The peril of laziness lost In his classic Programming Perl — affectionately known to a generation of technologists as "the Camel Book" — Larry Wall famously wrote of the three virtues of a programmer as laziness, impatience,…

"The problem is that LLMs inherently lack the virtue of laziness. Work costs nothing to an LLM." bcantrill.dtrace.org/2026/04/12/t...

A Visual Guide to Gemma 4 A great start to a new job ;)

Great explanation of the new Google Gemma 4 open source models. newsletter.maartengrootendorst.com/p/a-visual-g...

TurboQuant: Redefining AI efficiency with extreme compression We strive to create an environment conducive to many different types of research across many different time scales and levels of risk.

New compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss. research.google/blog/turboqu...

Supply Chain Attack in litellm 1.82.8 on PyPI litellm version 1.82.8 on PyPI contains a malicious .pth file that harvests SSH keys, cloud credentials, and secrets on every Python startup, then attempts lateral movement across Kubernetes clusters.

litellm package was compromised. futuresearch.ai/blog/litellm...
Good time for a reminder to pin your packages, and check before upgrades.

What Happens When AI Agents Build Societies? Inside TerraLingua Researchers built TerraLingua, a simulated world where AI agents survive and communicate. The result: emergent societies, governance, and shared knowledge.

www.cognizant.com/us/en/ai-lab...

"LLMs optimize for plausibility over correctness. In this case, plausible is about 20,000 times slower than correct." x.com/KatanaLarp/s...

Is AI singularity here already? x.com/karpathy/sta...

That's a new low.

Lego announces Smart Brick, the ‘most significant evolution’ in 50 years Starting with Lego Star Wars.

Lego announces Smart Brick, the ‘most significant evolution’ in 50 years share.google/e62z1cBDD6sr...

I asked their chatbot who was saying to break the cover of the meter first. I wasn't comfortable with that advice, so I asked the chatbot to get me in touch with their team.
Today, I got reimbursed of the non compatible sensor.
Glad I didn't break my water meter...

Companies will lose money and trust using LLM chatbots.

I recently bought a sensor for my water meter. It wasn't clear if the sensor was compatible. I wanted to email the company but there was only a chatbot that said it was compatible. After receiving it, I couldn't figure out how to install it

AI vs human code gen report: AI code creates 1.7x more issues We analyzed 470 open-source GitHub pull requests, using CodeRabbit’s structured issue taxonomy and found that AI generated code creates 1.7x more issues.

AI vs human code gen report: AI code creates 1.7x more issues www.coderabbit.ai/blog/state-o...

2025 LLM Year in Review 2025 Year in Review of LLM paradigm changes

"And while agent swarms running in the cloud feels like the "AGI endgame", we live in an intermediate and slow enough takeoff world of jagged capabilities that it makes more sense to run the agents directly on the developer's computer." karpathy.bearblog.dev/year-in-revi...

After moving our dev workload to the cloud (at Google we dropped our beloved workstations circa 2020, for more powerful and accessible cloud instances), are we going to move it back to a powerful machine with gpu, in order to run llm inference ourselves (maybe cheaper and def more private)?

Every airport should have a post box in the departure area. Really!

TIL: we can use Markdown in Google Docs! support.google.com/docs/answer/...

Disposable Code Is Here to Stay, but Durable Code Is What Runs the World Every day I seem to run into yet another post with someone solemnly opining that “writing code has never been the hardest part of software engineering. And hey, that’s smashing.

Disposable Code Is Here to Stay, but Durable Code Is What Runs the World www.honeycomb.io/blog/disposa...

The difficulty is how to move from disposable to trusted code imho? How to take a huge bunch of code and make it production ready?

Graphique issu d'un article de Libération L’organisation Anti-Defamation League a recensé, depuis 2005, les meurtres ou séries de meurtres commis aux Etats-Unis par des membres de groupes extrémistes. Sur 20 ans, une très large majorité des meurtriers (347 sur 371) sont liés à des organisations d’extrême droite. Pour la majorité d’entre eux (262), il s’agit de suprémacistes blancs. Les autres sont principalement liés à la mouvance complotiste et anti-gouvernementale, comme les «citoyens souverains».

Depuis 2005 aux USA, 93% des assassinats commis par des extrémistes sont le fait de l'extrême droite. Partout, l'antiwokisme s'impose pour invisibiliser ces crimes dont il est complice par ses silences et ses outrances.

Why do LLM coding models add so many emojis in the code/logs? Is it widely done in public repos?

Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity We conduct a randomized controlled trial (RCT) to understand how early-2025 AI tools affect the productivity of experienced open-source developers working on their own repositories. Surprisingly, we…

Less experienced engineers overly rely on LLMs without understanding the output, adding more work on the reviewers. metr.org/blog/2025-07...

Je cherchais des documentaires animaliers pour enfants sur Netflix sur Google. Pas certain que les options 1 et 3 soient appropriées ...

The quality of software engineering is worrisome these days. Less experienced engineers rely so much on LLMs, don't always check the generated code nor understand it. Thus reviews take much longer, as I need to ask more questions, and even sometimes help them understand their own code.

The Ingredients of a Productive Monorepo blog.swgillespie.me/posts/monore...

Understanding and Coding the KV Cache in LLMs from Scratch KV caches are one of the most critical techniques for efficient inference in LLMs in production.

Understanding and Coding the KV Cache in LLMs from Scratch magazine.sebastianraschka.com/p/coding-the...

Surfer-H Meets Holo1: Cost-Efficient Web Agent Powered by Open Weights We present Surfer-H, a cost-efficient web agent that integrates Vision-Language Models (VLM) to perform user-defined tasks on the web. We pair it with Holo1, a new open-weight collection of VLMs…

Surfer-H Meets Holo1: Cost-Efficient Web Agent Powered by Open Weights arxiv.org/abs/2506.02865

Bien que quand on voit les US, on se dit que les services publics ne sont pas toujours plus stables.

Some recruiter called my phone, they said they got my phone number through LinkedIn? Is that possible??

AI 2027 A research-backed AI scenario forecast.

Lots of people trying to guess the future of AI. Interesting story. ai-2027.com

johnoliverwantsyourraterotica.com

AI Coding in 2024 be like YouTube video by Programmers are also human

PLEASE! youtube.com/shorts/64TNG...