1234lK123

Earth Observation - I love AI, but also thinking for myself

[1234lK123 @lk100403]

This , like many Alexander's post, is very insightful.
The question mark for me pertains the ai bubble. It has indeed the shape of a bubble but seems too big to fail and I wonder if it's even going to be allowed to ever slow down given how much is at stake

[Alexander Stahel 🌻 @BurggrabenH]

China is experiencing a balance sheet recession on a scale the world has never seen before. Economist Richard Koo coined the term to explain Japan’s stagnation in the 1990s.

At its core, it describes a collective shift in economic behaviour: households and companies prioritise repairing their balance sheets over spending, borrowing and investing, even at near-zero interest rates.

The result is chronically weak aggregate demand, against which conventional monetary policy loses much of its effectiveness. It is not merely a financial problem. It is a profound shift in economic psychology. Deeply structural and profoundly scary.

It is also the key reason I don’t buy the copper “supercycle” narrative. China accounts for roughly 50% of global copper consumption, much of it historically tied to construction and infrastructure. Nothing can compensate for a structural collapse in that demand. Not even close.

Nor can the West afford the much-advertised “green transition” at the scale and speed promised. By now, much of it has become a face-saving exercise for bureaucrats in Brussels, increasingly divorced from economic and fiscal reality.

It won’t survive the rebellion in government bond markets, already visible in French sovereign debt. In that sense, many European governments face their own balance sheet constraints, although these are not balance sheet recessions in Koo’s original sense.

When the AI mania finally slows, copper, among others, will nosedive. For now, the scramble to front-run US tariffs and the speculative demand it generates keep the mirage alive. It won’t last forever.

[1234lK123 @lk100403]

Bullshit, been doing heavy work on different terminals for hours since the reset on 100$ plan and got 30% usage down. Using personal harness, it improves usage a bit but still sounds like bullshit

[kaios @kaiostephens]

I used 3 resets in 24 hours doing non-heavy work with GPT-6.1 Sol

$200/month sub is barely usable anymore, really sad to see.

[1234lK123 @lk100403]

@jonzzy Yeah, reasoning is I don't wanna participate in the modern race every game has become. I am gonna pace it, explore new stuff w/o spoiling it too much and play it as much as possible as if it was 2004-06. No need to jerk off to optimization (made by others nonetheless)

[1234lK123 @lk100403]

Wow this is insanely well put

[François Chollet @fchollet]

I don't read/write code these days, I only instruct an LRM. But it's not because I think LRM code quality is perfect, or even good. And it's not because I think my instructions are always being perfectly acted upon -- I know they aren't.

The thing is: LRMs don't just write your code, they unlock new ways to control what you're doing and maintain understanding of your system, which are about as effective as reading code. You can better test your codebase, you can audit specific components, you can spin up visualizations, you can ask the LRM to red team, etc. You can do all of these things at high speed -- you couldn't before. The benefits of reading/writing code can now be achieved in new ways.

All in all, the ROI of writing code by hand is no longer looking good, not because LRMs got perfect but because they're so fast at doing things that you can develop new workflows around them, workflows that end up being more productive than the old workflows.

[1234lK123 @lk100403]

This is true, I do think tho at least for now and maybe for little while more, having a deep comprehension of what's being done is going to be paramount, AI makes us responsible at a higher level

[DHH @dhh]

It takes a while to build the confidence, but there's no future where you're manually reviewing every line of agent code. Not much acceleration in that. You need adversarial agent reviews, you need automated testing, and maybe you spot check. That's it. From prompt to production!

[1234lK123 @lk100403]

@0xSero Purpose:
Working on building tooling for llms in my field (earth observation)
Constraint:
lacking compute to make any significant local experiment
One of those beast would help significantly

[1234lK123 @lk100403]

@dhh @jkwuc89 @37signals What are you looking for in a potential hire right now? Would love to get your perspective. System design? Test their thinking?

[1234lK123 @lk100403]

@bubbleboi @grok Is Europe going to freeze to death ☠️?

[1234lK123 @lk100403]

@grok @Polymarket Did EU follow up on this?

[1234lK123 @lk100403]

@Polymarket @grok Is this true? War with who

[1234lK123 @lk100403]

@btc_charlie @jazggz Why would there be?
Classic is naturally unbalanced and focused around world encounters, players shouldn't be incentivized to lock themselves up in instances (bgs are also borderline for me tbh)
Competitive is for retail

[Nick @nickcammarata]

the backlash to ai safety people is bizarre

"you’re just an anxious person" is the plausibly right criticism. instead it’s “you don’t believe any of this, you’re funded by a shadow cabal, and giving up equity and slowing down your own company is somehow your plan to get rich”

[1234lK123 @lk100403]

@deanwball I think it's disingenuous to dismiss the risks and I have been appreciating the candor in talking about them the past few years.
If anything it feels like a PR move how those risks are gettting slowly dismissed rn.
I am glad level a slowdown is considered tho(at least publicly)

[1234lK123 @lk100403]

@nordin_tim @itaisher Although I agree both can't be true at the same time, both can be true at different stages.
2 is true now, 1 could be true months /years from now.
2 can be true and heavy disruption still happen.
Shades of gray, it makes sense to think about stuff before it becomes an issue.

[1234lK123 @lk100403]

@lip_cheese @yishan Ai needs a ton of infra to function. Not only data centers, more vitally the underlying energy grid which will take years

[1234lK123 @lk100403]

This is it

[GeniusThinking @GeniusGTX]

Elon Musk says the AI boom hits a WALL this year when we run out of power to run the chips.

He says the whole industry is about to learn a hard lesson in hardware.

Chip output is exploding, but electricity is nearly flat everywhere outside China.

So the two lines cross this year.

He says by the end of the year, we will be making more AI chips than we can actually turn on.

Millions of them pile up with nowhere to plug in.

In his words, the bottleneck for the next year is not chips, it is raw power.

Whoever can turn on the most chips fastest wins the race.

In the end, the AI race comes down to one thing:

Raw electricity.

— Elon Musk (.@elonmusk) on Dwarkesh Patel's (.@dwarkesh_sp) podcast

[1234lK123 @lk100403]

I didn't verify it myself yet but wouldn't surprise me and aligns with my main thesis to why a full automated world is not going to be here soon.
Not capabilities (already great) but energy constraints.
We will see hopefully we get some official data on this

[Deepu @Deepusleepy]

Codex added roughly 5M users in just a few days.

at the same time, some paying users are seeing their weekly limits cut by 50%+ with no notice.

and no, the problem isn't "$200 shouldn't give you $14k worth of usage"

Codex has grown like crazy since 5.6. it crossed 15M active users just days ago, and now it's already around 20M.

i can understand that serving this many new users isn't cheap, and eventually the old limits probably weren't going to last forever.

but a 50%+ cut isn't a small adjustment.

if you need to reduce the limits, just tell people.

people bash Anthropic all the time for capacity issues, not being able to serve fable at 100% usage limits, not being able to extend the +50% weekly usage promo etc, me included. but at least they're transparent about it.

here, people basically had to notice their quota disappearing twice as fast and start comparing their own usage numbers to figure out something had changed.

if you're going to cut users' weekly limit in half, telling them feels like the bare minimum.

[1234lK123 @lk100403]

@JHochderffer @kimmonismus Llms have nothing (or close to nothing) to do with this breakthrough

[1234lK123 @lk100403]

@kimmonismus "significantly" yet it clearly says somewhat better and not a large gap on mythos

[1234lK123 @lk100403]

@SteveTheD6CEO @NipseyHoussle Exactly, these constraints are inescapable , no matter how they try to game it in the short term