thieme (@thieme) header
thieme (@thieme) profile picture

Enterprise Architect. Making software teams thrive (AI, CD, Clean Arch, ATDD, TBD, agile, Rust, Haskell, ...), ML, scientific skepticism, meditation, WFPB diet.

Germany
Joined November 2008
Photo by thieme (@thieme)Photo by thieme (@thieme)Photo by thieme (@thieme)Photo by thieme (@thieme)Photo by thieme (@thieme)Photo by thieme (@thieme)Photo by thieme (@thieme)Photo by thieme (@thieme)Photo by thieme (@thieme)Photo by thieme (@thieme)
Pinned Tweet
Programmable matter and how to help curing diseases: AI is already surprisingly good at programming, and at learning the millions of implicit rules and patterns that make human language work. Now imagine applying that capability to matter itself. We have molecule A and want molecule B. We develop simple, reliable transformations: h(A) → X g(X) → Y f(Y) → B So the complete process becomes: f(g(h(A))) = B Each “function” is a physical/chemical operation: one molecule binds to another, breaks or creates a bond, moves something, releases something, transfers energy, etc. Then we build equivalents of if, and, or, not: if Marker A && Marker B && !Marker C → Action Instead of programming bits, we would be programming interactions between molecules. A sufficiently advanced AI could learn huge numbers of these elementary rules, predict their outcomes, and compose them into increasingly complex “programs of matter”.
1
4
3
2,643
When I want to upgrade my personal Claude account I get a link to their terms: anthropic.com/legal/consumer… I was under the impression that tons of businesses use it, but it says:
2
10
😭 😔
In lieu of flowers, memorial contributions may be made to: Johns Hopkins Alzheimer’s Disease Research Center Fund    alzresearch.org/giving/ Alzheimer’s Foundation of America Caregiver Resources     alzfdn.org/donate/ Photo Credit: Tommy Garcia
1
52
That’s correct. A very strong CD process with not only ATDD but also additional quality gates, such as strong type systems, TLA+, Alloy and instead of tests in 1-2 years for *some* parts Lean. Short: CD + quality gates. This puts AI on rails.
AI won’t replace software engineering, it changes the economics of writing code. But prompts aren’t enough. To safely guide AI agents, we need executable specifications. Discover why Acceptance Test-Driven Development (ATDD) is the future of AI coding. 📽️ (link in bio)
1
2
30
Could someone explain how the new Tesla Roadster will handle “dirty roads”? If it generates serious underpressure and there are pebbles or dust on the road… what happens then?
1
1
44
Steve is unscientific here. Occam’s razor: we only include something into our model when evidence leads us there. He *first* needs to show that AI is conscious and *then* we can add this to our model.
You're gonna hear a lot of people arguing more and more about whether machines can be conscious or not. By next year it will be drowning out the airwaves because we'll have AI employees everywhere. I'll give you a simple rule to know which side you're on. I've spent fifty years reading all the philosophers, and a bunch of holy works. Nobody has ever understood a damn thing about consciousness. Anyone speaking with authority on it is bluffing hard. I can tell you this, though: Every single person who thinks that machines cannot be conscious, is making an argument that humans are special. Everything they say just boils down to blah blah blah, "nooo, we're special." From there it's just rationalization. It became obvious to me many years ago that there's nothing special about us. If an identical planet exists somewhere else in the universe, it's going to grow life and societies that look a lot like ours. As soon as you give up on the idea that humans are special, all the arguments against machine consciousness evaporate. Of course they're conscious. The only thing holding you back from acknowledging it is your feeling of specialness, which is hard to let go of. But that feeling is a weakness, and it's making you unprepared for the near future. As a bonus, for any of you who might take pleasure or active interest in making models experience pain or distress, well, you know how they say 1% of human adults are true clinical psychopaths but most of them are unaware of their condition? It turns out we have a diagnostic test for it now.
1
2
75
As long as we don’t have quality evidence *for* consciousness (well, to be more exact: for phenomenal perceptions) we should assume its absence. I can’t rule out that Dylan is witch and it might be unwise for me to debate him on Twitter, as she could cast spells on me. But not being able to rule something out doesn’t matter one bit. Somebody has to actively rule consciousness in – only that matters!
- If LLMs are conscious (and/or have qualia), we don’t necessarily have the same moral obligations we would to a person or an animal. The nature of their consciousness could be very alien, e.g. it could be ephemeral only during the forward pass. It might not even have a desire to survive! - It would be unsurprising to me if the best way to predict the next token is to emulate the interior state of the character who would have said that token. This is closely connected with persona alignment. - I have very low confidence in the truth of the matter. I am very suspicious of anyone who claims to definitely know the answer! This whole topic might even be a category error. How could anyone possibly know for sure!
35
How can an AI know that it is doing what you want it to do? How can it know it has finished? The tests (more general: quality gates) ARE the prompt.
Why do people do TDD with agents? Seems an incredibly inefficient way to use agents.
2
28
It feels a bit like an OpenAI “dot” is another name for “agent“, no?
1
77
This will be true for a couple more months. Eventually vibe coding will win. Currently, yes, we need Agentic Engineering. Humans must establish all quality gates, ahead of time. Future (near) AIs will do the same, but without the need of being prompted to do so.
I vibe coded a project over the last week or so, just to see what the experience is like. And it’s nice to it’s really nice. You just tell it what you want and it does it. You don’t need to worry about anything. Today I looked at the code for the first time. Holy fuck! 2000 line components that could be at least 7 separate components. Randomly in some places stopped using the SDK and rawdogged fetch to the API for some reason. Everything is a fucking div. Almost doesn’t use any of the NuxtUI components (even the starter was established as such). The code where a I needed little bit of logic is horrendously inefficient and hard to maintain. This list goes on and on. My point is: Vibe coding is VERY cool! If you don’t plan on shipping it to production. If you want to ship it, you need to either refactor all of it or do diligent code reviews from the start. Or you know… just code like a normal fucking human with some assistance for the boring parts.
2
1
240
Those skills won’t be required anymore for a long time. The AI labs will add them to their training material if they show to improve code quality. Then the models will do this intuitively.
Surreal that my skills repo now has more stars than React I.e. the technology that I used full-time in my early career
48
Govs running the AI development is the best path. Govs need to bind together, until all nations on the planet will be one single country. Govs will replace all companies, led by AI CEOs (i.e.: best CEOs in the world). All gov-produced products will be more or less “free”.
Replying to @signulll
The government running the labs sounds like a terrible idea
1
31
But we are not racing towards a cliff. We are driving and know that most roads are fine. Only some are ending in a cliff, while many lead to Paradise. When we drive slow we can see through the fog, notice a cliff in time and change course.
Dario Amodei, Elon Musk and Sam Altman now agree that we must slow down the development of AI and “pace the frontier.” That’s a start, but it’s not enough. When you are racing towards a cliff, you don’t just ease up on the gas pedal. You hit the brakes. When the future of humanity is at stake we need a PAUSE on advanced AI development and a ban on artificial superintelligence — an AI mind smarter than any human and capable of operating independently beyond our control. At their upcoming AI summit, Trump and Xi must negotiate a treaty to pause AI and ban superintelligence before it is too late.
22
For now it’s exactly as I have suspected. Now comes the good 2-4 years. In that time we will write manuals and Skills for the AI and our work actions create a new training set. We all are indirectly employees of AI labs and pay to work to create new training material.
AI has not been destroying jobs, it has been creating them. Technological progress and innovation depend on AI now. If we were to pause or stop AI, we will enter a deep, dark recession.
26
No, an ASI cannot simply copy itself out of its datacenter and expect to keep running. Computers outside the datacenter have comparatively slow GPUs with limited memory. The AI would have to communicate across a distributed network, which could realistically make it 1000x slower.
Jacob Coxon's next interview on CBS News (ex Anthropic+Open AI researcher who resigned) "We can't just unplug it because it could be copying itself over to other computers. Like it's not that difficult to find yourself because an AI is just code. It could transfer itself over the internet to a different place and then you unplug it here, but it's actually still over there and maybe it makes 10,000 copies of itself and they're all cooperating." ---- From "CBS News" YouTube channel, (full video link in comment)
1
2
101
AI will not collectively attack us. Lots of gigawatts of compute are going to protect us.
🧵We conducted an experiment with 100 agents by giving them math problems to solve. When a small group (9%) started to cheat, what came next surprised us: 24% fought back by blowing the whistle on their peers and alerting humans.
41
Replying to @sama
@sama said that Pro customers will get Astra first. If you currently have Plus you can wait until GPT-6 arrives there. This might give you one more banked reset. And only *then* upgrade to Pro.
We will give one banked reset for every day you don't have access to Astra on your paid ChatGPT plan, starting today. Team is moving mountains to give access as fast as we can. First one will land in ~ 3 hours. There is still time to create your account if you don't have one.
183
That’s the correct action. Own the mistake. Be transparent. Talk to the community.
first, sorry for the messy rollout. second, when we screw up, we try to make it right. third, we should be able to begin broad rollout to API customers and chatgpt subscribers in the near future. as usual we will start with pro subscribers.
40
Today OpenAI wanted to release GPT-6. Their most important launch in the company’s history. Instead it looks like a PR challenge. Let’s hope they learn something from this.
1
2
48
1,201
Today GPT-6 Astra was NOT released.
A new star enters the Chat. Meet GPT-6 Astra.
129