“the AI guy” at MGX. I try a new software every day. ex. EQT Motherbrain, Spotify, KPMG, Accenture
Stockholm, Sweden
Joined April 2009
- Tweets715
- Following2.8K
- Followers293
- Likes1.4K
Pinned Tweet
With the emergence of AI coders doing most of the work, I predict a future where it writes correct code with much less guardrails. In fact languages designed for addressing human flaws will be displaced by AI first programming languages, closer to the metal.
Pietro Casella retweeted
narrative violation: preferences will swing back from DIY vibe-coded software to super secure industry-leading SaaS solutions.
while it's super easy to build anything, endlessly protecting and securing it is super hard...and teams want to sleep at night.
Very important development. Formal verification is the holy grail for high risk systems like flight control, medical devices, power plant controllers and more.
Prediction: after all easy software is refactored, cracked engineers will mainly find joy building very complex systems that were not possible before.
… so best time ever to be or become one.
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
Predicting possible consequences of good interpretability research:
- LLM API will have concept suppression and boosting controls instead of fine tuning.
- LLM will ship and be priced by subdomains/ cognitive subsets.
- inference acceleration and iq focus by lobotomy instead of quantization/distillation.
有点震撼。。。我们可能真的快要搞清楚大模型内部到底是怎么运作的了。
OpenAI早期机械可解释性研究的参与者，昨天突然提出了一个很大胆的结论：
我们可能一直误解了神经网络。
昨天， @nickcammarata 发了一篇很有意思的文章，叫《Neural Networks Are Already Clean》。
这哥们早年就在OpenAI做机械可解释性研究，2023 年还牵头启动了@OpenAI 第一个大规模自动化神经元解释项目。
当时他们发现，现有工具只能解释大约 2% 的神经元。
Nick因此得出了一个很悲观的判断：神经网络内部存在大量纠缠，想搞清楚它们到底在干什么，可能非常困难。
结果三年后，他带着更先进的工具和Agent重新研究，发现情况完全不同。
在他们测试的现代视觉模型中，最后一层 99%～100% 的神经元都能找到相当清晰的解释。
甚至在Qwen3-32B这样的语言模型中，初步研究也发现，约90%的神经元都有一个可以识别的主要功能。
最有意思的是，他还重新研究了过去的一些经典案例。
比如曾经有一个神经元，既会对猫脸产生反应，也会对汽车前脸产生反应。
以前大家觉得，这是典型的Superposition，多个不相关的概念纠缠在了同一个神经元里。
结果重新研究后发现，这个神经元可能只是在识别一种共同的V形视觉结构。
猫脸和汽车前脸恰好都有这种特征。
全世界的家人们折腾了这么多年，可能连最基础的观察方式都有问题。
而且我觉得这篇文章最值得关注的，其实是他们使用 Agent的方式。
他们让Agent自动观察神经元、提出功能假设、设计实验、分析电路，再通过新的输入验证假设，甚至自动审查了10 万张神经元归因图。
这其实也是我最近一直关注的一个方向：让 AI 自己去做科学研究，把大量观察、假设、验证的工作自动化。
过去需要研究员花大量时间逐个分析的神经元，现在已经可以让Agent成规模地开展研究了。
当然，目前这些结果还处于早期阶段，距离完整解释大模型的内部机制，还有很多工作要做。
但我觉得Nick提出的这个观点非常有意思：
神经网络内部可能一直都很有规律，只是过去我们的工具太原始，根本看不清楚。
如果这个方向最终被验证，机械可解释性领域过去很多基础假设，恐怕都需要重新审视了。
I think of models as a mildly in the spectrum, engineer polymaths. Half my prompt input is nuance. Maybe that’s the human part of the equation. But i have great hopes for a knob to modulate that.
Prediction: next generation of models will be a renaissance or Affective Computing and Symbolic AI (inference)
Such a cool project!
Forget trying to design everything in Figma. You can build your own design tools with AI.
Introducing Toolcraft - a starter kit and UI library for building beautiful creative apps. It's free and available for you today.
npx @pixel-point/toolcraft create
Use it to build apps to stylize images, create WebGL shaders, Three.js scenes, animations, photo editing tools, or whatever else needs a canvas and controls.
Toolcraft gives you the full architecture in a single solution. It comes with:
- plenty of UI components: sliders, pickers, timelines, curves, and more
- built-in canvas, export, and toolbar setup
- AI instructions to help the agent deliver a great result to you
With Toolcraft, you don’t need an extensive prompt. Just explain what type of visuals you want to build, attach references, and wait.
It’s been a game changer for our design process.
Watch the full video on our YouTube channel with all the details and examples.
I’ve been saying this a lot. Think about the Agent(‘s) Experience above all - how the AI perceives the system and the user. AND the AI experience (how the user perceives the AI). Most AI agents are blind to most stuff, let’s change that in the future or software.
This is a very cool idea. Great to see something new coming out. Paradigms shifting.
Introducing Fractal — the recursive language model CLI agent.
Fractal is a terminal agent that can answer questions, plan tasks, read and write files, run commands, inspect codebases, and work through problems turn by turn.
But at its core, Fractal is not just another CLI harness around an LLM.
It is powered by predict-rlm, our self-harnessed Recursive Language Model runtime. In practice, that means Fractal can recursively reason through context, interact with its own REPL environment, and take action through code as it works.
We built it to make RLMs easy to try and actually useful on real tasks.
If you've been hearing about RLMs but could never come around to using one, this is the best way to do so.
And if you already know what RLMs are capable of, Fractal is the easiest way to use one on your own tasks.
Install it today:
→ curl -LsSf fractal.trampoline.ai/instal… | sh
Drop us a ⭐️:
→ github.com/Trampoline-AI/fra…
Or check out the website:
→ fractal.trampoline.ai/
Pietro Casella retweeted
Introducing Fractal — the recursive language model CLI agent.
Fractal is a terminal agent that can answer questions, plan tasks, read and write files, run commands, inspect codebases, and work through problems turn by turn.
But at its core, Fractal is not just another CLI harness around an LLM.
It is powered by predict-rlm, our self-harnessed Recursive Language Model runtime. In practice, that means Fractal can recursively reason through context, interact with its own REPL environment, and take action through code as it works.
We built it to make RLMs easy to try and actually useful on real tasks.
If you've been hearing about RLMs but could never come around to using one, this is the best way to do so.
And if you already know what RLMs are capable of, Fractal is the easiest way to use one on your own tasks.
Install it today:
→ curl -LsSf fractal.trampoline.ai/instal… | sh
Drop us a ⭐️:
→ github.com/Trampoline-AI/fra…
Or check out the website:
→ fractal.trampoline.ai/
Prediction: UX will disappear for a while (for chat UX)… but will come back big time for post AI specialized experiences.
Similar to the client to server circular trend.
Pietro Casella retweeted
Almost every SaaS app inside Vercel has now been replaced with a generated app or agent interface, deployed on Vercel.
Support, sales, marketing, PM, HR, dataviz, even design and video workflows. It’s shocking.
The SaaSpocalypse is both understated and overstated. Over because the key systems of record and storage are still there (Salesforce, Snowflake, etc.)
Understated because the software we are generating is more beautiful, personalized, and crucially, fits our business problems better.
We struggled for years to represent the health of a Vercel customer properly inside Salesforce. Too much data (trillions of consumption data points), the ontology of Vercel was a mismatch to the built-in assumptions, and the resulting UI was bizarre. We generated what we needed instead. When you don’t need a UI, you just ask an agent with natural language.
We’ve also been moving off legacy systems with poor, slow, outdated, and inconsistent APIs, as well as just dropping abstraction down to more traditional databases. UI is a function 𝑓 of data (always has been), and that 𝑓 is increasingly becoming the LLM.
What happens when the moat disappears in software — when everyone can copy and everyone can ship fast?
Companies won’t converge. They’ll contrast.
Paradoxically, enterprise software may have a renaissance. There are now so many building blocks that the combinatorial space is exploding. Categories won’t collapse into “everything is a chatbot.”
Instead we’ll see sharper opinions, more differentiated products, and fewer copycats.
Speed and Complexity will be table stakes. Conviction and Taste will be the “moat”. People will choose by betting on future innovation advantage.
A new paradigm is emerging: Skill-Oriented Programming.
Skills become the primitive for deterministic, quasi-deterministic, and non-deterministic behavior in business systems.
Example: quasi-determinism → a skill with a static function but llm as the driver. Instead of config tables, you simply describe the behavior. Allow the AI to change the file, expands the expressivity of the system, retaining some degree of control. Block that and you establish rules of physics for the environment.
This enables simpler, more powerful architectures where you design the persona first, then build the system around it to achieve the desired outcome.
Working on this, super exciting to see it taking shape.
Fixing bugs on a live system by asking its assistant to retry a task using a tool… while I’m actively AI coding that same tool.
Feels like performing brain surgery on a patient who’s awake and talking.
Thinking like an agent is the most important mindset for the modern architect, designer, engineer, org designer. I call it the “AI Experience” - how the AI sees and interacts with you and your business - and has been a major unlock for how we build apps and rethink processes.
As more Next.js apps are built with AI, we've been focusing on improving the agent experience.
Here's what we've learned, and how we're thinking about agents as first-class users ↓
nextjs.org/blog/agentic-futu…
Jobs that do the work become jobs designing how the work is done.
No one does the work. The work gets done.
Pietro Casella retweeted
MGX co-leads @AnthropicAI's $30 billion Series G funding round at a $380 billion post money valuation. The investment will accelerate frontier research, product development, and infrastructure expansion to meet growing enterprise demand for advanced AI.
#ArtificialIntelligence #MGX
Idea: enterprise as code. One repo with all (agents) you need to run the company: product, ops, sales, everything... later, enterprise in a box….
