained on it but they don't know if that training data resulted in an "improvement" to the model. That is, they can't rule out that the only reason the model found this solution was because it had been trained on this approach.

The term ruled out is very open ended and gives them significant flexibility of meaning. They may have the information to determine exactly what happened, but they haven't looked so they can't "rule it out".

I am surprised that this happened to liquid. Many of the other hacks seem to the sort of mistakes that I expect projects to make. RNG-failure resulting in bad keying material is a mistake that people have been making since at least the last world war almost 100 years ago.

> You have accurately described a email mailing list. Where in the value-add here?

Pre-prints are basically a mailinglist where you post your paper prior to peer review.

The value over a simple mailinglist is:

1. Stable URL and citation to enable other work and discussions to cite and reference it.

2. Versioning of the paper, allowing updates to be made without having mail out the paper, while allowing everyone to find all prior versions

3. Host for a PDF and data that might be quite large

4. Centralized searchable long term archive of scientific papers

5. Scalability, arxiv gets 30,000 submissions a day, no one wants to receive 30,000 PDFs in their inbox everyday

They wouldn't be the only party, but they are in my opinion the most likely party.

What the people that can monetize it? Microsoft, X, IBM, Salesforce, Discord, Google

Microsoft would likely buy OpenAI due to their investment and integration with OpenAI. Making Anthropic not that valuable.

Google would likely buy Anthropic due to their investment and ties to Anthropic.

X, Salesforce maybe. Not sure Discord has the money. No idea if IBM capable of growing.

All this changes if Anthropic figures out a moat/sticky product that doesn't just depend on having the best LLM. If they pull off https://claude.com/solutions/healthcare then all bets are off.

And that might be why Google loses, they are competing with companies that are willing to operate outside the law (or at least pay Trump to have the DoJ look the other way).

Gemini seems like a better search engine than the alternatives because it has access to Google's massive crawler feeds. I think Google could charge 30 dollars month for Gemini + all the data Google has locked up (Scholar, Books, crawled webpages, Google Groups, Usenet archives + search of your personalized datasets such as calendars, photos, emails, docs, slides, etc...). I'd pay that easily.

Anthropic has no moat, their models just get distilled and resold. They can maybe get good margins when LLM improvement rate is very high but as it slows down they are looking at commodity margins. Google has the products that makes that commodity more than just cost of electricity + 0.5%.

My case is based on the assumption that Anthropic will not hit RSI or if it does RSI rapidly hits a wall. Faster your growth curve, the faster you eat all the low hanging fruit and s-curve. I could be wrong here, maybe RSI will cause a hard takeoff singularity by 2030 and just keep going, but if that happens the world fundamentally changes.

I disagree. Google is one of the few parties that can monetize AI because Google has a massive moat in the form of their products: gmail, chrome, photos, search, etc... and Google already has the custom-built chips and datacenters.

Google spending money on competing head to head with your LLMs is a waste of money for Google. If anthropic wins, Google copies their approach, buys anthropic for cheap or both. All the investors throwing money into OpenAI, Anthropic, are just accidentally subsidizing Google's product development. Google shouldn't spend its AI research capital in an arms race with Anthropic but instead should invest in AI approaches that no one else is investigating at scale. That way Google can hedge against LLMs hitting a wall.

There is a real but small danger to Google that Anthropic replaces Google as a search engine, but that is an uphill fight for Anthropic. Google has massive brand recognition, network effects with gmail and chrome, Anthropic can't just copy what works from Google. On the other hand Google can copy what works for Anthropic. Google would have to play poorly to lose that fight.

Probably the worse case for Google is that software becomes so cheap and easy to create and maintain that all of Google's product offerings become commoditized. Even in that world Google has a lock on infrastructure. Perhaps ASI software creation completely removes that as well? If so we are living in a post-singularity world and probably the stockmarket doesn't exist anymore either.

You can use a self-hosted identity provider with OPKSSH. OPKSSH is just the protocol and the software that plugs an identity provider into SSH. Lots of people use it with authentik instances they run themselves.

> Still, using certificate in OpenSSH (instead of public key) is easier for me once it starts to scale beyond 5 or more hosts. Even do dual-cert, by host and by user combo, while using brittle OpenLDAP-OpenSSH for other class of users.

SSH certificates are fantastic. OPKSSH wouldn't work without them.

> And smashed all my notes on certificates and OpenSSH on my ad-free, revenue-free, JS-free Pelican blog. It's personal on both note-wise and server.

Will check it out.

Exactly! Assume that you will have bad apples and build a system that protects against them and minimizes the harm they do.

Those jobs protects aren't necessarily wasteful in that they stimulate the economy and cutting them would have serious impacts for those states. The essential question is, could that subsidy to a state have more impact. For instance, rather than building tanks the military does not want or need, fund and hire more teachers, build things the military actually does need like ships, fund startups, fund science projects in those states, etc...

Because finding people that are 100% ethical is extremely difficult. Even if we are wildly optimistic and say 20% of the population is 100% ethical. You aren't likely to weed out unethical people, so you are hiring people, training them, and then firing them 4 out of 5 of them. There are many cases where an experienced but occasionally unethical worker is better than an unexperienced but ethical worker. When faced with this dilemma it is likely that more police debts would simply cheat or cover up police abuses to retain valuable staff or staff at all.

The solution is not making humans more virtuous but reducing the capability and the harm done that unethical humans can do.

> If we cannot trust them with the very basics of ethical behavior they are absolutely in the wrong job and there need to be very clear consequences.

Police should not be trusted because they are police. There should be audits and controls that prevent abuse and unethical behavior. Small unethical behaviors should result in corrective measures but not termination, since when the punishment becomes too great you create incentives for cover ups or scapegoats. A small number of minor punishments, that catch people as soon as they step over the line, functions better as a deterrent than a large scale punishments that are unlikely to be actually enforced. Granted if a police officer does a major crime, they should face serious consequences, but the goal should be to creating a system that makes major crimes by police less likely. If they know they will get caught for minor crimes, they are less likely to commit bigger crimes.

A Surveillance state is not the same as police state. We already live in a surveillance state, but we are protected by three things:

1. The cost of arrest. Actually sending someone to pick the person up and put them in handcuffs, process and change them.

2. The cost of prosecution. Having lawyers build a case, schedule a judge and courtroom.

3. The cost of enforcement. Putting someone in prison or threatening to put them in prison if they don't pay back what is stolen. Tracking if they paid it back. If they get probation it is cheaper but a probation officer is required.

These are expensive and use limited resources. Policing and governments prioritize these resources for more serious crimes like murder. Constitutional rights, including due process dramatically increase these costs. Mass surveillance made investigation and determining who did what for minor crimes done in public much much cheaper, so we just hit the next cost bottleneck.

The solution to the enforcement bottleneck is automated enforcement. This will likely be rolled out over the next ten to twenty years. If you steal from a store, you get banned from the entire story chain. This means living life very difficult, so you pay a fine. It is likely that stores will make lots of mistakes and flag innocent people, but most people will pay the fine because it is more convenient.

The situation is entirely different in a police state with no constitutional protections. The government can simply lock you out of everything you need to live because you did buy enough Dear Leader portraits this quarter. You can't use your car, can't call a ride share, can't catch a bus, can't buy food, can't get electricity to heat your house to keep your family comfortable. Instead of a fine, you need to walk to closest detention center, confess your guilt, lock yourself in and do manual labor and reeducation until the state allows you to leave. This system requires no guards.

This is a very cool idea, surprised no one has done it before!

It was an accident, you can read the investigation. No one claims Nagasaki or Hiroshima were accidents.

> I think americans have the false belief that US is some of kind of benevolent force acting for the good of the world and promoting freedom and democracy.

A state can still make mistakes without saying it is good in everyway

> Iran does not think they are dealing with a madman.

No one knows but the Iranian leadership. The Iranian leadership has been famously bad at modeling the intentions and motivations of other nations leaderships. A bolt of the blue decapitation strike, followed by the US having plan if Iran closes the straits which is the obvious response by Iran, does at face value appear to be the work of a madman. Now in the US we might conclude that Trump and Hegseth are just wildly incompetent and unprepared, but it seems likely to me that Iranian leadership see irrationality instead of incompetence.

Threatening autocrats might work, but just bolt out of the blue decapitation strikes undermine future threats because they figure they'll get no warning. If you are threatening, you are bluffing and when you aren't bluffing, there are no threats.

> I think "Iran was considering a peace deal" is a bit of a stretch here.

Iran was considering a peace deal. I agree that the most plausible was they would reject it.

> What does Iran still have to lose? Well, a lot.

The US could do this, sure, but then Iran would have even less to lose. This might work if the US started small and threatened escalation to try to compel Iran, but the US started at massive escalation so any additional airstrikes are likely to be less escalatory and thus less of a threat.

Even worse, there is a fundamental problem with madman theory, if Iran believes they are dealing with a madman, then threats aren't effective because a mad man doesn't keep promises. If you think your opponent is not rational, then you should not expect them to follow cause and effect.

> Iran is never going to capitulate, until it capitulates. Their rhetoric is going to remain that the US has no more levers and can't change anything, because admitting otherwise invites those levers to be engaged.

I agree that we don't know exactly how much pressure is on Iran. Iran historically has been willing to suffer almost any cost. During the Iran Iraq war then sent enormous numbers of teenagers in human wave attacks over and over. It is my estimation that the current war with the US has helped to stabilize the Iranian government and that they benefit more from the war continuing than from a peace deal.

The only military lever the US has left on the table is an invasion of Iran. Maybe limited to the coastline or maybe complete regime change. Trump has not even attempted to bluff that he is doing this.

The bigger issue is the tankers. The US Navy isn't going to be happy patrolling the strait sure, but even if they did they wouldn't be able to protect the tankers enough for it to make sense for tankers to take the risk.

The last time this happened the US opened the strait by accidentally shooting down an Iranian passenger plane after sinking a large chunk of Iranian navy. The Iranians assumed the US shoot the passenger plane down on intentionally as a war crime and assumed the US would was planning to escalate the conflict. This fear deterred further Iranian attacks on tankers.

This isn't going to work this time because the US started the war by performing of the most serious escalations possible, a decapitation strike against top Iranian leadership in a surprise attack using a diplomatic negotiation as cover. The US did this while the strait was open and Iran was considering a peace deal.

Threats of escalation are no longer effective at deterring Iran because Iran now believes the US will take such actions regardless of what Iran does. What does Iranian leadership have to lose by staying the course? Very little. On the other hand if Iranian leadership back down, they loose all their leverage, they look weak internally, they look weak externally and the US might decide to attack them out of the blue again.

This is why decapitation strikes are generally not done. They remove options and they undermine deterrence and paint belligerents into a corner.

I'm not arguing against anything you have said, but there is an important connection here that I want to make between disillusionment and funding.

> Besides the people in this thread bemoaning the state of research funding, international students, etc. (all of which are valid), a lot of people are becoming disillusioned with academia. ... grueling work for terrible pay, all for difficult job prospects given the current market.

If there was more research funding and more jobs for researchers in academia this would result in both better pay for PhD students and more academic jobs post-PhD. The disillusionment is related to the funding.

Imagine if the US currently had a new massive project for physics research, you would get a lot more people doing physics Phds and much faster progress in physics. We know this because Edward Teller tricked Reagan into pumping billions of dollars into optics research and that resulted in optics being one of few areas in physics to see breakthrough after breakthrough.

r/K selection theory but for protocols?

This unofficial Bitcoin roadmap and the official ETH (Ethereum) PQ roadmap present two very different philosophies.

* As one would expect the ETH PQ roadmap is ambitious bordering on hubristic. It proposes a very short timeline and aims to deploy protocols that will to push the frontiers of current technology.

* This proposed Bitcoin roadmap is conservative and focused. It is designed around managing risk and hedging against unknowns using currently available tech.

Reminds me of that Arthur C. Clarke short story [0]:

> "We realize now that this was our first mistake. I still think that it was a natural one, for it seemed to us then that all our existing weapons had become obsolete overnight, and we already regarded them as almost primitive survivals. What we did not appreciate was the magnitude of the task we were attempting, and the length of time it would take to get the revolutionary super-weapon into battle. Nothing like this had happened for a hundred years and we had no previous experience to guide us."

[0]: Arthur C. Clarke , Superiority (1951) https://www.baen.com/Chapters/1439133476/1439133476___5.htm

> but some portion of folks can attest to their ownership.

How? Alice pay's Bob 1 BTC at random address 0x1234. Someone shows up and says, I own that address and here is signature proving it. But the signature scheme is broken so anyone can do that. So you ask for documentation they own that address, well they have screencap of a message asking for payment from Alice. Is that real? Maybe you find the email of that user and ask them, but they could be lying. Now if you paid from coinbase, coinbase could vouch for you.

So you need some sort of court that sits in judgement over who owns what. That is going to be very expensive. While you are doing this, no one can move funds. What is the most likely outcome of such a system, well there is not CEO of Bitcoin, so you would probably end up with multiple courts producing conflicting rulings that no one would respect.

The whole notion of ownership courts is anathema to Bitcoin's philosophy and would completely undermine the social trust that makes Bitcoin valuable. If we are going to save Bitcoin from a CRQC we must act before a CRQC recovers everyone's private key.

There are three workable schemes:

* For public keys that in hashed addresses such as P2PKH (Pay-to-Public-Key-Hash) et al., if the public key is not known, then you could produce a ZKP that you know the public key (proof of pre-image). The main problem with this approach is that it only protects hashed addresses where the public key has not been leaked or exposed on-chain. It doesn't have enough coverage.

* You can do commit-reveal schemes, this makes miners far more trusted and again only helps with hashed addresses that haven't exposed the public key.

* You can do ZKP proof of HD Seeds, from most modern wallets have HD seeds. AFAICT You'd have to use STARKs but STARKs for HD seeds are too big for on-chain proofs. Not all HD seeds are protected and not all addresses have HD seeds. Just today Laolu published this demo for doing this, the proofs at 1.7 mbs https://groups.google.com/g/bitcoindev/c/Q06piCEJhkI

> In practice there are very few bitcoin outputs that aren't linked to an offline identity and most users could easily produce a proof of ownership.

Any who is going to in charge of reading that proof of identity and moving the coins? A trusted centralized party? The point of Bitcoin is to avoid exactly that sort of trust relationship, otherwise use the banking system.

> Satoshi's wallet alone could crash Bitcoin's value as a currency if dumped on the open market.

No one knows, but the incentives are aligned with a softfork to burn Satoshi's coins.