Dirk Roeckmann

AI = Deep Learning + Causal Inference + Symbol Manipulation A I = ∇ w + P ( Y = y | d o ( X = x ) ) + λ #AI #GenAI #CausalAI #KnowledgeGraphs #NeuroSymbolics

Users converse in natural language with #AI. Vibe coders too. Skilled developers should talk in a less ambiguous, concise language in an agentic coding session. First approaches are emerging: DSPy, sudolang, DSLs. I have been advocating for this for long now. Saves tokens too. #SAP #ABAP

You falsely assume that everybody else is like you and you completely leave out that staring at code you didn’t code and don’t understand is a very weak posture by any stretch of the imagination. You can’t let for example use AI to create a motor, the engineers don’t know how it works, then the motor will be installed and shipped. It’s an enigma to me how some assume this is ok for software

With all due respect to Andrew, in his motivational post, he didn't explain why anyone would write code by hand. I can code, but I consider coding by hand a waste of time. So, if I, the one who already knows how to code, consider this a waste of time, why would anyone learn something which is very hard to learn only to then consider it a waste of time, like I do?

We need to tell the ‘good vibes’ from the ‘bad’ ones. #Neurosymbolic paradigms are of paramount importance. #Agentic coding is also not the ‘ozempic of labor costs’ either - smart companies will keep their (skilled) staff get more stuff done and expand #JevonsParadox #SAP #ABAP 2/2

We should not demonize #vibecoding, the agent only reflects the #humanintelligence it’s presented with - it doesn’t have any of it’s own. Vibecoding paired with real coding, engineering and scientific skills is the way to go using #symbolicverification. #SAP #ABAP 1/2

Enterprise software is not ready for #agenticworkflow. Mutable #OOP is problematic. Critical systems are too big and have too many dependencies to be pulled locally. Therefore a #functionalprogramming based on immutable values is needed and a local lang runtime independent of the data. #SAP #ABAP

As auto-regressive token predictors #LLM predict syntax patterns and have zero senantic understanding: esolang-bench.vercel.app #SAP #ABAP

@lxgicstudios @fchollet Being at the frontier doesn’t help you though if you per definition can’t look outside the library - bummer

@fchollet the library gets you to the frontier though. knowing everything that's been tried is half the work of finding what hasn't.

Current AI is a librarian of existing knowledge. Science requires an explorer of the unknown. You don't win a Nobel Prize by staying in the library.

Formal abstractions and not natural language are the lingua franca of #science and consequently of #intelligence. 2/2

Natural language will not continue to be the pinnacle language of prompting. There must and will be an evolution towards a #6GL formal, declarative/functional statically typed homoiconic meta-language serving as the deterministic and formally verifiable substrate to prompt future #AI paradigms. 1/2

The time to learn how to think for yourself was before genAI, if you missed your chance, good luck

The persisting importance of prompt engineering -- and now harness engineering -- is one of the best indicators of how far we are from AGI. A general system doesn't need a task-specific harness. And when provided with instructions, it is robust to phrasing variations.

@fchollet This is why Konrad Lorenz said: “Thinking is acting in an imaginary world”. Our minds navigate counterfactual worlds in the past and in the future.

We underestimate how much "abstract" thought is just repurposed sensorimotor control circuitry. A lot of reasoning is essentially about moving through idea-space the way we move through physical space.

Paper Link: arxiv.org/abs/2510.26130

We have to come to terms with the fact that the rate of tool calling by an #agent is inversely proportional to the capabilities of the underlying core which is the #LLM. If it could do it on it’s own - no need to call a tool. Progress nowadays happens mostly in the scaffolding. #SAP #ABAP

Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached. The result is a blackbox model (the generated codebase): an artifact that performs the task, that you deploy without ever inspecting its internal logic, just as we ignore individual weights in a neural network. This implies that all classic issues encountered in ML will soon become problems for agentic coding: overfitting to the spec, Clever Hans shortcuts that don't generalize outside the tests, data leakage, concept drift, etc. I would also ask: what will be the Keras of agentic coding? What will be the optimal set of high-level abstractions that allow humans to steer codebase 'training' with minimal cognitive overhead?

Natural language will not stay the pinnacle language of prompting. There must and will be an evolution towards a 6GL formal, declarative, homoiconic meta language serving as the deterministic and formally verifiable substrate to prompt #AI paradigms of the future. #SAP #ABAP

First of all #math, #physics are #science. Scientific theories are formal statements about the informal world. These formal statements are expressed in #formallanguages. Formal languages (#algebra etc) are abstractions and therefore #code. These kind of statements made by the two gentlemen @elonmusk and @nvidia CEO Jensen Huang are void of any meaning and can be disregarded with a good conscience🤦

Elon Musk and Nvidia CEO Jensen Huang are both steering students toward physics and math as AI takes over more of the routine work in coding. Read more: linkedin.com/posts/evolving… Jensen Huang recently said that if he were graduating today, he would focus on physics because the future of AI will require a deep understanding of the physical world.

@pmddomingos #NostradamusDisorder befalls a lot of people lately 😂

And I give him 18 months before he’s fired.

Languages like #prolog #ocaml #lisp which are much better suited for #AI will potentially have a huge revival with the advent of #tensorlogic and #textdiffusion models 2/2

#agenticcoding with left to right auto-regressive transformers works well with imperative languages. #declarative and #functional languages being far superior for #AI are step-children for lack of training data and the transformer architecture. That could change with #textdiffusion models! 1/2