Philippe Modard

Account not used anymore. Find me on

[Philippe Modard @Philmod]

How Kaggle load-balances a gRPC application across multiple GKE clusters. https://t.co/LstgIPYOSK

[Philippe Modard @Philmod]

LLMs can't jump https://t.co/3PhrJ8ZNnX

[Philippe Modard @Philmod]

I'm already tired of this frontend style we find everywhere now https://t.co/dO2Dz00J19

[Philippe Modard @Philmod]

"The problem is that LLMs inherently lack the virtue of laziness. Work costs nothing to an LLM." https://t.co/PFALqMeuoL

[Philippe Modard @Philmod]

@Eurostar I cannot submit a claim on your website, looks like the reCAPTCHA site key is wrong. Smart way to avoid claims indeed :)

[Philippe Modard @Philmod]

Great explanation of the new Google Gemma 4 open source models. https://t.co/FggeDtK870

[Philippe Modard @Philmod]

@yuvadm That seems to work locally to bind a directory as a npm package. In my case, the CI/CD pipeline needs to install the latest package from our private repository. We obviously trust the latest version and don't want to wait 7 days. yarn has `npmMinimumReleaseAgeExclude`,but not npm

[Philippe Modard @Philmod]

@tjlu_artwork @yuvadm Thanks, I was able to do that for `uv`. But I haven't found any solution with `npm`.

[Philippe Modard @Philmod]

@yuvadm My main repo has private dependencies (of my own). Using these tricks means I cannot install the latest versions of my own private packages.

[Philippe Modard @Philmod]

@giffmana I wonder if these models can also be creative at new types of attacks? I'm guessing not if they haven't seen them in their training data?

[Philippe Modard @Philmod]

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. https://t.co/IUw8emWHHm

[Philippe Modard @Philmod]

litellm package was compromised. https://t.co/I7DnyScrlO
Good time for a reminder to pin your packages, and check before upgrades.

[Philippe Modard @Philmod]

@doodlestein @Beetcoin I haven't been able to find the reply you're referring to. That said, I’m just looking at the data—there’s no need for personal insults when we’re just discussing technical performance.

[Philippe Modard @Philmod]

"LLMs optimize for plausibility over correctness. In this case, plausible is about 20,000 times slower than correct." https://t.co/ESK2RPzr9Z

[Hōrōshi バガボンド @KatanaLarp]

https://t.co/h9C7YZeCx3

[Philippe Modard @Philmod]

Is AI singularity here already? https://t.co/V33xQzjWlc

[Andrej Karpathy @karpathy]

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:

- the human iterates on the prompt (.md)
- the AI agent iterates on the training code (.py)

The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.

https://t.co/YCvOwwjOzF
Part code, part sci-fi, and a pinch of psychosis :)

[Philippe Modard @Philmod]

@ryzerth I got the exact same email.

[Philippe Modard @Philmod]

AI vs human code gen report: AI code creates 1.7x more issues https://t.co/fv0c3UU9MX

[Philippe Modard @Philmod]

The venue (@lotto_arena) decided to change my @ElectricCallboy concert ticket from the pit to the seats, unilaterally. Really not cool! What's the point of booking my ticket year in advance if I can't enjoy the show the way I'd prefer?

[Philippe Modard @Philmod]

Surfer-H Meets Holo1: Cost-Efficient Web Agent Powered by Open Weights https://t.co/qihM1ZurVK

[Philippe Modard @Philmod]

Nice vizualization of a LLM's transformer https://t.co/WHBFzVjPSe