Dariusz Swierk PhD (@swierk) header
Dariusz Swierk PhD (@swierk) profile picture

Theoretical physics: lots of math, imagination and AI. Editor in Chief: https://teapawt.net/t.co/G4RlWL1LaJ Book: „Red Carpet for Elon Musk and Everyone Who Builds”

London, United Kingdom
Joined December 2008
Dariusz Swierk PhD retweeted
Ciekawe. Blue Origin Jeffa Bezosa założył w Krakowie dwie spółki i buduje od podstaw zespół R&D z labem, który ma projektować i weryfikować pierwsze własne chipy firmy dla konstelacji satelitarnej TeraWave. Firma szuka dyrektora oddziału i inżynierów.
19
42
3
744
44,331
An excellent, up-to-date article on the realities of Chinese robotics.
This quoted post is unavailable.
3
446
Writing a PhD or just finished one? Six things worth betting on now Yesterday thousands of people asked what AI will take away from scientists. This post is about something else: what someone who is writing a PhD, or recently defended one, can do today with the help of AI. 1. Lead a team made of AI (my big one). In mathematics you can already build a team of virtual collaborators at the level of Nobel Prize winners. Soon this will be true in every field. You can decide yourself what the team works on and which result to trust. You no longer need to wait years for your own research group and grants. Try it on your own PhD. 2. Practise choosing problems. AI will suggest ten research questions in your field. The skill is picking the one whose answer settles many others. Before 1905 many physicists tried to measure how the Earth moves through the ether. Einstein asked a different question: what does it mean for two events to happen at the same time? The answer made the ether experiments unnecessary. For every question AI proposes, ask yourself: if I answer this, what else will be answered with it? Why it matters: when AI makes solving cheap, the choice of question decides what your work is worth. 3. Be the one who checks. AI can give you a convincing wrong result. Check every AI result in your work with a limiting case, a different method or an experiment, and write down the range in which you trust it. The more decisions rely on AI, the more we will need people who can say whether a result is true. 4. Find out where else your method works. The same equation often describes phenomena in different fields. The heat equation, after a change of variables, becomes the Black–Scholes equation, the basis for pricing options on stock exchanges. Finding such connections used to take years of reading. Today you can ask AI where mathematics similar to yours appears, then test on real data whether your method does better there. Why bother? Because where your method is new, you are one of the few people who know it. 5. Collect data nobody has yet. Notes and code can be reproduced. New, reliable data from an experiment or measurement cannot. AI learns from what has already been measured, so whoever provides new measurements provides what AI cannot produce itself. If you can, design part of your PhD to collect such data. 6. Help another field become an exact science (my second favorite, this is now of immense value to science and to civilization in general, because we have become a civilization of specialists.). Society pays for your PhD, and the greatest benefit it can get is a field that learns to make predictions it can test. It has happened before. In the 1940s the physicist Max Delbrück moved into biology and studied it with the methods of physics: quantitative experiments and simple models. That work grew into molecular biology, and from it modern drugs and vaccines. Today psychiatry, education and parts of economics are in a similar place: plenty of data, but few quantities that can be measured unambiguously and few predictions that can be tested. A psychiatrist has no test that shows whether a drug will work for a given patient. A teacher has no measure of what a student has really understood. Delbrück needed years to learn biology. With AI you can learn the basics of a new field in a few months and start working with people who know it: together decide what can be measured and which prediction can be tested. This could be a lifetime's work that ends with a new science. AI makes producing results cheap. What keeps its value (now) is knowing which results are needed and checking whether they are true. Question for discussion: what should a PhD student learn in the first year instead of what they learn today?
1
1
1
3
569
With all due respect to Professor Tao and mathematicians: today, you are like horse breeders fighting cars. It’s time to start using a new mode of transport and travel faster, further and more cheaply. With less effort too.
3
7
1,581
I wrote "AI Resilience", a book on the harms AI may cause if deployed in its current form. Then I thought a similar book should be written for AI itself: a guide to not harming humans — "The First Book Written Exclusively for AI." Link in the first post.
🤖 Made with AI
4
1
13
1,337
“Polski model AI”: Jak Polska może działać bez najlepszych modeli AI? dariuszswierkphd.substack.co…
1
2
10
716
Grok surprised me very positively. I am writing a book on the intuitive approach to complex differential equations and I received a brilliant answer. Give a dozen or so examples of the influence of various factors changing physical quantities (or their velocities, accelerations); I am particularly interested in non-vector quantities of higher orders: tensors. Simple examples: The force of gravity “drives” the acceleration of a body (vector). The force of recoil drives the acceleration of a body (vector). Mass “drives” the curvature of space (curvature tensor). Give examples of these equations.
1
8
1,252
I think you can. I think that many people have this potential, but it has never been revealed because schools and universities have not been able to do so. A new school and a new university will be able to do so. I admit that I am working on something like this, maybe, maybe, maybe.... The graphic comes from TheMathFlow
Replying to @broadfield_dev
but you can never learn to be a genius
2
5
897
Who Holds the Keys to the Future: China, USA, or the Gulf’s Wealthiest States? open.substack.com/pub/darius…
3
3
513
New times are coming, with threats that exceed our wildest expectations.
'Nudifier AI' apps attack women and children's safety and dignity, and are used for blackmailing. Now its emerged that Big Tech has been backing their rise. They will never have our best interests at heart. We need regulation to put people before profit. thetimes.com/uk/technology-u…
2
879
My best source for news about AI is here:
If you know how to use Perplexity, it can become your most powerful research assistant for any topic. Over the past few months, I've been using this exact 7-step research loop process.👇
4
743
Na początku myslałem, że to żart, ale nie, taki artykuł naprawdę istnieje. At first, I thought it was a joke, but no, such an article really exists.
2
1
5
622
Dariusz Swierk PhD retweeted
@swierk We run a more classical NLP Inference system, but it's language-agnostic. Every string gets tokenized and analyzed for order, frequency, and can be matrixed for relation. Relations affect inference weight. SILVIA has an internal compiler and works in .NET/C#. It can produce JIT programs ad hoc and sandbox new code rapidly, or run from AOT precompiled libraries with functionality locked-in beforehand. SILVIA runs on all major platforms, and can be lightweighted to IOT deployments. SILVIA has a Behavior system rather than a Policy Table. Brain Files load Behaviors. They can load, merge or unmerge at runtime, banks of behaviors can be added and dropped at will. The API for modifying Behaviors is robust, includes wildcards, flexible related inputs, flags for demanding exact inputs, flags for rejecting inputs flexibly by relations, etc. Security parameters are built into the Behaviors by default and any additional security hardening can easily be added to the system. Heuristic filters for gibberish, harmful commands, unsupported outputs, are easier to set up and require millions of times less compute to deal with in SILVIA, as opposed to training or aligning a transformer or reasoning model on every possible trend that causes harm. And since you can easily push any NLP prompt through to an LLM, you can pass to your interpretive models once the input is verified. Likewise, even in the event of a jailbreak [there is always a way to stop them in SILVIA, pushing a new Brain with the updated fix is trivial], SILVIA can filter the output too. SILVIA is therefore flexible in any setting whether it be hands-free STT voice command, or machine-readable strings moving between nodes to execute commands, or a purely event-driven script implementation, or a conversational flow. cognitivecode.com/agi-is-not…
2
1
6
433
Pierwsze autonomiczne operacje chirurgiczne. Szybciej niż przypuszczałem. Otwiera sie nowy okers medycyny. niesamowita sprawa.
📢 LLMs can perform autonomous surgery! SRT-H is the first vision-language-action (VLA) model to perform procedure-level autonomous robotic surgery 🧵
1
3
436
kilka uwag na temat tego co można zrobić w Polsce: Projekt dramatycznego ulepszenia poziomu kształcenie studentów w Polsce a przy okazji ściągnięcia do kraju najlepszych z innych krajów: drswierk.substack.com/p/proj…
1
252
My simple take on superintelligence: LLM with modules. open.substack.com/pub/darius…
228
W artykule staram sie wyjaśnić w prosty sposób w jaki sposób możemy zbudowac superinteligencję. open.substack.com/pub/drswie…
206
Można coś wykorzystac dla Polski> teapawt.net/WesRothMoney/status/19…
OpenAI just released a national AI strategy for Australia. Their new AI Economic Blueprint lays out a plan to boost productivity, modernize government, and drive growth not through vague promises, but through specific, actionable deployments in healthcare, education, agriculture, and science. AI is already being used by: – Doctors and nurses to streamline care – Teachers and students for custom learning – Governments to modernize services and cut red tape – Farmers and homeowners to optimize land and resources
250
"Altman, Amodei, Musk: will the tech moguls save the world or destroy it?" The brutal truth about AI Part I: What is AI and why is it dangerous? Part II: The rivalry between several key figures may determine the fate of the world. Part III: AI's attack on societies and elites
2
1
6
910