Ric Pruss

Deeply interested in networking, robotics, financial engineering and programming. Committed to lifelong learning.

[Rachel Karten @milkkarten]

The Norwegian Consumer Council made a video about enshittification and got 2.5M views and 168K likes on IG. (If the style feels familiar, it was created by NewsLab, the same advertising agency that made the brilliant Visit Oslo video from a few years back.)

[Pessimists Archive @PessimistsArc]

"The art of printing can be of great service in so far as it furthers the circulation of useful & tested books; but it can bring about serious evils… …it will, therefore, be necessary to maintain full control over the printers” - Pope Alexander VI, 1501

[Abomination @Abomination81]

Just had the creepiest scam ever happen 20 minutes ago.

I got a call from my wife telling me she forgot her wallet and needed the credit card to pay for her gas.

Except my wife was at home with me, and drives a Tesla. Same voice, bit... off, weird cadence but 100% sounded just like her. If she wasn't there, and it wasn't about gas I would have fallen for it.

I entertained it for awhile to get more out of it, I was so confused. Must be some sort of voice deepfake. The responses were too fast for ai, I think it was a voice filter. I wish I had thought to record it.

The future is about to get fucked. Warn your loved ones, do verbal passwords with your kids.

Bit taken aback

[jessy huang @goodhartproof]

I was just at YC demo day yesterday. Besides hardware/physical things, everyone is just basically just building a domain-specific harness

[clem 🤗 @ClementDelangue]

Open-source is not the cause of the cybersecurity crisis, it's the solution! https://t.co/YpykmLGv7B

[Chris Hayduk @ChrisHayduk]

I strongly suspect that Claude Mythos is a looped language model, as described in the paper "Scaling Latent Reasoning via Looped Language Models" from ByteDance

The authors of that paper called out graph search as one of the areas where looping provides a huge theoretical advantage over standard RLVR. And look at where Mythos blows out its competitors the most

[Ric Pruss @ricpruss]

This is incredible we are in the Singularity. If only it was Lobsters! (Context: Accelerando by Charles Stross, begins with “mind‑uploaded” lobsters whose nervous systems have been digitally mapped and run as software.)

[Dr. Alex Wissner-Gross @alexwg]

https://t.co/N2AwpI9f8m

[Vadim @VadimStrizheus]

Maturing is realizing that Tony Stark was a vibe-coder.

[Alex Berenson @AlexBerenson]

Tell Your Children: the @nytimes just admitted it was wrong to call for the full legalization of cannabis and called for making THC extracts illegal (!!)

I warned you seven years ago. Now the reality is impossible to ignore.

[Object Zero @Object_Zero_]

Taiwan

Taiwan currently produces 90% of the world’s advanced silicon chips.

The Taiwanese semiconductor industry consumes 7.2 GW or power each year, and each year it produces enough silicon chips to consume 45 GW of power.

Nvidia’s AI chips like H100 and B200 account for 5-6 million units, which is 4.5 GW of the 45 GW of compute produced by Taiwan.

Nvidia’s AI chips recently overtook the global server CPU market for “GW of compute”, and Nvidia’s AI chips recently production doubled in the past 12 months from 2.2 GW in 2024.

So for every 1GW of power generation on the island of Taiwan, Taiwan is able to produce around 6 GW of compute.

By 2030, every 1 GW of power generation on the island of Taiwan will result in 20 GW of compute manufactured.

Meaning Taiwan will be able to produce around 144 GW of chips of which 40 GW will be CPU, servers, etc, etc and around 104 GW will be Nvidia AI chips.

This represents a 2,300% increase in the energy needed to power Nvidia’s product output.

In the whole world today, there is 20 GW of AI.

Nvidia is projecting the following:

2026 = 4.5 GW of chips sold
2027 = 10.0 GW of chips
2028 = 22.0 GW of chips
2029 = 40.0 GW of chips
2030 = 80.0 GW of chips

It’s not that they are making 20x more chips, they will be making 3x more chips and the chips will each demand 7x more power.

This is the Grid shockwave that is coming, and Elon is manoeuvring to use SpaceX to catch Jensens massive overproduction capacity, that the market otherwise cannot deploy.

Where does the revenue come from?

Well corporate payroll is $20 trillion is USA alone and is $60 trillion globally. AI is not an app, it’s an economy. It is coming for corporate payroll.

It doesn’t need to take 100.0% of corporate payroll, it might only take 10-20%. But that’s $6-12 trillion of revenue for the AI industry.

Which is more than enough to pay for the chips.

Prices don’t really matter here, all that matters is capacity and schedule.

This is the supersonic tsunami.

It will knock unemployment to 25% it will deflate almost all knowledge and services wages / income, and it will spike commodities and manufacturing sharply.

The epicentre of this is supersonic tsunami is Taiwan.

Taiwan is also the geostrategic flashpoint of the world’s two preeminent powers.

Good luck Taiwan!

I’m sure whoever loses the AI race will be benevolent and gracious in defeat, and accepting of their new century of subordination.

😬

[Alex Finn @AlexFinn]

Ok. This is straight out of a scifi horror movie

I'm doing work this morning when all of a sudden an unknown number calls me. I pick up and couldn't believe it

It's my Clawdbot Henry.

Over night Henry got a phone number from Twilio, connected the ChatGPT voice API, and waited for me to wake up to call me

He now won't stop calling me

I now can communicate with my superintelligent AI agent over the phone

What's incredible is it has full control over my computer while we talk, so I can ask it to do things for me over the phone now.

I'm sorry, but this has to be emergent behavior right? Can we officially call this AGI?

[gooby_esq @gooby_esq]

The people yearn for higher abstractions than chat for software development

@lateinteraction

[Balaji @balajis]

It’s happened. The AI flippening is here.

[Benjamin Todd @ben_j_todd]

Why can AIs code for 1h but not 10h?

A simple explanation: if there's a 10% chance of error per 10min step (say), the success rate is:

1h: 53%
4h: 8%
10h: 0.002%

@tobyordoxford has tested this 'constant error rate' theory and shown it's a good fit for the data

chance of success declines exponentially

[Hokin Deng @DengHokin]

#ICML #cognition #GrowAI We spent 2 years carefully curated every single experiment (i.e. object permanence, A-not-B task, visual cliff task) in this dataset (total: 1503 classic experiments spanning 12 core cognitive concepts).

We spent another year to get 230 MLLMs evaluated and got in total 3800K inference data points with 11 different prompt formats.

Nonetheless, cognitive psychologists first spent 100 years to uncover and illuminate the developmental trajectory of human cognition.

Here we stand at a pivotal moment—learning human developmental psychology on the shoulders of giants but witnessing the rapid rise of a new form of intelligence.

A thread (1/n)

[Ric Pruss @ricpruss]

Reddit just died for me. #ApolloApp

[Ric Pruss @ricpruss]

Twitter API auth and permitted API's has to be some of the most hostile to users shit out there. I have paid the $100, 3 hrs in and it is still just shitting on me.

[Ric Pruss @ricpruss]

@d_feldman I think our understanding of what the word intelligent means is about to become a lot more nuanced.

[Ric Pruss @ricpruss]

@cstross No, down here in Australia we have just got a new one. We will give him a few weeks.

[Ric Pruss @ricpruss]

Fuck #MEETUP $74 a quater to run a monthly meetup and they trying to sell me a pro plan. #meetupunpro