I write code. I do a few other things too...
LX
Joined August 2009
- Tweets 8,885
- Following 1,065
- Followers 400
- Likes 2,400
Paulo Gaspar retweeted
Rust is special. The bincode library moved away from Github because of 'immorality', and now bans and scolds contributors who work in 'evil' industries like oil (you know, the guys who keep the lights on).
Folks, it's not that hard--just keep politics out of tech.
Makes one think... piped.video/shorts/CHH4kaOmP… via @YouTube
Paulo Gaspar retweeted
AI Tools don't replace expertise - they amplify it.
Think of AI as a power-tool in the hands of a craftsman: it doesn't make you the craftsman, but it magnifies what you can do when you already know your trade. This elevates your use of Cursor, Claude Code, Gemini CLI and others.
First, it's worth recognising what expertise brings to the table: domain knowledge, pattern-recognition, judgement, trade-offs, system-thinking.
When you're an engineer who has internalised core concepts - performance, scalability, reliability - you're not just running commands or following recipes. You're asking good questions, choosing the right abstractions, understanding context. AI tools alone don't understand context the way you do; they can generate options and surface patterns, but you still decide which path makes sense.
Second, when your expertise is strong, you can leverage AI tools much more effectively. If you know how to frame a problem, break it into sub-problems, assess options, apply constraints, test and iterate, then the AI becomes a multiplier.
For example, AI is used for a spectrum of code-gen, to reduce repetitive work, and explore large design spaces. But in order to exploit that speed you still need the skill to interpret the results, catch edge-cases, know when to trust the output and when to probe deeper.
Third, the trajectory isn't "tools will replace engineers" entirely but rather "tools will raise the ceiling of what engineers can do". AI frees us to focus on higher-level tasks rather than repetitive ones. So the message is: ramp your core engineering capabilities - architecture thinking, domain fluency, product impact - and then use AI to accelerate your reach and explore more ambitious outcomes.
Fourth, there's another dimension: the richer your skillset, the better feedback you can give the AI, and the better the AI becomes as a partner. If you are good at prompt-design (and more recently context engineering), good at crafting the right constraints, good at validating and refining outputs, then the AI contributes more. If instead you treat it like a black-box oracle, you risk mis-use or over-dependence. In engineering contexts, guardrails, interpretation and a critical eye remain vital.
In short: expertise is the foundation. AI tools are the amplifier.
The stronger the foundation, the louder the amplifier becomes. When you bring the skill, the judgment, the systems-level perspective, you unlock far more than you would by simply running the latest tool or model in isolation.
@simonw covers more on this in a recent write-up: simonwillison.net/2025/Oct/7…
Best two articles I have found so far about AI assisted real world serious software engineering (2 of 2):
steipete.me/posts/just-talk-…
Best two articles I have found so far about AI assisted real world serious software engineering (1 of 2):
simonwillison.net/2025/Oct/7…
Paulo Gaspar retweeted
This is excellent - long, information dense, full of actionable advice on getting the most out of GPT-5 and Codex CLI
📢 Time for an update on my workflow. This one's a 23 min read, so buckle up. 100% organic and hand-written, like an animal. steipete.me/posts/just-talk-…
Great product story, great keyboards:
piped.video/tu6_cMJwqxw?si=OZWd…
Paulo Gaspar retweeted
Hard to say what's more striking, the amateurishness or the brutality.
Paulo Gaspar retweeted
Unbelievable. Hierarchical reasoning seems like such a promising new direction.
My brain broke when I read this paper.
A tiny 7 Million parameter model just beat DeepSeek-R1, Gemini 2.5 pro, and o3-mini at reasoning on both ARG-AGI 1 and ARC-AGI 2.
It's called Tiny Recursive Model (TRM) from Samsung.
How can a model 10,000x smaller be smarter?
Here's how it works:
1. Draft an Initial Answer: Unlike an LLM that writes word-by-word, TRM first generates a quick, complete "draft" of the solution. Think of this as its first rough guess.
2. Create a "Scratchpad": It then creates a separate space for its internal thoughts, a latent reasoning "scratchpad." This is where the real magic happens.
3. Intensely Self-Critique: The model enters an intense inner loop. It compares its draft answer to the original problem and refines its reasoning on the scratchpad over and over (6 times in a row), asking itself, "Does my logic hold up? Where are the errors?"
4. Revise the Answer: After this focused "thinking," it uses the improved logic from its scratchpad to create a brand new, much better draft of the final answer.
5. Repeat until Confident: The entire process, draft, think, revise, is repeated up to 16 times. Each cycle pushes the model closer to a correct, logically sound solution.
Why this matters:
Business Leaders: This is what algorithmic advantage looks like. While competitors are paying massive inference costs for brute-force scale, a smarter, more efficient model can deliver superior performance for a tiny fraction of the cost.
Researchers: This is a major validation for neuro-symbolic ideas. The model's ability to recursively "think" before "acting" demonstrates that architecture, not just scale, can be a primary driver of reasoning ability.
Practitioners: SOTA reasoning is no longer gated behind billion-dollar GPU clusters. This paper provides a highly efficient, parameter-light blueprint for building specialized reasoners that can run anywhere.
This isn't just scaling down; it's a completely different, more deliberate way of solving problems.
Saved by the Germans. Shame on us Danes.
We did it: 🇩🇪Germany will OPPOSE Chat Control! 🥳
Thanks everyone for writing to the ministers. 🫶
#ChatControl will not get a majority in the EU Council - at least for now.
Not Using a Mouse For 30 Days piped.video/YrEOnSv0C4I?si=SsQH… via @YouTube
Paulo Gaspar retweeted
JetBrains is adopting ACP.
Every @jetbrains IDE will support any ACP-compatible agent. Combined with Zed, Neovim, and Emacs, that's one protocol implementation reaching developers everywhere.
Agent developers: the ecosystem just got real.
Paulo Gaspar retweeted
One protocol to connect all IDEs with different coding agents.
- No vendor lock-in
- Confidence and control
- A great US in your IDE
Read more at blog.jetbrains.com/ai/2025/1…
JetBrains is adopting ACP.
Every @jetbrains IDE will support any ACP-compatible agent. Combined with Zed, Neovim, and Emacs, that's one protocol implementation reaching developers everywhere.
Agent developers: the ecosystem just got real.
Eagerly waiting for text diffusion: A new paradigm for LLMs piped.video/bmr718eZYGU?si=CqcN… via @YouTube #ai #aicoding
Paulo Gaspar retweeted
GLM 4.6 is surprisingly good. I think it is my third favorite coding model after gpt-5 & sonnet 4.5. I have a feeling that by the end of year the best models for coding are going to be:
- GPT
- Claude
- Gemini
- GLM
- Grok
