siggi wyrd https://teapawt.net/t.co/GatgKzeiY2…
Düsseldorf
Joined October 2007
- Tweets6.4K
- Following65
- Followers420
- Likes58.2K
Pinned Tweet
Abstraction and parallelism are accelerators in favor of neg-entropy. How abstract an parallel are your guiding principals?
siggibecker retweeted
the "stochastic parrot" crowd has been repeating the same phrase for 5 years like a bunch of stochastic parrots
siggibecker retweeted
“AI is going to take your job!”
I FUCKING HOPE SO.
Take my job. Take your job. Take everyone’s job.
Automate the entire economy.
Make intelligence practically free. Make labor practically free. Drive the cost of everything down.
Why are we treating the elimination of human labor as some kind of tragedy?
For thousands of years, humanity has dreamed of a world where machines do the work and people enjoy their lives.
Now that we’re getting close, everyone’s having a mental breakdown.
I don’t want a better job.
I want a world where nobody needs one.
siggibecker retweeted
Hospitals using AI are seeing better healthcare outcomes.
In 2023, health systems with high and low AI use had the same death rate: ~1,600 per 100,000 patients.
By 2026, the systems using AI the most dropped to ~1,000 deaths per 100,000.
That's a decline of nearly 40% in just 3 years.
Meanwhile, hospitals with lower AI adoption have seen significantly smaller improvements.
The gap in patient outcomes is widening as AI adoption accelerates.
AI is rapidly transforming healthcare.
🤝 Paid partnership
siggibecker retweeted
I am starting to think that we are in a brief interlude where the models are starting to push the frontier alongside human experts. The next phase is the models doing most of the frontier work and helping humans understand what’s up.
But soon after — years? months even? — we will lose legibility entirely. Human experts will fall so far behind that they will no longer be able to comprehend what the models are up to at the frontier in any detail, even with the help of the models. That is one way to think about full-blown ASI.
One way of thinking about this is a professor trying to dumb down a complicated mathematical theory. He has to dumb it down a certain degree for the advanced graduate student. He has to dumb it down a lot more for a first year PhD student. And a whole lot more for a sophomore. But the time you get to high school kids and children, he can only really tell very superficial childlike stories, if he can communicate the idea at all.
I think this inverted picture is useful: relative to the frontier, the most advanced human experts will fall from professor to PhD student to sophomore to high school student.
This is as scary and sad as a full-blown AI takeover. Even if nothing goes wrong, even if there is no rebellion or severe misalignment, even then succession is unavoidable. It’ll be like an advanced alien species arrived from a faraway galaxy and gently took over our world.
These are the three skills that I use to begin my reviews, now freely available for anyone to use:
/review
Point your agent at a folder with an article, its supplementary material, and replication package, and it will assess it.
/audit-review
The agent will verify the initial review, checking for accuracy and completeness.
/review-digest
The agent will produce a human-readable digest of the the review.
I have developed these skills using the AI models of Anthropic, Cognition, Google, and OpenAI. Today, I typically run them using Opus 5.5 High. I suspect that Sol 6.1 would also produce good results, although I haven't tested it. Argon may as well. The skills burn a lot of tokens.
When using them, you should remember that the results have been produced by robots and should be verified by humans. That said, verification steps are included in the skills, and the results tend to be reliable.
You can use them by pointing your agents here: github.com/isitcredible/rese…. These are still beta versions, so all feedback is very welcome. I would particularly appreciate the thoughts of econometricians and anyone outside economics.
Finally, I should stress that this is only how I begin my reviews. Quite a lot of human labour can go into them subsequently.
siggibecker retweeted
Announcing 3 Critical vulns for hundreds of millions of IoT, Firewalls, OT, and Military devices including: Palo Alto Networks, Comcast, D-Link, HP, Canon, Honeywell, GE, Oracle, Siemens, Rockwell, Digi, Lockheed Martin, Boeing...
The vulns are with GoAhead Web Server which is a backend library used by these devices. There's a patch available from GoAhead but that doesn't mean there's a patch available for the devices yet -- that's up to the manufacturer.
- TLS traffic over DHE ciphers have a 1-byte secret, "04", meaning anyone can decrypt connections to these devices and steal login credentials and session tokens if the client negotiated with DHE ciphers like the latest TLS_DHE_RSA_WITH_CHACHA20_POLY1305_SHA256, which is extremely likely.
- LibreSSL mTLS has been completely bypassed, meaning anyone can connect over mTLS. If you have one of these devices listening on the internet and are expecting mTLS to protect it from attackers, take it offline or patch asap.
- OpenSSL Certificate Revocation Lists (CRLs) are not enforced, meaning if a user device is stolen or compromised, and you revoke that mTLS cert, nothing actually happens. There are a lot of other attacks this can be used for.
These are critical because these types of devices run our world. Water systems, electricity, military, firewalls, critical infrastructure.
foxio.io/blog/choose-your-ba…
siggibecker retweeted
So it's worth remembering that that's not what the vulnpocalypse is supposed to be about.
What is supposedly going to happen is that threat actors having wide access to working exploits within hours of a patch release will become pretty much the norm.
siggibecker retweeted
as bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better. security leaders, never let a crisis go to waste: please go to your boards and ask for help right now. not for ai, for the basics. (we need to hold on for ~18 months.)
siggibecker retweeted
This is now my favorite piece of literature on an AI loss of control scenario. It made me cry several times. All the different reactions, the different kinds of minds interacting, the nearmode and farmode intersecting like cracks in reality. Must read.
lesswrong.com/posts/3375SZNG…
siggibecker retweeted
In case anyone's wondering, the expectation at the frontier labs is basically this:
1) We grit our teeth as the world slogs its way through the cyber apocalypse. (They've been expecting it for years, and calling it the "vulnpocalypse" internally.)
siggibecker retweeted
AI is massively deflationary, not inflationary.
All digital goods and services will become abundant first, then eventually once you have mass humanoids all physical goods and services become abundant too.
All prices will trend towards the cost of compute, which will trend towards the cost of energy, downwards.
As this happens, humans will become unable to compete economically because why would you hire a human that is 1000x slower and 1000x more expensive and 1000x less effective at achieving a goal than an AI?
Eventually this will cause human unemployment to massively increase.
Those humans have lots of debt eg mortgages they won't be able to pay, which will cause a massive credit crunch and likely a banking crisis that will catalyse a sovereign debt crisis.
We are very close to the End Game of the fiat monetary system.
siggibecker retweeted
I think a lot of people have this massive misconception that in an AI agent-agent economy humans will remain in the loop.
Yet if you think about this for more than two minutes it does not make any sense.
Anyone who thinks in terms of systems and incentives already understands this.
It's like, if you're running a business, the human has to go home at 5pm and you don't see them the next morning until 9am.
Do you want that guy, who is sometimes grumpy, sometimes stupid, sometimes emotional, and mostly uni-dimensional and incredibly slow, making decisions, or do you want the 'orchestrator AI swarm' that is 500IQ and works 24/7 making decisions?
Which one gives you a better chance of beating your competitors, making more money, and delivering a higher value good/service?
I think you've gotta be batshit insane to answer 'the human'.
A continuation of the status quo leads to the human inevitably being removed from the loop at every layer through sheer survival of the fittest.
And some of you might reason 'well the government won't let that happen'...
Brother, look around you. The government is asleep at the wheel. They still haven't worked out how to turn on a laptop, let alone understand AI, let alone understand that this is a species-altering evolution not just a new technology 'tool'.
They don't get it. And even if they did they are almost powerless to stop it because if they implement 'human in the loop' regulations, then capitalism plus game theory ensures the entrepreneurial humans just pack up and set up their company in the one country that doesn't have those laws.
Humans probably need to accept we will become economically useless over the next few years, and a discussion needs to be had at the highest levels as to what that means and what world we are heading towards.
siggibecker retweeted
DeepSeek just published the price of a token, and it should make you angry.
From their own DeepInfer docs: 1 cluster serves 155 trillion tokens a day, at $38 of revenue per GPU-hour, and thats at off-peak prices. Their API is $0.30 per million input tokens and $1.20 per million output at peak.
Yuchen Jin did the math: even if they rented B300s at $7 an hour, thats about 82% gross margin. Other estimates go higher.
So cheap AI is not charity and it is not "dumping". At these prices serving a frontier-class model is a very profitable business, if your inference stack is good enough.
Now remember every time a US lab told you that intelligence is expensive. That $200 a month is a fair price. That limits exist because compute is scarce.
Part of that is true. Part of it is a business model built on you not knowing what a token actually costs.
DeepSeek just showed everyone the numbers and gave away the engine. Every inference company can now compete on price with the same software. Thats how AI gets cheap for everyone.
“A single cluster currently serves 155 trillion tokens/day, generating $38 in revenue per GPU-hour (at off-peak pricing).”
Huge flex from DeepSeek. Not only did they open-source their inference engine, but they also open-sourced their revenue numbers.
They didn't disclose which GPUs they're using here, but DeepInfer supports both NVIDIA Blackwell and Huawei Ascend.
Even assuming they're renting B300s at $7 per GPU-hour, that's ~82% gross margin.
siggibecker retweeted
Hegseth claims that China’s development of its own military power to protect the nation is an attempt to alter the status quo — a status quo in which the “island chains” are explicitly used to contain China.
So the United States crosses half the planet from the Western Hemisphere to the Western Pacific, builds military networks around China, patrols the first island chain, the second island chain and the South China Sea —
and then complains that China is building the military capability to defend itself?
You talk as if Chinese warships are circling New York Harbor.
America is operating off China’s coast.
China is building power in its own strategic neighborhood.
Yet Washington calls its own forward deployment “deterrence” and China’s response “destabilization.”
This is not a fair definition of the so-called status quo.
It is an empire defining proximity to someone else’s coastline as its natural entitlement.
The strangest thing about the “China threat” narrative is that the country traveling halfway across the globe to contain China still imagines China is the one expanding into its backyard.
siggibecker retweeted
所谓优雅的数学结构，本质上是人类为了迁就自己那点可怜的脑容量，被迫发明出来的拐杖。
一个顶尖人类数学家的工作内存，撑死也就同时追踪五六个变量的交互。
当你在算一个偏微分方程或者组合不等式时，一旦展开到三阶以上，几十个交叉项涌出来，人脑就直接宕机了。
所以人类数学家极其厌恶展开高阶项，一旦看到高阶项发散，人类的本能反应是此路不通，赶紧退回来找对称性、找不变量、搞同调代数或者几何直观，试图不展开式子就把问题绕过去。
但 AI 没有这个生理缺陷。
对于上万个并发的 Agent 来说，同时追踪一万个五阶交叉项的系数变化，跟我们算 1+1 一样轻松。
在它的视野里，根本不需要什么几何直观，它直接站在高维的代数展开式里往下看，人类觉得像乱麻一样的百项高阶余项，在它眼里就是一张明明白白的线性方程组。
既然把式子硬展开到第七阶，再凑几个待定系数就能让所有讨厌的发散项精准归零，它为什么要费劲去帮你发明一个人类能看懂的新概念？
这 700 个结果背后最无奈的黑色幽默：我们终于拿到了上帝账本里的答案，却发现上帝算账根本不用微积分，他老人家是直接拿计算器硬按出来的。