#ai

Well, "one engineer one month one million lines of code" -> "who the hell knows what's in there" -> "you need some AI coder architect".

https://www.linkedin.com/posts/galenh_principal-software-engineer-coreai-microsoft-activity-7407863239289729024-WTzf

#ai

It's just conflicting thoughts patched together.



AI Is Changing How We Learn at Work
https://hbr.org/2025/12/ai-is-changing-how-we-learn-at-work

Harvard Business Review

AI Is Changing How We Learn at Work

As artificial intelligence rapidly transforms the workplace, it is also reshaping how people learn, develop expertise, and form their professional identities. Although gen AI promises to accelerate learning and boost productivity, it risks undermining the…

#ai

What GitHub imagined about everyone creating their own app is almost here. Github Spark is not that good, vercel ship 0 is also buggy. Cloude is doing a fairly good job here.

The other day I needed a special checklist and instead of writing the code myself, I asked Gemini to do it and it was surprisingly good.


---
Build interactive diagram tools | Claude | Claude
https://www.claude.com/resources/use-cases/build-interactive-diagram-tools

Claude

Build interactive diagram tools | Claude | Claude

Turn a detailed prompt into a working educational app with the data depth interactions and visual polish you specify.

#ai

Interesting... Going from finite Eucleadian space to Hilbert space

https://openreview.net/forum?id=1b7whO4SfY

openreview.net

Gated Attention for Large Language Models: Non-linearity, Sparsity,...

Gating mechanisms have been widely utilized, from early models like LSTMs and Highway Networks to recent state space models, linear attention, and also softmax attention.
Yet, existing literature...

#ai

Yeah, just import antigravity.

Google Antigravity
https://antigravity.google/

Google Antigravity

Google Antigravity - Build the new way

#ai

This is more than language models. Somehow many forecasting models also almost fall inside the realm. Somehow the root of this is the latent space. Time series models even classical ones, have enlarged latent space, which is more or less embedding patterns with higher dimensions.

However this paper is a bit fishy. I just can't trust the proof of theorem 2.2.

Language Models are Injective and Hence Invertible
https://arxiv.org/abs/2510.15511

arXiv.org

Language Models are Injective and Hence Invertible

Transformer components such as non-linear activations and normalization are inherently non-injective, suggesting that different inputs could map to the same output and prevent exact recovery of...

#ai


I ran into this quite interesting paper when exploring embeddings of time series.

In the past, the manifold hypothesis has always been working quite well for me regarding physical world data. You just take a model, compress it, do something in the latent space, decode it, damn it works so well. The latent space is so magical.

To me, the hyperparameters for the latent space has always been some kind of battle between the curse of dimensionality and the Whitney embedding theorem.

Then there comes the language models. The ancient word2vec was already amazing. It brings in the questioin of why embeddings works unbelievably well in language models and it bugs me a lot. If you think about it, regardless of the model, embedding has been working so well. This hints that language embeddings might be universal. There is the linear representation hypothesis, but it is weird as it is missing the global structure. This paper provides a bit more clarity. The authors used a lot of assumptions but the proposal is interesting in the sense that the cosine similarity we used is likely a tool that depends on the distance on the manifold of the continuous features in the backstage.



https://arxiv.org/abs/2505.18235v1

arXiv.org

The Origins of Representation Manifolds in Large Language Models

There is a large ongoing scientific effort in mechanistic interpretability to map embeddings and internal representations of AI systems into human-understandable concepts. A key element of this...

#ai

AI to juniors is more or less a "fuck you in particular" thingy.

https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/

#ai

https://x.com/Google/status/1981403118803046686

X (formerly Twitter)

Google (@Google) on X

Google Earth AI brings together our geospatial models for floods, wildfires, cyclones, air quality and much more to help with everything from environmental monitoring to disaster response. Today, we’re advancing its capabilities and expanding access around…

#ai

Souly, Alexandra, Javier Rando, Ed Chapman, Xander Davies, Burak Hasircioglu, Ezzeldin Shereen, Carlos Mougan, et al. 2025. “Poisoning Attacks on LLMs Require a Near-Constant Number of Poison Samples.” arXiv [Cs.LG]. arXiv. http://arxiv.org/abs/2510.07192.

arXiv.org

Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples

Poisoning attacks can compromise the safety of large language models (LLMs) by injecting malicious documents into their training data. Existing work has studied pretraining poisoning assuming...

#ai

OpenAI open weight models. The benchmarks are amazing.

https://huggingface.co/collections/openai/gpt-oss-68911959590a1634ba11c7a4

#ai

PedanticAI already partially supports A2A .

Agent2Agent (A2A) Protocol
https://a2aproject.github.io/A2A/dev/

a2aproject.github.io

Agent2Agent (A2A) Protocol

An open protocol enabling communication and interoperability between opaque agentic applications.

#ai

Well, I figure many people actually "think" like this.

The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
https://ml-site.cdn-apple.com/papers/the-illusion-of-thinking.pdf

#ai

Extending Minds with Generative AI | Nature Communications
https://www.nature.com/articles/s41467-025-59906-9

Nature

Extending Minds with Generative AI

Nature Communications - As human-AI collaborations become the norm, we should remind ourselves that it is our basic nature to build hybrid thinking systems – ones that fluidly incorporate...

https://ai-2027.com “We predict that the impact of superhuman AI over the next decade will be enormous, exceeding that of the Industrial Revolution.” 不管怎样，这个页面的 interaction 很棒 #ai

Ai-2027

AI 2027

A research-backed AI scenario forecast.

#ai

2 trillion parameters...

The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation
https://ai.meta.com/blog/llama-4-multimodal-intelligence/

Meta AI

The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation

We’re introducing Llama 4 Scout and Llama 4 Maverick, the first open-weight natively multimodal models with unprecedented context support and our first built using a mixture-of-experts (MoE) architecture.

#ai

I'm also a firm believer of it. Keep the foundation simple and build on top of the foundation.

https://youtu.be/vRTcE19M-KE?si=cdpZ9_nRcaYu4zOa

YouTube

Stanford Webinar - Large Language Models Get the Hype, but Compound Systems Are the Future of AI

For more information about Stanford's Artificial Intelligence programs visit: https://stanford.io/ai

In recent years AI has taken center stage with the rise of Large Language Models (LLMs) that can be used to perform a wide range of tasks, from question…

#ai

Funny article

On Chomsky and the Two Cultures of Statistical Learning
https://norvig.com/chomsky.html