Hacker News

Chose amusante, alors que ce matin même, j'ai découvert l'existence de o1, sortie il y a seulement quelques jours, le 5 décembre 2024.
Voilà que je découvre ce soir, dans ce thread Hacker News la sortie de o3 le 20 décembre 2024 : "OpenAI O3 breakthrough high score on ARC-AGI-PUB".

Les releases sont très réguliers en ce moment, il est difficile de suivre le rythme 😮 !

Dans ce thread, j'ai découvert le prix ARC (https://arcprize.org), lancé le 11 juin 2024, par le français Francois Chollet, basé sur le papier de recherche "On the Measure of Intelligence" sorti en 2019, il y a 5 ans.

ARC est un outil de mesure de AGI.

#JaimeraisUnJour prendre le temps de lire On the Measure of Intelligence.

Je lis ici :

OpenAI o3 Breakthrough High Score on ARC-AGI-Pub

OpenAI's new o3 system - trained on the ARC-AGI-1 Public Training set - has scored a breakthrough 75.7% on the Semi-Private Evaluation set at our stated public leaderboard $10k compute limit. A high-compute (172x) o3 configuration scored 87.5%.

This is a surprising and important step-function increase in AI capabilities, showing novel task adaptation ability never seen before in the GPT-family models. For context, ARC-AGI-1 took 4 years to go from 0% with GPT-3 in 2020 to 5% in 2024 with GPT-4o. All intuition about AI capabilities will need to get updated for o3.

source

Plus loin, je lis :

However, it is important to note that ARC-AGI is not an acid test for AGI – as we've repeated dozens of times this year. It's a research tool designed to focus attention on the most challenging unsolved problems in AI, a role it has fulfilled well over the past five years.

Passing ARC-AGI does not equate to achieving AGI, and, as a matter of fact, I don't think o3 is AGI yet. o3 still fails on some very easy tasks, indicating fundamental differences with human intelligence.

source

Donc, j'en conclus qu'il ne faut pas s'emballer outre mesure sur les résultats de ce test, bien que les progrès soient impressionnants.

La première partie du thread semble aborder la thématique du coût financier de o3 versus un humain : 309 commentaires.

Dans ce commentaire #JaiDécouvert le papier de recherche "H-ARC: A Robust Estimate of Human Performance on the Abstraction and Reasoning Corpus Benchmark" qui date de 2024.

#JaiDécouvert l'outil de "version manager" nommé aqua, une alternative à Mise et Asdf codé en Golang.

Ce projet semble avoir débuté en août 2021.

J'ai fait quelques recherches au sujet d'aqua sur Hacker News, j'ai trouvé très peu d'occurrences. J'ai trouvé "Ask HN: Homebrew, Asdf, Nix, or Other?".

Je pense qu'aqua est bien moins populaire que Asdf et Mise.

Au 4 décembre 2024 :

• aqua : 901 stars GitHub
• Mise : 10 400 stars GitHub
• Asdf : 22 200 stars GitHub

#JaiLu le thread Hacker News : What's Next for WebGPU (WebGPU).

En lisant la release note v3.0.3 de wal-g, j'ai découvert l'extension PostgreSQL nommée OrioleDB.

OrioleDB is a new storage engine for PostgreSQL, bringing a modern approach to database capacity, capabilities and performance to the world's most-loved database platform.

OrioleDB consists of an extension, building on the innovative table access method framework and other standard Postgres extension interfaces. By extending and enhancing the current table access methods, OrioleDB opens the door to a future of more powerful storage models that are optimized for cloud and modern hardware architectures.

Le projet OrioleDB a commencé en février 2022 par un développeur de Supabase : Alexander Korotkov.

Les commentaires de ce thread Hacker News semblent très enthousiastes https://news.ycombinator.com/item?id=30462695.

Dans la page "Introductions" de la documentation, je lis :

Differentiators

The key technical differentiations of OrioleDB are as follows:

No buffer mapping and lock-less page reading

In-memory pages in OrioleDB are connected with direct links to the storage pages. This eliminates the need for in-buffer mapping along with its related bottlenecks. Additionally, in OrioleDB in-memory page reading doesn't involve atomic operations. Together, these design decisions bring vertical scalability for Postgres to the whole new level.

MVCC is based on the UNDO log concept

In OrioleDB, old versions of tuples do not cause bloat in the main storage system, but eviction into the undo log comprising undo chains. Page-level undo records allow the system to easily reclaim space occupied by deleted tuples as soon as possible. Together with page-mergins, these mechanisms eliminate bloat in the majority of cases. Dedicated VACUUMing of tables is not needed as well, removing a significant and common cause of system performance deterioration and database outages.

Copy-on-write checkpoints and row-level WAL

OrioleDB utilizes copy-on-write checkpoints, which provides a structurally consistent snapshot of data every moment of time. This is friendly for modern SSDs and allows row-level WAL logging. In turn, row-level WAL logging is easy to parallelize (done), compact and suitable for active-active multimaster (planned).

J'ai lu le billet "Rethinking PostgreSQL buffer mapping for modern hardware architectures". Je pense avoir compris que l'implémentation actuelle de PostgreSQL utilise un "buffer mapping" autrefois bien adapté aux contraintes matérielles.

J'ai compris qu'OrioleDB propose une nouvelle approche, spécialement conçue pour tirer parti des SSD rapides, ce qui lui permet d’atteindre des performances nettement supérieures à celles de l’implémentation existante.

Dans le thread Hacker News "Rsbuild – A Better Vite?" #JaiDécouvert :

• Rsbuild : The Rspack-based build tool. It's fast, out-of-the-box and extensible.
• SWC : (stands for Speedy Web Compiler) is a super-fast TypeScript / JavaScript compiler written in Rust.
• VoidZero : a company dedicated to building an open-source, high-performance, and unified development toolchain for the JavaScript ecosystem.
• Oxc : is building a parser, linter, formatter, transformer, minifier, resolver ... all written in Rust.
• Rolldown : Rolldown is a JavaScript/TypeScript bundler written in Rust intended to serve as the future bundler used in Vite.

Voici ce que j'ai compris.

Tous ces outils sont écrits en Rust.

Rsbuild est une alternative à : Vite, Create React App et Vue CLI et qui offre d'excellente performance (les tâches de build… sont exécutées bien plus rapidement).

Jiahan Chen, développeur de chez ByteDance, a commencé le projet Rsbuild en octobre 2023.

Dans le thread HackerNews je lis ce commentaire :

The better, faster, Rust-powered Vite is… Vite.

https://voidzero.dev/posts/announcing-voidzero-inc

J'ai creusé le sujet et j'ai compris que le créateur de Vite, Evan You a fondé une société nommée VoidZero, composée de core développeurs des projets Oxc, Vite, Rolldown.

Accel a injecté 4,6 millions de dollars dans VoidZero avec comme objectif de financer le développement de Rolldown qui sera intégré dans une future version de Vite.
D'après ce que j'ai compris, Rolldown utilise Oxc.

Je me demande si Accel envisage de tirer des bénéfices directs de VoidZero ou si cette initiative relève davantage du mécénat. Du côté des intérêts indirects : plusieurs sociétés du portefeuille d'Accel utilisent la stack Javascript, ce qui permet de financer et de mutualiser le développement d'outils clés.

Voici les points principaux que je retiens. Rsbuild semble une alternative performante Vite qui est utilisable dès aujourd'hui.
Le projet Vite est bien structuré et financé, ce qui lui permettra de sortir une nouvelle version optimisée.
Pour ma part, j’espère voir le projet VoidZero réussir afin d’éviter une dilution des efforts au sein de la communauté Javascript dans une multitude de projets.

#JaiLu l'article Ode aux perdants de Ploum qui commente l'article fedi is for losers ( wingolog) qui traite de Fediverse.

La question est provocante et intelligente : le Fediverse semble être un repère d’écologistes, libristes, défenseurs des droits sociaux, féministes et cyclistes. Bref la liste de tous ceux qui ne sont pas mis en avant, qui semblent « perdre ».

Je n’avais jamais vu les choses sous cet angle.

-- from

Personnellement, je ne suis pas surpris, j'ai souvent reçu cette remarque de la part de profils non hackers, comme des CEO, des responsables marketing, ou des personnes issues de la culture startup.

J'utilise Fediverse, Reddit, Hacker News, Lobster, les forums… parce que j'y trouve des signaux faibles, des Maven, des hackers.

Je suis conscient que le Fediverse représente une barrière à l'entrée, mais je considère cela comme une qualité. Cela contribue à éviter, dans une certaine mesure, le phénomène du septembre éternel.

Personnellement, je trouve peu d'information sur Twitter et encore moins sur LinkedIn.
Je trouve que ces plateformes sont majoritairement saturées de bruit, de contenu promotionnel déguisé et de messages de signalement de statut social.

Pour être totalement transparent, jusqu'à présent, je trouve que mon feed Mastodon est de mauvaise qualité, j'y trouve peu d'Information. Je préfère nettement le contenu de mon flux RSS, qui agrège des sources provenant de Hacker News, Reddit et de divers blogs.

J'aimerais voir sur Fediverse plus de chercheurs, plus d'informations qui proviennent de revues à comités de lectures.

#JaiLu l'article Pledging $300,000 to the Zig Software Foundation de Mitchell Hashimoto. Très bonne nouvelle pour Zig 🙂.

Deux threads Hacker News à ce sujet : https://news.ycombinator.com/item?id=41712239 et https://news.ycombinator.com/item?id=41711601.

#JaiDécouvert au nouveau SaaS de signature électronique : OpenSign (from).

Le code source est disponible sous licence AGPL. Le projet semble actif.

Le projet semble avoir été créé en 2023 par un anglais nommé Amol Shejole.

Je viens de découvrir l'existence d'un long thread Hacker News qui date de novembre 2023 : Show HN: OpenSign – Open source alternative to DocuSign. Je n'ai pas encore pris le temps de lire tout le thread, mais il semble contenir des commentaires intéressants.

Voir aussi la note 2023-07-24_2046.

#OnMaPartagé la vidéo YouTube nommée Pourquoi les projets informatiques vont dans le mur ?.

Je ne connaissais pas la dérive de ce projet de migration de SAP vers ERP Oracle Fusion de la mairie de Birmingham : La 2e ville du Royaume-Uni s'est déclarée en faillite, plombée entre autres par les dérives d'un projet de migration vers l'ERP Oracle Fusion. Après des années de retards, de problèmes de contrôle, de gestion hasardeuse, la facture du projet a quintuplé pour atteindre 115 M€.

J'ai fait quelques recherches dans les commentaires Hacker News et j'ai trouvé ceci :

• Birmingham council's "equal pay" bankruptcy provided cover for Oracle disaster

My state of Oregon paid Oracle something like $250M for a healthcare system that never materialized. The lawsuit that followed was settled for $100M, but most of that was “free” Oracle licenses and no less than $60M of customer support.

-- from

🙈

Actually it's the opposite. I worked for an NGO half a decade ago, and they wanted to add 2FA authentication to their login system used by ~400 staff. I created a quick demo using Google Authenticator in less than a week.

However the director of IT didn't like this solution. He insisted we use RSA keys and hire IBM to build a solution using that - I think the original estimate was a few million $ and it would take six months or so for their team (of basically new graduates) to build.

I asked my boss why the director was pushing so much for IBM to build it:

He told me that if we build it, and it doesn't work, then the director has to take the blame. If IBM build it, and it doesn't work, then IBM take the blame.

-- from

😭

Yes. There’s also the revolving door problem. The bureaucrat making the decision is often angling for a cushy role at the contractor. And the contractor is making the offer under the table to get the gig. From the decision makers perspective, it doesn’t matter if the project succeeds, they’ll be long gone. I’ve seen this with my own eyes.

-- from

😔

How do they stay in business?

Oracle's main line of competency is not providing good software services. They are in fact, specialists of acquiring government contracts.

-- from

Has anyone ever heard of an Oracle project that has ever ended well?

After 25 years in IT consulting all over the US in different businesses, Oracle is never NOT a 4 letter world, where projects involving them always over-promise, under-deliver, and project costs end up some 3-10x any initial projection. Particularly any ERP, CRM, now EMR in hospitals as well since acquiring Cerner. Anyone that does use them only do as a necessary evil of some dubious or shady circumstances, otherwise Oracle is a term almost universally reviled and hated amongst end users and organization leadership alike.

• How poor IT project management failed Birmingham Oracle implementation

An insider at Birmingham City Council who has been closely involved in the project told Computer Weekly it went live “despite all the warnings telling them it wouldn’t work”.

Discussing how the Oracle system failure impacted the council’s ability to manage its finances, the insider said: “We were withholding thousands of supplier payments because we couldn’t make any payments. We didn’t have any direct debits for cash collection. We had no cash collection, no bank reconciliation. When you do a project of this size, you must have your financial reporting and you must have a bank reconciliation system that tells you where the money is, what’s being spent and what’s being paid.”

Since going live, the Oracle system effectively scrambled financial data, which meant the council had no clear picture of its overall finances.

The insider said that by January 2023, Birmingham City Council could not produce an accurate account of its spending and budget for the next financial year: “There’s no way that we could do our year-end accounts because the system didn’t work.”

-- from

The June 2023 Birmingham City Council report to cabinet stated that due to issues with the council’s bank reconciliation system (BRS), a significant number of transactions had to be manually allocated to accounts rather than automatically via the Oracle system. However, Computer Weekly has seen an enterprise resource planning (ERP) implementation presentation given in 2019, which shows that the council was made aware of these issues at that time, three years before the go-live date.

...

The lack of a functioning BRS has directly contributed to the council’s current financial crisis. In BCC’s April 2024 audit report, councillor Grindrod said: “We couldn’t accurately collect council tax or business rates.”

As of April 2024, it is believed the manual intervention needed for the bank reconciliation process is costing the council £250,000 per month.

-- from

• City council audit trail is an audit fail after disastrous Oracle ERP rollout

Peoplesoft/Oracle ERP has had over 30 year of experience selling to local governments globally.

When dealing with procurement in countries that aren't the US, riles and regulations are much more difficult and unintuitive, and also provide marginal RoI.

This is why companies like Workday and Salesforce don't care to compete with Oracle or SAPs in these kinds of contracts - they don't have the right relationships with channel partners and systems integrators needed.

When a city council places a tender for an ERP system, they won't be doing the work in-house due to regulatory and budget allocation reasons. Instead they'll farm out the work to local contractors, MSPs, and Systems Integrators instead.

PLG driven companies like Workday and Salesforce dislike working with SIs and MSPs as much because channel partners don't care about upselling features in the products they bought - they wanna keep the customer satiated instead.

Also, the dollars spent getting the contract might not have a significant RoI when factoring the contract size itself.

-- from

J'ai trouvé ceci sur Reddit :

• Why is ERP software still so terrible? And so expensive?

I work as an implementation consultant for finance software, albeit a much smaller scale one than SAP/Oracle, but I’ve been in this line of work for several years. This kind of thing is fairly common, and I think a several year long project that Lidl undertook to do the same thing went the same way, which is money down the toilet.

All of the cost is in the services. The clients are billed at an hourly or daily rate for meetings, project management, issue resolution, emails etc. The problem with huge projects like these is that institutions like councils have their own very specific processes and are unwilling to change, because more often than not the employees are set in their ways and don’t want to learn anything, but also changing one or two things could have a huge effect on other things. In finance systems there are often several integrations, both incoming and outgoing, and the client will need a tailored solution to migrate everything. It would be easy if the client accepted that some things would have to change, but SAP and Oracle are very very customisable, and depending on the company doing the implementation everything will get customised to where there are all these new moving parts, new problems, and new things to learn. People also change their minds about what they want all the time, especially if the project wasn’t scoped or managed correctly.

In short, the software may be established, but the way it is implemented never is. Templates exist, but every business is different, and the public sector is particularly messy to deal with (underpaid and undertrained staff, tend to be a bit less motivated than private sector IME).

-- from

I tried several different SaaS solutions, mostly aimed at breweries. It was infuriating watching these things fit 98% of the requirement, but the missing 2% rendered the whole system inoperable. But the choice is just take it for £50/user/month or leave it.

I tried some open source options which were better, but were still far too rigid to a point where we'd be shaping our whole operations around the way our ERP wants us to do things, and getting to that point would take a lot of development.

In the end, it was legitimately easier to do it ourselves. And by that I mean me, a distiller, to develop a system from scratch. We now have a fully integrated ERP system that works around our processes, but was built using good practices so that things are very versatile and don't inherently depend on working a certain way. Some of these systems had moronic limitations that wouldn't even allow for an output to be used as an input into another process. Apparently everything has to be made in one process! Can you imagine how many use cases that single, completely unnecessary, restriction... restricts?

-- from

#JaiDécouvert pgroll (from).

PostgreSQL zero-downtime migrations made easy.

#JaiLu en partie ce thread Hacker News de 2023.

Après avoir lu partiellement la documentation, j'ai l'impression que pgroll est simple à utiliser pour des migrations qui restent simples.
J'ai lu la section Raw SQL et #JeMeDemande si pgroll reste pratique à utiliser pour des migrations complexes, par exemple, split d'une table en plusieurs tables, merge de tables…

Je ne suis pas très motivé pour apprendre un nouveau DSL, c'est-à-dire, le format de migrations de pgroll à la place des instructions DDL (Data Definition Language) SQL (create, alter…).

Pour le moment, j'ai réussi à réaliser "à la main" des migrations en douceur : mise en place de view, de triggers… qui sont par la suite supprimés.

Je pense que pgroll serait très pratique avec une fonctionnalité Skew Protection pour un projet où les déploiements en production en journée sont fréquents et qui ne souhaite pas imposer aux utilisateurs de rafraîchir leurs pages.