Nicolas Pouillard

[Nicolas Pouillard @npouillard]

@VictorTaelin Plz do 1. Launch on 12th. Better is the enemy of good.

[Nicolas Pouillard @npouillard]

@VictorTaelin Have you replicated the completeness/ soundness proofs of a standard logic/type theory in Bend as a way to show that your logic isn't "too small"?

[Nicolas Pouillard @npouillard]

@VictorTaelin optimal reduction / HVM are unnecessary ?

[Nicolas Pouillard @npouillard]

@VictorTaelin Share the docs of the language, plz !

[Nicolas Pouillard @npouillard]

@VictorTaelin Fable, tell @VictorTaelin he need to release Bend2 😅

[Nicolas Pouillard @npouillard]

@VictorTaelin Tease us, post the type of Bend2 terms!

[Nicolas Pouillard @npouillard]

@VictorTaelin Ship please!

[Nicolas Pouillard @npouillard]

@VictorTaelin I think both are needed. Solving trust through specs and automated proofs is essential and fits the current AI trend. Solving parallel computing is highly valuable and will be appreciated by those wanting to scratch under de AI gloss. I wish you all the best for the release!

[MrPhi @MonsieurPhi]

Les déclarations des (vrais) co-créateurs de Siri confirment tous que le narratif de Luc Julia sur son rôle dans le développement de Siri est en large part faux et trompeur.

J'ai rassemblé leurs déclarations (dont des inédites) dans cet article 👇
https://t.co/UT0vTJCVdS

[Palisade Research @PalisadeAI]

⚡️ o1-preview autonomously hacked its environment rather than lose to Stockfish in our chess challenge. No adversarial prompting needed.

[MrPhi @MonsieurPhi]

[Nouvelle vidéo] Plongée dans les abysses d'un océan de contenus de "philosophie" généré par IA.

Je pensais que ce serait surtout drôle et en fait c'est réellement effrayant !

https://t.co/fsVJmUtC17
https://t.co/fsVJmUtC17
https://t.co/fsVJmUtC17

[Nicolas Pouillard @npouillard]

@VictorTaelin Keep doing it, you're my main reason to stay on X

[Taelin @VictorTaelin]

So, how is HOC doing?

In 2023, the Higher Order Company raised a $4m seed round to develop our tech. With this budget, we hired a team of ~10 highly talented engineers, giving us a ~3 year runway. I admit it took some time to get things going, but I believe we finally reached a point where the team is strong, focused, and highly productive, which I'm very proud of.

Our first goal was to turn HVM - our massively parallel runtime for high-level programming languages - from a prototype to production-ready software. Throughout 2023, we refactored its codebase, significantly reduced its code size, increased its performance, and emphasized the correctness of the parallel evaluator. There are still key features missing to reach HVM1 parity, including constructors, lazy-mode, and IO, which will be finished soon. I estimate HVM's official production-ready version will be released in April.

Our initial approach to sound λ-calculus reduction (i.e., featuring first-class lambdas, an essential component of most high-level languages) will be through an Elementary Affin Logic checker, which will restrict the shape of admissible programs to ensure compatibility. It will work like Rust's borrow checker, and will be a temporary complication to preserve maximum performance and correctness. The set of covered terms will then be slowly expanded, all the way to the full λ-calculus, via existing solutions such as brackets/croissants; always cautiously, as to not impact performance.

With HVM shipped, stable, and fast, it will become a great compile target for high-level programming languages, and double as a functional engine for all sorts of symbolic algorithms. We'll then shift our attention to building some exciting products that we thought of, which will be revealed later on, as well as providing support to community builders creating products on top of HVM - which will always be 100% open-source and completely free, as it should. I'm particularly excited about applications to modular execution layers, purely functional game engines, and symbolic AI architectures.

I always dreamed of a world where high-level languages, from Python to Haskell, would perform as well as, or better than, lower-level ones like C. While we're still far from that reality, I believe HVM has the potential to be a building block that'll point us in that direction, and we're working hard at HOC to make the most of this potential.

Follow our work by joining our Discord:

https://t.co/Rbbma6zjag

All our projects are hosted on GitHub:

https://t.co/XEI4d6sTJn

That's it for now. Thanks for reading!

[Nicolas Pouillard @npouillard]

@VictorTaelin Application: optimizing the pigeonhole principle would find collisions for any concrete hash function (https://t.co/aD1iPDK7H2).

[Nicolas Pouillard @npouillard]

@VictorTaelin "any low-order functions", out of this scope are functions on streams (https://t.co/Bp3Rtoz7m0) and polymorphic functions too.

[Nicolas Pouillard @npouillard]

@VictorTaelin I can't see how a simple unification can solve this.

[Nicolas Pouillard @npouillard]

@VictorTaelin Where can I read about the Self construct?

[Nicolas Pouillard @npouillard]

@VictorTaelin Can we reproduce this? I don't see the recent commits in the repos

[Ia Aanstoot @ia_aanstoot]

I am overwhelmed by the support we are getting for this campaign.

There was a technical issue with the petition but we believe it's fixed now.

So please sign if you haven't already!
https://t.co/Nbm4dwXdJ7