Lesmes retweeted
Llegiu tot l'article, és molt aclaridor sobre la falsedat eco en que vivim...
Unes anàlisis de laboratori confirmen que ACIN Recycling Metals està abocant concentracions d'alumini, zinc, crom, arsènic i níquel molt superiors als límits màxims autoritzats a una riera a Arbúcies directa.cat/entitats-darbuci…
Lesmes retweeted
I'd like to share a few thoughts and observations I've seen on my timeline here, I find it interesting when people wonder whether software is becoming a solved problem, whether AI will eliminate the moats in software, or whether there is anything left to build.
LT;DR. Software is still the MOAT. let me share my side of the story.
I never assume malice when I read these posts. I understand where the concern comes from, and I want to share how I see it.
1. Excel might be the best example of why simple primitives win.
Excel is arguably one of the most successful pieces of software ever built. Why? Rows and columns give people a simple way to organise data, build models, run calculations, and present information however they want.
The power is not in what Excel does out of the box. It is in what people can make it do.
Much of the software industry has been trying to improve on that basic idea: give people primitives they can use to solve their own problems.
2. Software factories are about personalised software.
I often see the term Software Factory to describe a new way of producing software faster and thousands of PRs, but i see software factory as more of its that is personalised, adaptable, and shaped around the needs of the people using it. look how @badlogicgames did with pi and @steipete did for openclaw.
Until now, if your software did not do what you needed, you had to wait for the vendor to ship an update, pay someone to change it, or build something else.
That constraint is starting to disappear.
You can change the software yourself. You can make it fit your work, rather than changing your work to fit the software.
3. Being able to build something does not mean everyone will.
Walk into Home Depot and look at all the circular saws. The tools to build things are available, but not everyone buys one. Not everyone wants to build their own furniture, either.
Some people want to make things. Others want to buy them. Many will do both.
Software will be no different. The ability to build your own tools does not eliminate the market for products. It gives people more choices.
4. Software is becoming accessible to a new group of builders.
What is software, after all? It is a way to turn needs, ideas, and processes into something a computer can execute.
For decades, the cost and expertise required to produce software created a high barrier to entry. AI is lowering that barrier.
People with good ideas can now build tools for themselves, their families, their communities, and their businesses. They do not have to convince a vendor that their problem is worth solving.
That creates a much larger group of people who can participate in the software economy.
5. Software 3.0 opens up more problems to solve, not fewer.
@karpathy and others have written about Software 3.0. Whatever terminology we settle on, we are entering a new phase in how software is created and used. or @tobi with Qmd, out of request, it would cost him way more than a few prompts and tokens, it would cost more money and more time to build.
One vendor, one model, or one agent harness will not serve every need in this new economy.
Think about it for a second. If AI agents are already solving so many mundane problems, why are my friends working at frontier labs and AI startups sleeping less?
Because we are working on a new set of problems. The old constraints are falling away, and we are discovering what becomes possible when they do.
6. Every layer of the stack can become adaptable.
Think about the entire chain, from the software used to design a transistor to the software powering a fighter jet, an industrial machine, or a mobile app.
Every layer has the potential to become more adaptable to the needs of its users.
That includes the models themselves. We should be able to distil and adapt models to the needs of the people using them, including through private and local deployments. The data and knowledge used to train these systems came, in large part, from human activity and collective work. We should be building ways for people to benefit from that intelligence directly.
7. New abstractions create new jobs.
Look back at the rise of reactive architectures and digital transformation. People worried about how new tools and architectures would change the roles of frontend and backend developers.
The industry did not stop needing software engineers. New abstractions helped create the full-stack developer, a role that could work across boundaries that previously separated teams.
We should expect similar changes again. Some jobs will disappear, others will change, and new roles will emerge around problems we have not yet learned how to solve.
Making something easier to build removes a constraint. It also creates room for new products, businesses, and careers.
Consider dental floss and toothpicks. Dental floss did not make toothpicks obsolete. In Brazil, Gina has built a major toothpick business, with reported production of around 40 million units a day.
A better or newer solution does not automatically eliminate the demand for an existing one. People have different needs, habits, and preferences.
8. Taste and understanding the user become the moat.
If everyone can build software, simply having the ability to build it becomes less of an advantage.
What matters is understanding which problems are worth solving, how people want to solve them, and what makes a product worth returning to.
That requires more than waiting for feedback. It means observing how people use software, where they struggle, what they change, and what they build on top of it.
It means giving users the tools to adapt the product themselves, while learning from those adaptations to improve the experience for everyone.
This is why I started with Excel. Its lasting value is not just its spreadsheet. It is the freedom it gives people to create things the original developers never anticipated.
I believe the next generation of successful software companies will understand this.
The opportunity is not simply to build software faster. It is to create software that people can make their own, let them choose, ship in a way they can truly own their data and the way you ship will get the attention it deserves.
The barrier to creating software is falling. The number of problems we can solve is growing.
And I think we are still at the beginning.
- I wrote about the future of new roles soon, xitter.cf/compose/articles/edit/… and @GeoffreyHuntley recently wrote about interviewing process for this new kind of role here xitter.cf/GeoffreyHuntley/status…
gist.github.com/igorcosta/bb…
🗞️ if your team is too busy doing their 'normal job' to experiment with AI, you're preparing them to be replaced
ghuntley.com/replaced/
In this post, I guess this's the last time I'll say it. AI use is no longer optional if you wish to be employed. What follows is advice for current employees and hiring managers on how to hire in 2026 to source AI-first candidates.
Lesmes retweeted
Real world example. I bought a bunch of electronics from Ali Express, fed it to classcad AI inside Claude: here you go, find the exact dimensions in the internet and build me a nice housing around it so everything is seated nice and tight. gave it some design clues, and out came ...
Lesmes retweeted
Inference is the second way to raise a massive round right now. After the labs, you can still justify massive rounds by pointing at data centers and compute, which means more inference companies keep getting funded even as the underlying product stays extremely easy to switch away from. On the flip side, many neolabs are also renting out their own compute, adding yet another source of supply into the same market.
Companies that tried to verticalize around batch workloads and post-training are now pivoting back to "pure inference" and joining the same undercutting cycle. The result is a permanent reverse auction: providers undercut each other over and over: half off Fireworks, half off Baseten, 10% cheaper GLM or Kimi—and every new entrant or neolab offering capacity just accelerates the cheaper pricing
At the same time, app companies have so much choice right now that they can bake off and compete prices down extremely hard. They move millions of run-rate spend overnight whenever a meaningfully cheaper option appears, because the technical switching cost is essentially "one line of code".
That combination: easy capital, extra compute from neolabs, low stickiness, and continuous price pressure makes inference one of the less defensible layers in the stack even as it continues to attract large valuations.
However, if you are a smart team of infra people, it is the easiest way to raise 100M seed after a neolab
Researchers: neolab <> engineers: inference
Lesmes retweeted
This is absolutely crazy
Buy a dataset of chats with Fable, and you get passwords to hack half of the world
Lesmes retweeted
After six years in stealth, Atomic Machines is out. Our mission: on-demand, universal command of matter. First beachhead: the Matter Compiler, an AI-native manufacturing system that builds working micro-machines from code alone. No per-product tooling. No process development. Different code, different machine.
Lesmes retweeted
How to get top-1 on the Decision Index Vision?
Step 1: Look at the data
Step 2: Realize we are underperforming tasks that need precise in-image reading
Step 3: Connect the dots and realize our HF implementation caps at 512*512 resolution
Step 4: Profit
PSA: if you're using our decider for vision tasks with the HF model, don't hesitate to raise the max resolution (in model.py)
Right now it's running 512*512 max, which cost a lot on tasks that requires reading/precision in the image. The model is most likely already around the SOTA in the vision task as well, thanks to the Qwen backbone
The API is fine though, we already support 2M pixels images :)
Lesmes retweeted
Jensen creando capacidades sobre las que no tiene realmente usuarios, de nuevo probando cosas en un espacio competitivo que no existe.
Reading delays quarterly reports I found a new component of the NVQLink family, Ising.
open AI, do not be confused with OpenAI.
joapen.com/blog/2026/10/06/n…
Lesmes retweeted
it’s coming
developers.google.com/worksp…
Lesmes retweeted
And if you are wondering what Pi Durable is all about, I wrote a little piece on it here:
earendil.com/posts/pi-durabl…
With code examples. How quaint.
People of Pi: We've shipped Pi 1.0 with Pi Durable. Go make them yours.
earendil.com/posts/pi-1-0/
Lesmes retweeted
Twelve years after launching Universal SSL, Cloudflare is applying to become a certificate authority. By combining an established root, an ACME-first approach, and Merkle Tree Certificates, we are building a post-quantum CA for the open web. cfl.re/4AFIczO #BirthdayWeek
Lesmes retweeted
I tested Sonnet 5.5 and couldn't believe my eyes. 2 real repos, 105 bugs, find and fix what you can.
It took it seriously. The results:
- Sonnet 5.5 (max): 55.5
- GPT-6 Astra (max): 45
- GPT-5.6 Sol (max): 43.5
- Fable 5.1 (max): 43
- Opus 5.5 (max): 41.7
- Sonnet 5.5 (xhigh): 39
The secret? Sonnet 5.5 (max) might be cheap and fast for most tasks, but it's the least lazy model I tested. It leads in bug hunting by spending turns.
It took about 6× Astra's turns and about 3× Opus 5.5's for 10–14 more bugs. xhigh cut its turns by more than half and dropped to 39, below Opus 5.5 max, which used fewer turns on average:
- Sonnet 5.5 (max): 1,330 turns (n=2)
- GPT-6 Astra (max): 222 turns (n=3)
- GPT-5.6 Sol (max): 1,083 turns (n=2)
- Fable 5.1 (max): 459 turns (n=1)
- Opus 5.5 (max): 476 turns (n=3)
- Sonnet 5.5 (xhigh): 588 turns (n=1)
More effort levels for Sonnet 5.5 with n=3 drop in this thread today 🧵
Spain was also deindustrialized during 80s and 90s by the same people.
Germany is not merely being deindustrialized.
It is being dismantled.
Piece by piece, factory by factory, job by job.
And the people driving this destruction are not hiding in the shadows.
They sit in government offices, corporate boardrooms, and political institutions.
German journalist Patrik Baab puts his finger on what Berlin would rather keep buried: Germany’s corporate giants do not appear to fear the collapse of German industry, because many of their decision-makers no longer have their interests anchored in Germany.
Cheap energy is gone.
Factories are closing or moving abroad.
Jobs are disappearing.
The industrial backbone is being hollowed out.
And yet the DAX executives remain remarkably calm.
Why?
Because Germany, for them, is increasingly just another balance-sheet entry.
American investment funds hold enormous influence.
Production is being shifted to China, the United States, Uzbekistan and elsewhere.
Germany is being treated less like the industrial heart of Europe and more like a cost center that can be cut, downsized, or written off.
Meanwhile, billions are being poured into war.
Not into rebuilding factories.
Not into modernizing infrastructure.
Not into securing Germany’s industrial future.
Into weapons.
And while taxpayers carry the burden, arms manufacturers and investors stand to benefit.
For BlackRock and the global financial establishment, critics argue, Germany’s decline does not necessarily represent a disaster.
It can represent an opportunity.
That is the brutal logic of the system being described:
One side loses factories, energy security and jobs.
Another side collects contracts, profits and influence.
War and deindustrialization begin to look less like separate crises and more like two sides of the same destructive process.
And still, Berlin talks about “transformation”
“Solidarity”
“Adaptation”
“Modernization”
But behind the carefully chosen words, Germany’s industrial foundation is being stripped away.
The factories disappear.
The skilled jobs disappear.
The energy advantage disappears.
The economic independence disappears.
And the political establishment keeps insisting that everything is under control.
The real question is no longer whether Germany can afford this course.
It is whether Germany can survive it as the industrial power it once was.
Because a country does not remain an industrial nation after its factories, energy security, skilled workforce and productive capacity have been dismantled.
At some point, there is nothing left to "transform"
There is only something left to rebuild.