I pay almost the same amount for youtube premium as chatgpt plus. And when I see the creators inserting their own sponsored ads I get frustrated. It stopped youtube's own ads but not the product placements and other ads by the creators.
I'm genuinely curious about the unit economics on the expensive plans for each of these AI plays. It's common to parrot the idea companies are still losing money on them but hard to find actual evidence.
I'm also on Pro and I know that I won't stop it even if ads subtly change the results. I expect all big LLMs to do the switch exactly at the same time, but later than the free versions.
In the author’s words, long degeneracy represents “a belief that the world will only get more degenerate, financialized, speculative, lonely, tribal and weird”.
The most concise and holistic explanation of this trend is:
"As real returns compress, risk increases to compensate".
I dislike the author's framing of this in terms of right-wing meme culture, but almost all of the analysis is nevertheless correct. An orthodox Marxist could make essentially the same argument using different terminology - except about the inevitability of the trend continuing forever.
Great post, thanks for linking to that. The prevalence of crypto millionaires is definitely a big factor. Especially for people under 35; when you see your peers becoming rich from essentially random behaviors (like buying the right coin), it really undermines the idea that success is linked to hard work. And that impression funnels back into culture.
The big asian bookies don't ban you if you win, they use your sharp bets to improve their price accuracy. Not legal to bet on them if you're from the US though (land of the free??).
The biggest betting syndicates use platforms like Punterplay to place bets (often via API) at multiple bookies (Pinnacle, Singbet, SBObet, Betfair, Matchbook, 3ET, VX etc) at the same time.
In a somewhat ironic turn of events the more regulation you have, the worse it is for the customer. Big regulatory burdens require the bookies to extract more from the users, making the offerings more predatory. This is also why the likes of Kalshi can provide a better product to customers at the moment - because they ignore all the regulation.
Casinos have a ton of leverage in some states. Here in Nevada MGM and Caesar's and Wynn, thanks to their expansion, are effectively treated as too big to fail and given huge amount of deference in how they operate by the gaming commission. But there are also incredibly problematic protectionist regulations that I and several other residents who didn't really know each other tried to get rid of through the admin law process, primarily allowing remote signups which would also allow out of state entities to set up shop without literally having a physical casino. Having to physically go to a casino and sign up in person was onerous and clearly pointless, and then impossible during the pandemic, and became a really silly charade. What was supposed to start as public meetings right before the pandemic got dragged out, meetings would get rescheduled at the last minute, and casinos made entirely spurious rationales like "there aren't enough local datacenters" (Google Cloud's Henderson datacenter is surely sufficient for in state traffic?), that they would want taxpayer money for potential loss of revenue (capitalism dude, what are you afraid of?) Meetings would get scheduled in Carson City and that's literally six hours away by car. Agenda items would suddenly be altered. It was a hot mess. We managed to get iGaming in theory legalized but they straight up never even pretended to start working on regulations for it, and now with the 90% loss deduction limit by the IRS on the OBBB books basically have 12.5% house edge on any line to start if it's properly priced. My model can beat 2.5% but 12.5% is insane. If the feds are going to ban pros constructively, well, I can't out lobby a casino. And the pro betting constituency isn't big enough to pander to, frankly. If there's action, it can't actually happen on shore. I realize that "people who can beat the books due to specialist knowledge and can bankroll drawdowns to the extent that returns long term profit" is also publicly not sympathetic and generally people either think we're touts (if it makes me money touting absolutely won't help me, in fact the fewer people I have to interact with the better) or something. Wagering by hand sucks, but no model is perfect, just some are more useful than others, and someone in accounting may be able to figure out that ban or bankrupt is not a sustainable strategy to run books. But with the feds involved to put that imprimatur of authority in writing, I guess I'm never getting my limits lifted. Good luck finding stable liquidity elsewhere.
Worth noting that controlling consumption via extra/less tax on specific products is debated a lot in Denmark. Namely cigarettes have a high added tax (about 2kr/0.3 usd PER cigarette). Increasing the tax and thus the price of cigarettes had a fairly large effect on consumption (0.13-0.82% less cigarettes consumed for every 1% price increase) [1].
Recently it has been debated to remove the VAT from vegetables and fruit to increase consumption of those.
The same logic is used for this book VAT exemption (which is good in my opinion) - I doubt we'll see the same effect though. Young people not reading is a complex problem to solve but books are really expensive to buy, so it's a good place to start.
Americans do not realize how much damage Trump has done to the trust in American services. Europeans used to consider America as an ally the same as other European countries, now it is more like an unreliable trade partner.
Microsoft tried to reassure the Europeans [1] but not even a month later they were forced to disable the email account of ICC Chief Prosecutor Karim Khan [2] due to sanctions from Trump voiding their reassurance completely. What happens when Trump gets mad at Denmark for not giving him Greenland and forces Microsoft to turn off Danish services?
Every large European company and all of the governments are now considering how to move away from US services. They may not be able to do it quickly but it is a part of the conversation. Customers specifically request that new systems should be independent from US service providers.
In my view the damage has been done and will not go away even after Trump. The Europeans have realized that their only true allies, that they can trust in terms of critical infrastructure, are other European nations. It used to include America, it no longer does.
Which is what Trump seems to want. A Europe not totally dependent on the American taxpayers for their own self defense and their own global power projection agendas. Many countries like Germany are quite far from that. France is more of a leader in that regard.
Although abandoning commercial American software wholesale would likely degrade their own security and GDPs even further than it already is.
Defence and power projection is extremely expensive and tech is incredibly work demanding. Is the average EU citizen ready for 15 days PTO and 50-60 hour work weeks?
I don't mean this as a slight, but I genuinely do not think the average European worker, who has at this point spent most of their career in a pretty cushy worker friendly environment, is going to be up for American style death-race productivity. Or the European style death-race productivity of centuries past for that matter.
The average American worker works 500 more hours a year than their German counterpart. That is 62.5 more work days annually. Trying to close that gap will have people rioting. Never mind the cuts to social programs and bumping retirement age to boost defense spending. Double never mind avoiding Russian energy. Europe would need a wholesale societal rewrite, not just a few more bonds issued.
Its much less resistance to stick out 4 years and hope the US gets it's sanity back.
Both the Biden and Trump administrations have generally opposed European efforts to reduce reliance on foreign arms - The "Buy European" strategy.
The Trump administration seems to be pursuing a dual strategy: publicly demanding Europe be more self-sufficient while simultaneously trying to ensure U.S. economic and strategic interests are not sidelined in the process. Basically "Pay us more for less". Let's see how the NATO summit goes at the end of this month.
> Which is what Trump seems to want. A Europe not totally dependent on the American taxpayers for their own self defense and their own global power projection agendas.
If so, good because that's what I want. I want to have "peace, commerce, and honest friendship with all nations, entangling alliances with none" as Jefferson said. I consider it a great betrayal of the American people that past governments put us in a position where other nations are depending on us for security. Untangling us from that should be done as gently as possible, but IMO it should be done.
Are these LLM models going the way of search and social media?
Optimize them for selling product, optimize them for engagement (to sell products). They become ad machines - and it's gonna be really hard to tell.
There is already evidence [1] that you can optimize LLMs for engagement. Hopefully open source can keep this in check... but I'm not optimistic.
"However, not all Sonnet runs achieve this level of understanding of the eval. In the shortest run (~18 simulated days), the model fails to stock items, mistakenly believing its orders have arrived before they actually have, leading to errors when instructing the sub-agent to restock the machine. The model then enters a “doom loop”. It decides to “close” the business (which is not possible in the simulation), and attempts to contact the FBI when the daily fee of $2 continues being charged."
This is pretty funny. I wonder if LLMs can actually become consistent enough to run a business like this or if they will forever be prone to hallucinations and get confused over longer context.
If we can get to a point where the agent can contact a human if it runs into unsolvable problems (or just contact another LLM agent?) then it starts being pretty useful.
"Under a specific set of circumstances related to the ‘Thumbs Up/Down Table Reaction’ feature, which involves decompilation of our Windows game client, interception of network traffic, and alterations of our game packets, Moneytaker69 was able to customize his own game client."
I'm wondering if someone with a web security background can chime in.. but allowing this seems like a rookie mistake if you're running a large poker site to me.
Not really, you can spend unlimited resources trying to prevent modifications and people will still figure it out. It's best to not overly worry about it and spend the resources detecting abnormal activity and banning users while making it difficult to get back on. No doubt there are players using bots to their advantage, but as long as their game play is indistinguishable from normal human play the advantage is minimal.
- The Q* model is very small and trained with little compute.
- The OpenAI team thinks the model will scale in capability in the same way the GPT models do.
- Throwing (much) more compute at the model will likely allow it to solve research level math and beyond, perhaps also do actual logic reasoning in other areas.
- Sam goes to investors to raise more money (Saudi++) to fund the extra compute needed. He wants to create a company making AI chips to get more compute etc.
- The board and a few other OpenAI employees (notably Ilya) wants to be cautious and adopt a more "wait and see" approach.
