thundergolfer

Work in USA for exposure to good tail outcomes, have Aus citizenship as a refuge from v. bad tail outcomes.

Modal was not hacked. A Modal customer left an unauthenticated web endpoint running on Modal which could be used for code compilation and execution.

See https://modal.com/blog/a-note-on-the-hugging-face-agent-inci... for details.

Modal eng here. Getting the model running is quite challenging, but its accessible right now on Modal via Endpoints: https://modal.com/blog/kimi-k3-by-moonshot-now-available-on-...

Having been early at a startup that relied on Postgres I think this post doesn't put enough focus on monitoring and alerting. Postgres has a few key failure modes that you want to avoid ever happening, and you can use alerting to get early warning that you're danger of it happening.

For example, AWS will send you an email if you're approaching XID wraparound. In a startup that email is very likely to be missed, especially if it's sent on Boxing day. You want whatever AWS is watching to send you that email to be something connected to a pager.

Not really that related. Refinement culture is concerned with evolving aesthetics and marketing which is partly a response to globalization and the rising middle classes of asia, partly related to digitization, and partly just a normal evolution of style.

What George is talking about here is much more related to the ideas of Nick Land, technocapital, Marshall McLuhan, and man's relationship to industrialization.

> Isolation is basically impossible because the Internet follows you everywhere. And it’s perfectly uniform, there is no other Internet, just a place with five corporate towns and some Chinese ones that are really hard to visit if you don’t speak Chinese.

This is McLuhan's "global village".

> I don’t think I’m properly capturing the scope of the machine. First you build the fence to keep the animals out then you build the fence to keep the animals in. It’s a Fullmetal Alchemist homunculus maybe it has already eaten your soul.

This is Nick Land.

Major Sandbox providers (e.g. Modal) run on non-hyperscaler bare metal not AWS and so don't need to markup on AWS's markup. Thus, prices are comparable or better than AWS.

You're talking about AWS Lambda?

- Their use of bare metal isn't necessarily the latest gen hardware - AWS Lambda is part of AWS, and obviously has privileged access to supply

Unfortunately supply is quite limited. If you want to horizontally scale on these instances you need to have a good relationship with AWS so they'll give you a big allocation before c9i is a thing.

> Well, we tried switching back to GitHub Actions, and… yeah it still sucks. Blacksmith has grown explosively because it makes an increasingly frustrating bottleneck in the dev cycle faster.

Where's the "grown explosively" bit coming from? Google doesn't have growth news from 2026, only early 2025.

No a Ponzi scheme involves not output, but here there is very much output in the inference being sold by Anthropic. Pretty big difference.

> That means each employee's AI spending cap is ~11% of that median compensation package.

Probably better to use the fully-loaded cost of the engineer, which is much higher than their compensation package. The fully-loaded cost is the total cost paid for the labor power of the engineer, and it includes big ticket items such as office space, food, equipment, insurance, payroll tax, fringe benefits, recruiting costs.

If the median compensation package is $330k/year then the median fully loaded cost is probably around $450-500k.

This is downvoted but's it pointing out a fundamental dynamic in capitalism. Labour activists had to intervene in this dynamic to protect workers from being exhausted by the constant need for capital to increase labour exploitation to increase profits.

Almost this entire thread is people discussing a labour issue with no reference to the fundamental antagonism between labour and capital.

“Opt-in by default” is just Opt-out. We already have the term.

https://teachyourselfcs.com/ is an excellent list

I just can’t handle this guy’s “it’s a tulip bubble!” approach when just yesterday Thinking Machines released a demo which would have been taken as literal magic if it was shown 20 years ago.

The tech is astonishing and Ed spends 98% of his writing acting like the industry is a scam.

Completely agree that it's off-putting. The author indeed has only ever worked in academia per his LinkedIn.

But disagree that this is a path to unemployment. At work we go very fast and yet I think fast is compatible with each of those points, just not in all situations.

Marc Brooker, distinguished eng at AWS, gives much more useful advice for industry, as you'd expect given his almost 30 years in industry.

https://brooker.co.za/blog/2026/03/25/ic-junior.html

> AI-accelerated tradecraft. The CEO publicly attributed the attacker's unusual velocity to AI augmentation — an early, high-profile data point in the 2026 discourse around AI-accelerated adversary tradecraft.

Attributed without evidence from what I could tell. So it doesn't reveal much at all.

This article is a brilliant skewering of the 'em dash means LLM' heuristic as a broken trick deployed by those too-clever-by-half.

1. https://www.scottsmitelli.com/articles/em-dash-tool/

I think a bigger blocker is that this is a read-only environment for the computer when we need readwrite.

It’s fantastic that computers can be so effective at this read-only work but so much of what I do needs write feedback from the machine.

The comments pretending Marvel actors (e.g. Benedict Cumberbatch) are their Marvel characters in other movies (e.g. Sam Mendes' 1917) kill me.

Their algorithm has a toxic positivity problem where they weight positivity so much the most moronic, saccharine crap sits at the top and you'd be hard pressed to distinguish the comments from LLM slop.

Broadcom builds the TPU chip. Google designs it. You can’t avoid partnering with Broadcom if you want TPUs in significant volume .

It’s a geographic legibility, but you could also say that “C. Chavez ave” provides a historical legibility or an ability to see what the city values.

In Melbourne the North-South street go: King, William, Queen, Elizabeth, Swanston, Russell. It communicates the commonwealth history and founding history of the city. Cute, but decidedly less useful and memorable than number avenues.

Yea ok, but move to a numbered city and you notice the boost. I’ve lived in Melbourne, Sydney, and NYC and found Manhattan’s numbering is great.

Lower Manhattan also still has a large non-numbered area so you can get the old charm when you want it.

Dwarkesh gets far more technical and in the weeds than TPBN. It’s very different. I can’t listen to TPBN though it seems fun but I’ll relisten to Dwarkesh episodes more than once.

There’s a potential irony here that a commenter lamenting the decline of education in the West is leaning on the “critical thinking over memorization” trope in contemporary Western education, when that trope has contributed to a decline in educational effectiveness.

The massive success of information retrieval allowed people to trick themselves that they no longer needed to remember things, and remember them easily. They should instead turn focus on critical thinking.

But critical thinking is knowledge based. At least, I buy E. D Hirch’s argument that it is.

Yeah I'd say the critic was most likely affirming the idea that film is a middlebrow medium. Seeing Hamlet at the Globe is high brow, but seeing Hamlet as the cinema is middlebrow.

The ‘brow’ standards have dropped significantly, in a process Fussell has described as the general proletarianization of culture.

For a long time films that would be considered niche and arthouse were middlebrow, because film itself was at best a middlebrow medium.

To people still concerned with the various brows, Marvel films are below low. They are sign of a debased and infantile film culture that caters to childish tastes and merchandising, not art.

At some point, yes, they absolutely would struggle.

Few things push AI bull spirits on me like seeing these kind of (pretty much correct) diagnoses of the challenges of AI in society.

The proposed solutions are utterly fanciful. They rely on the presence of social and political competencies which have almost completely disappeared.

The OP at least points to the plausible outcome of "protocol lockdown" instead of healthy adaption. Ezra Klein recently made a similar point that AI could end up being over-regulated like Nuclear because irresponsible private industry and weaknesses in our political systems cause a chronic allergic reaction in the demos.

This is an aside, but it always irks me when people throw out the "critical thinking" thought-terminating cliché.

> Critical thinking taught alongside AI literacy.

Critical thinking is not a skill unto itself. You cannot think critically about things you do not understand. All critical thinking is knowledge-based. Where one does not have knowledge, they must rely on trust, or in substitution a theory of incentives which leads to a positive outcome without understanding of details and dynamics. But that substitution theory is itself knowledge.

As to "AI literacy", we could have started on computing literacy 30 years ago when it became obvious that computing was going to dominate society. You can't understand AI without understanding computing.