Opinions in my tweets are my own and do not represent the views of my employer, past or present. d-led
Munich, Germany
Joined September 2010
- Tweets5.6K
- Following1.2K
- Followers270
- Likes28.6K
DLed retweeted
This is an old hat. That's exactly what the neural fast weight programmers did in the 1990s (including the 1991 unnormalised linear Transformer). Overview with links to the original papers: people.idsia.ch/~juergen/fas…
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
DLed retweeted
"Discipline" generated a new vocabulary for Crimson; forced the players to abandon clichés & adopt different ways of thinking; used new technology & instruments; and balanced front line & rhythm section with greater equity than the 1973-74 Crim.
➡️ robertfripp.substack.com/p/d…
DLed retweeted
Hooray, the Elixir issues tracker is back to a single page, with fewer than 20 open issues, and only 3 open pull requests at the time of writing! 💪
DLed retweeted
Hooray, Erlang Distribution over InfiniBand for machine learning workflows: github.com/marcnnn/flame_slu…
DLed retweeted
Popcorn v1.0 is out! 🍿
It runs the latest Elixir and is fully compatible with the BEAM, because… it runs on the BEAM! It turns out that with a few tweaks, the BEAM fits in under 1 MB gzipped and runs just fine in your browser. Switching to it also brought major improvements in stability and performance.
Read more in the blogpost: swmansion.com/blog/popcorn-1…
DLed retweeted
We have been fans of Erlang and the BEAM ecosystem for years. The actor model and hot reloading are perfect for the current moment in AI, however lack of trustworthy and high performance isolates inside the VM has been the one gap in the ecosystem that held back serious agent development. Today we are releasing HypErlang - smallest, safest and fairest sandbox for the BEAM that runs native Erlang code. Here's to the next 30 years! Link to the repo in replies.
DLed retweeted
Genies are being designed as a RoboGym for the mind.
Button labelled "Lift." Weight goes up. Down. A counter tells you how many joules you burned.
That's not a workout.
I love that my genie does my toil — reformatting, boilerplate, drudgery. It teaches me nothing.
The struggle is different. Three hours on a train proving a theorem I knew was true changed who I was.
Let the genie carry the toil. Leave me my struggle. I need it. More in my newsletter.
DLed retweeted
Tools of the trade guide disciplined adoption and ongoing progress.
modeler.domo.tools
1. Those who don't use #DDD are generally in a world of hurt within 6 - 12 months. That is self-inflicted pain.
2. On the other hand, those who attempt to use #DomainDrivenDesign often fall into the trap of prefect-is-the-enemy-of-good because they want to spray a can of "DDD Magic" over the whole enterprise.
3. The US has benefited from move-fast-and-break-things mentality and also suffered for it. See above #1.
4. Europe leads interest in #DDDesign but often fail to apply it well.
Given #GenAI or #AgenticCoding or the flavor of the week...
There's never been a better and more urgent time to apply Domain-Driven Design. Use the end-to-end guidance of @kalele_domo by @kalele_io
ALT Domo Modeler by Kalele.io
DLed retweeted
Researchers in @ScienceAdvances have demonstrated that brief physical contact between healthy and bleached coral can restore algae to the bleached coral, leading to their full recovery. scim.ag/4s40Tt3
ALT Two coral samples are shown side by side in a comparison. On the left, labeled "Week 0," the corals appear lighter, with distinct white and brown patterns. On the right, labeled "Week 9," the same corals are noticeably darker with more prominent brown hues, indicating changes over time.
DLed retweeted
Reminder - the entire @BBSJournal collection on 'conscious AI and biological naturalism' is completely #OpenAccess until October 23, 2026 - one more week only! Get yourself some weekend reading 🤓👇🏼doi.org/10.1017/S0140525X250… (click 'save pdf')
🎇 Good news! 🎇 The kind people at @BBSJournal have made the new collection on "Conscious AI and biological naturalism" completely #OpenAccess for a limited period. This include the target article, all 50 commentaries (many of which disagree with me), and my response. Go to the link and click 'save pdf': DOI.org/10.1017/S0140525X250…
For easier ways into the debate, try my @NoemaMag essay on "The Mythology of Conscious AI" (winner of the 2025 @berggruenInst Essay Prize), and this 2026 main-stage @TEDTalks:
noemamag.com/the-mythology-o…
ted.com/talks/anil_seth_why_…
DLed retweeted
“LLM’s simulate the OUTPUT of intelligence. They do not simulate intelligence itself. They do not learn. They are static piles of human data compressed in high-dimensional math. The whole idea they are intelligent is absurd.”
This is Nobel Laureate Geoffrey Hinton showing precisely where the entire AI industry went wrong.
He went along with the trend: “If you have any theory of how the brain works, you can simulate it on a digital computer” unless it’s quantum effects.
This is false.
The brain is a physical learning system that uses the laws of physics to calculate. It *cannot* be reproduced in digital form, nor can it be simulated. The moment you digitize information, you lose the part that makes it real. By converting it to bits—no matter how many—you lose the infinite resolution of reality.
All forms of biological intelligence are 100% ANALOG. This is not an incidental feature of intelligence. It is fundamental to it. We have never simulated intelligence or consciousness on a digital computer because you can’t.
LLMs simulate the OUTPUT of intelligence. They do not simulate intelligence itself. They do not learn. They are static piles of human data compressed in high-dimensional math. The whole idea they are intelligent is absurd.
You will never get an intelligent digital computer, much less a conscious one. The future of AI is analog computing. When computers use the universe to calculate, when they adapt based on input from the physical world, we’ll be getting closer.
Until then, AI is a dead end.
DLed retweeted
So I was trying to understand why pull requests speeding up the Elixir version were being denied by the clankers while they were being accepted to Go/Express/Rust.
It turns out that, not only each project had a different starting point and a different prompt, but also each project also has a different set of verifications that they are required to pass!
For example, Elixir is the only guided to "preserve the pinned Rails reference behaviour", the only with a explicit verification that we are using Redis and draining its queue (so we cannot remove it while others can), and the only one actively enforcing it.
So it turns out that the people telling you to not read the code cannot even use AI to set up fair benchmarks (probably because they would have to read the code in order to do that).
You don't have to believe me, you can do your own research or ask your agent to do it for you. E.g. here is what Claude Code came back with: gist.github.com/josevalim/ae…
Also extracted a separate verification project that's responsible for the benchmarking and checking of new implementations and improvements to the existing ones. The full report from the latest runs is there too. github.com/basecamp/once-cam…
DLed retweeted
See it before you build it.
Workshop: Implementing Domain-Driven Design (nearly sold out)
4 days: $400 (save $1,600)
Dates: 12-15 October
- DDD Strategic and Tactical Design
- EventStorming, Impact Mapping
- Ports & Adapters / Hexagonal Architecture
- Event-Driven Architecture
Three seats remaining, and then their gone. You can still join my workshop that starts Monday. There will never be a lower price, and I'm told my next event has a pricing increase.
Just a taste 👇 Registration below.
kalele.io/iddd-workshop
ALT Vaughn Vernon Implementing Domain-Driven Design Workshop #DDD #DDDesign #DomainDrivenDesign #ContextMap #ImpactMapping #FlowArchitecture #EventDrivenArchitecture #CQRS #EventSourcing #EDA #ComplexEventProcessing
DLed retweeted
I provide 16 heuristics for identifying Subdomains and Bounded Contexts. Even so, you still have to think 🙂
Beware:
- Don't be fooled by cookie-cutter solutions—absolutes, archetypes
- #DDD is not an architecture, or architecture style
- Archetypical models are based on known solutions
- See Cynefin Framework:
- Clear and Complicated: Known archetypes; ordinary is enough
- Complex: Unknown unknowns; requires differentiating innovation
Workshop: Implementing Domain-Driven Design (nearly sold out)
4 days: $400 (save $1,600)
Dates: 12-15 October
- DDD Strategic and Tactical Design
- EventStorming, Impact Mapping
- Ports & Adapters / Hexagonal Architecture
- Event-Driven Architecture
Just a taste 👇 Registration below.
Limited registration--same seat limit as full price.
Avoid this human-generated slop when you build software with
#DDDesign #DomainDrivenDesign #DDD
kalele.io/iddd-workshop
ALT Vaughn Vernon Implementing Domain-Driven Design Workshop #DDD #DDDesign #DomainDrivenDesign #ContextMap #ImpactMapping #FlowArchitecture #EventDrivenArchitecture #CQRS #EventSourcing #EDA #ComplexEventProcessing
DLed retweeted
A right-hand circular motion, thumb and index striking a rumba rhythm, fast. This is "Putugizi," a fingerstyle technique from Malawi, shown by master guitarist Daniel Kachamba.
He said the name means Portuguese, the style likely traveled to Malawi from neighboring Mozambique.
Filmed in 1972 by ethnomusicologist Gerhard Kubik, who spent his life documenting African guitar.
DLed retweeted
Ok, now I understand why people want to get rid of unit tests. Apparently they overrely on mocks and there aren't proper boundaries between the different parts of the system. If that's the case, go ahead! You should have gotten rid of those before AI. :D
DLed retweeted
🧩How can Monty discover on its own that a handle can belong to many different utensils? @cortical_canon discusses merging and splitting object models, and hierarchical Learning Modules for learning reusable object parts without supervision.
DLed retweeted
An ATM protocol parser took 100 lines of C per message. In Erlang, 10. With the bit syntax, 3 or 4 lines you can actually read. The story behind it, and why a payment is just a binary message: beamtheredonethat.com
DLed retweeted
Very exciting feature coming to LiveView: subscribe to a PubSub topic from any LiveView (or LiveComponent!!!) and LiveView will automatically manage the subscription for you. Simply:
subscribe(socket, "user.updates", &handle_user_updates/2)
PR: github.com/phoenixframework/…
