Is there a good source for total token consumption between the various labs / openrouter / etc?

Comments URL: https://news.ycombinator.com/item?id=48660159

Points: 4

# Comments: 1

Paul Graham has mandatory essay on this - https://www.paulgraham.com/submarine.html

(1) More than 50% of Americans at this point are more concerned about AI than excited for it - https://www.pewresearch.org/short-reads/2026/03/12/key-findi...

(2) Popular media is feeding into this zeitgeist with headlines like - "Prepare for an AI jobs apocalypse" eg - https://www.economist.com/leaders/2026/05/14/prepare-for-an-...

(3) There is a bright line between these articles and growing concern / pushback on the development of new data centers with both moratoriums and significant municipal cancellations.

(4) Perhaps more materially the architects of AI are being challenged directly - in April Sam Altman's home was (a) bombed and then (b) shot at and weeks later the entire industry was just taken to task by The Pope! himself calling for acknowledgement of human limitation, grace, and dignity.

(5) Meanwhile Sam and others are reframing including launching a new foundation to "increase quality of life and individual freedoms for people around the world" and pivoting messaging to AI "accelerating everyone in achieving their goals" https://x.com/sama/status/2059677202917331431 & https://x.com/sama/status/2057218997503086888

Is this because the architects don't believe AI will be as disruptive as planned or . . . ?

Couple of highlights on Ryan

- Built and sold Chewy from a startup to the largest ecomm acq of all time - Became #1 individual shareholder of Apple early on - Bought a 10% share of Gamespot in 2020 becoming largest personal shareholder - Took over as CEO after being a proactive board member, works for no salary

https://en.wikipedia.org/wiki/Ryan_Cohen

First most of the data center build out is happening in areas that have had little other opportunity so local resistance is muted. Abilene, Texas is referenced and is the kind of place my grandfather would lovingly say is where you go to learn to be a “dirt farmer”

Second, every environmentalist in the US is fighting 100 different battles with the most anti-regulation, pro-energy administration in decades (ever?) and has limited bandwidth.

And third, the AI narrative around national security, longevity, and super-intelligence-enabled abundance provides massive national coverage - the implications being that AI will solve any environmental and or human economic disasters that they enable.

https://www.imdb.com/title/tt1935156/

On the cover it's a story about the failed production of Alejandro Jodorowsky's Dune script but the deeper story was the aggregation of an unbelievably talented pool of visual artists including Jean "Moebius" Giraud (mentioned as central artist in 5th element), H.R. Giger, Chris Foss, Salvador Dali, & Dan O'Bannon.

That group would go on to centrally influence the visual style of a huge body of science fiction work including Alien, Blade Runner, Total Recall, Star Wars, The Matrix, Guardians of the Galaxy, etc etc.

The art and creativity on display in the film is absolutely sonic.

Kind of like the original PayPal mafia!

Anybody have a good solution that's utilizing actual traveler data vs the (non existent atm) TSA data?

Comments URL: https://news.ycombinator.com/item?id=47408441

Points: 185

# Comments: 74

https://closingline.substack.com/p/the-takeaway-kalshi-non-s...

PS - Christopher Cantwell - one of the writers and showrunners - has written a library of wonderful comic books worth investigating

PPS - ATX TV did a 10 year anniversary interview with a handful of the cast and crew that's worth watching if you're a fan - https://www.youtube.com/watch?v=q6L1suN-mGE

But mortality credits (pooling) don't solve the math of the discount rate - they add 100 - 150 basis points of reduction so retarget to 5.5% vs 4% if generous

So they are still structurally designed where they HAVE to allocate towards risk to meet their targets which is at core of issue

The "Safe Withdrawal Rate" assumed by many private individuals planning for their own retirement assumes a withdrawal rate in the 3 - 4% range based on the "trinity study" - https://en.wikipedia.org/wiki/Trinity_study

Meanwhile, American public pensions are structurally engineered around a 7%+ SWR - this was recently confirmed again by the median goal by the National Association of State Retirement Administrators.

The perpetual "under funded" nature, and all the return hunting etc in pension fund management can be explained by that disconnect.

But this then belies a very uncomfortable acknowledgement which is that we cannot afford the government workforce currently in place requiring us to either:

(a) Raise taxes to increase contributions.

Or

(b) Somehow make due with less government :)

"The best technical analysis anyone has done on bitcoin."

https://www.youtube.com/watch?v=XbZ8zDpX2Mg

Comments URL: https://news.ycombinator.com/item?id=46893050

Points: 4

# Comments: 0

https://www.youtube.com/watch?v=T2OxO-4YLRk

This is a fascinating point - if Neo / Tesla deliver a teleoperated hybrid at their <$30k price point the low-skill US labor force is going to be significantly disrupted on a shorter timeline than I would have previously estimated.

These are being pitched as "home robots" but clearly corporations will go all in - 24/7 operation (with multiple remote operators), no labor law / healthcare / pensions, spin up / down at will.

This is a quote from Alex Proyas (director of Dark City, irobot, the crow etc), he has long observed that humans are biologically under-adapted for a modernity of silicon-fueld complexity

https://www.starburstmagazine.com/features/alex-proyas-dark-...

And that's very apparent even in this article which leads with the logical fallacy of moral equivalence

". . . and America’s preeminent real estate fraudster who bankrupted six rigged businesses is all of a sudden concerned with supposed mortgage fraud"

BOTH are guilty, BOTH are crooked!

This directly leads to a lot of "the whole system is broken, may as well get mine"

eg - GPT-5 beats GPT-4 on factual recall + reasoning (HeadQA, Medbullets, MedCalc).

But then slips on structured queries (EHRSQL), fairness (RaceBias), evidence QA (PubMedQA).

Hallucination resistance better but only modestly.

Latency seems uneven (maybe more testing?) faster on long tasks, slower on short ones.