Felix Bach (@FelixBach7) header
Felix Bach (@FelixBach7) profile picture

Science/Tech Nerd, dep. head research data @FIZKarlsruhe NFDI4:Chem,Culture, Memory,Objects #Repositories @RADAR_Service #RDM #RSE. Own opinion, RT≠like

Nerdistan
Joined June 2020
Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)Photo by Felix Bach (@FelixBach7)
Attention @arxiv authors: Our Code of Conduct states that by signing your name as an author of a paper, each author takes full responsibility for all its contents, irrespective of how the contents were generated. 1/
136
894
270
6,431
1,102,221
Your DNA is like a computer program according to @blaiseaguera. Von Neumann figured this out before scientists even discovered how DNA works -- that for anything to make a copy of itself, it needs: * Instructions (the code/DNA) * A machine to read those instructions (ribosomes) * A "photocopier" for the instructions (DNA polymerase) Blaise reckons that your cells are running actual computations to build proteins and replicate i.e. life is "software" running on hardware.
9
34
4
181
15,818
Felix Bach retweeted
“In the beginning was the Word. Then came the fucking word processor. Then came the thought processor. Then came the death of literature” — Dan Simmons: Hyperion, 1989 (anticipating the LLM)
22
33
1
399
They killed this one
131
8,920
243
197,470
4,011,036
Couldn’t make it to every session at #OSFair2025? No worries…recordings are now online! Hosted at @CERN, you can now catch up on all the talks & panels here: 🔗 cds.cern.ch/search?f=490__a&… #OpenScience #OSFair2025 #CERN #OpenAIRE
6
3
496
Today gives this video a whole new weight. Michio Kaku's famous statement on US H1B Visa He is a renowned theoretical physicist, known for his work on string theory.
H1B US Visa now costs $100K per year. This new rules begins in a week. SF Bay Area is about to feel the disruption. If you are outside the U.S. and need to start or resume H-1B work, your employer must budget for the $100K per year.
117
208
57
1,445
302,105
True. And computer science is so much less than 'Informatik'. But it reflects what you learn during the bachelor of computer science now.
Calling informatics ‘computer science’ was like naming psychology ‘head science’ or economics ‘money science’.
39
The binding energy of atomic nuclei has contours that seem to map some weird underlying grid structure. This is deeply weird, or is it?
6
3
1
357
21,763
Very well done article about current attempts to test the quantum properties of gravity by @dcastelvecchi
58
30
5
292
24,842
Felix Bach retweeted
Quantum computing is the largely speculative attempt to create better error correction on the rulial substrate than naturally occuring particle dynamics can provide
31
8
3
166
14,781
After some more thinking: Intelligence is the ability to model intelligence. @Plinz ?
1
1
49
MIT researchers pulled off something rare, they updated an LLM’s weights during test-time and saw a major performance boost. Usually, models are trained on massive datasets, fine-tuned on domain-specific data, and then frozen. Once in production, even a thousand real-time examples won’t change how the model behaves. But MIT flipped that. They converted examples in the system prompt into real-time training data, effectively teaching the model as it runs. The result? An adaptive model that gets more accurate with each query. It’s early days. The examples are still too domain-specific, and there’s no universal way to generate them. But the direction is exciting.
13
31
5
140
13,487
we need to do something about democracy
unavailable
1
2
837
Felix Bach retweeted
Today, we release LEAP, our new developer platform for building with on-device AI — and Apollo, a lightweight iOS application for vibe checking small language models directly on your phone. With LEAP and Apollo, AI isn’t tied to the cloud anymore. Run it locally when you want, for speed, privacy, and reliability, using LEAP’s end-to-end toolkit for on-device AI. 1/ – at San Francisco, CA
12
44
14
281
1,677,460
Felix Bach retweeted
New Paper! Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents A longstanding goal of AI research has been the creation of AI that can learn indefinitely. One path toward that goal is an AI that improves itself by rewriting its own code, including any code responsible for learning. That idea, known as a Gödel Machine, proposed by @SchmidhuberAI over two decades ago, is a hypothetical self-improving AI. It optimally solves problems by recursively rewriting its own code when it can mathematically prove a better strategy, making it a key concept in meta-learning or “learning to learn.” While the theoretical Gödel Machine promised provably beneficial self-modifications, its realization relied on an impractical assumption: that the AI could mathematically prove that a proposed change in its own code would yield a net improvement before adopting it. Sakana AI, in collaboration with Jeff Clune’s lab at UBC, proposes something more feasible: a system that harnesses the principles of open-ended algorithms like Darwinian evolution to search for improvements that empirically improve performance. We call the result the Darwin Gödel Machine. DGMs leverage foundation models to propose code improvements, and use recent innovations in open-ended algorithms to search for a growing library of diverse, high-quality AI agents. Applied to practical tasks, we implemented Darwin Gödel Machine as a self-improving coding agent that rewrites its own code to improve performance on programming tasks. It creates various self-improvements, such as a patch validation step, better file viewing, enhanced editing tools, generating and ranking multiple solutions to choose the best one, and adding a history of what has been tried before (and why it failed) when making new changes (see the attached video). We believe that Darwin Gödel Machines represent a concrete step towards AI systems that can autonomously gather their own stepping stones to learn and innovate forever!
36
199
24
1,031
105,158
Introducing *ARC‑AGI Without Pretraining* – ❌ No pretraining. ❌ No datasets. Just pure inference-time gradient descent on the target ARC-AGI puzzle itself, solving 20% of the evaluation set. 🧵 1/4
35
176
38
1,293
226,607
Felix Bach retweeted
Meta FAIR researchers, in collaboration with @bcbl_ are sharing two breakthroughs that show how AI can help advance our understanding of human intelligence. 1️⃣ Research that successfully decodes the production of sentences from non-invasive brain recordings. 2️⃣ A study toward understanding the neural mechanisms that coordinate language production in the human brain.
63
116
32
511
51,293