
Homo Ludens | CTO @ Unknown Group | Founder of @SciSportsNL & @BallJamesAI (now @Inmotio)
Amersfoort, Nederland
Joined October 2010
- Tweets7.7K
- Following1.2K
- Followers2.9K
- Likes4.1K
Giels Brouwer retweeted
We spent $2m so you don't have to.
And so you can learn how to train MoEs.
Mixture-of-Experts LLMs are what powers the AIs you use everyday, and few people in the world know how to train them at scale (a handful at each top AI lab).
This handbook distills everything we learned in meticulously crafted chapters and figures.
Disclaimer: it may take days to read.
We hope to see 1000x more people able to train their own frontier models all over the world! 🤗
Congrats @Nouamanetazi @lvwerra @Thom_Wolf
Giels Brouwer retweeted
Echt ongekend dit.
Hoe losgezongen van de werkelijkheid kan een minister van volkshuisvesting zijn?
Nederlanders hebben weinig tot niks te verwachten van dit kabinet. 😞
telegraaf.nl/video/woonminis…
Giels Brouwer retweeted
In Sweden today, there was a match with 11 GOALS in the SECOND HALF!
48': AIK 0-1 Brommapojkarna.
55': AIK 1-1 Brommapojkarna.
59': AIK 1-2 Brommapojkarna.
61': AIK 1-3 Brommapojkarna.
65': AIK 1-4 Brommapojkarna.
70': AIK 1-5 Brommapojkarna.
73': AIK 2-5 Brommapojkarna.
81': AIK 3-5 Brommapojkarna.
83': AIK 4-5 Brommapojkarna.
90+8': AIK 5-5 Brommapojkarna.
90+9': AIK 5-6 Brommapojkarna.
INCREDIBLE.
Giels Brouwer retweeted
Aviation solved AI slop in 1986.
Karpathy is telling people to ask their LLM to write in ASD-STE100, a controlled language built for aircraft maintenance manuals. I went down the rabbit hole on where this spec comes from, and the backstory makes the hack even better.
Back in the late 1970s, European airlines were handing English manuals to mechanics around the world who mostly didn't speak English as a first language. One ambiguous sentence in a fuel line procedure can get people killed. So the industry built a version of English where ambiguity is banned at the dictionary level.
The spec allows roughly 900 approved words. Each word gets one meaning and one part of speech. "Close" can only be a verb, because "close the door" and "the close door" can't be allowed to coexist in a repair manual. "Follow" only means to come after. If you want obey, you write obey.
Procedural sentences max out at 20 words. One instruction per sentence. Active voice only. Around 1,200 common words are explicitly banned, each with a mandated replacement. You don't "commence pumping." You start the pump.
English has about 170,000 words in current use. Aviation decided mechanics get 900, and planes became safer to maintain because of it.
Now flip it to LLMs. Models hedge and synonym-cycle because the training data rewards sounding fluent. STE is 40 years of accumulated rules for stripping exactly that out, written by people whose readers would die if a sentence could be read two ways. It might be the strongest anti-slop prompt in existence, and it's a free PDF.
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
Giels Brouwer retweeted
jev is insane 🫣
it makes realtime virtual try-on hauls possible.
built this experiment for Drape with @typesafeai
> i talk
> jev reads transcript + what i'm wearing
> picks from my closet
> changes my outfit in realtime
cost: $0.0011 per decision
time: ~620ms per decision
imagine getting ready like this:
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
De Transferzomer van Jordi Cruijff: linkedin.com/pulse/de-transf…
Giels Brouwer retweeted
Tigers look orange to humans because we are trichromats, but they appear green to deer and wild boars because they are dichromats.
Toch knap dat Ziggo dit altijd voor elkaar krijgt bij cruciale live events als de Klassieker of spannende WK wedstrijden.. ongekend onbetrouwbaar helaas. @ZiggoWebcare
Giels Brouwer retweeted
Noord-Korea kan nog een puntje zuigen aan deze vorm van propaganda.
Children are asking for a safer online world.
We asked Europeans aged 12–17 for their views on apps, websites, games & social media:
🚸80% support rules for loot boxes
🚸78% support age limits
🚸48% want controls on features like infinite scroll
link.europa.eu/CGDqYJ
Readers added context they thought people might want to know
The survey was a voluntary online consultation with 4,786 self-selected respondents, not representative of all EU children aged 12-17, with 38% from Germany and 33% from Romania. eu-for-children.europa.eu/digital-fairne… eu-for-children.europa.eu/system/files/2…
Giels Brouwer retweeted
North Korean leader Kim Jong-Un assured that the new intercontinental missiles programme would be examined and debated within the relevant committees
Dans une interview donnée au média suisse « Bluewin », le président de la FIFA, Gianni Infantino, a assuré que l'élargissement de la Coupe du monde à 64 équipes serait examiné et débattu au sein des comités compétents. > l.lequipe.fr/1FM
Giels Brouwer retweeted
🚨🇪🇺 LE PARLEMENT EUROPÉEN VOTE AUJOURD'HUI SUR LE « CHAT CONTROL » !
Le Parlement européen vote aujourd'hui sur une procédure d'urgence qui pourrait permettre un vote final dès jeudi sur le projet « Chat Control », malgré son rejet par les eurodéputés en mai dernier.
Le texte autoriserait le scan automatisé des messages privés sur des applications comme WhatsApp, Signal ou iMessage. Ses opposants dénoncent une « backdoor législative » qui contournerait le vote précédent du Parlement et pourrait concerner près de 470 millions de citoyens européens.
Giels Brouwer retweeted
C'est passé. 331 contre 304.
Cet après-midi le Parlement européen a voté l'urgence sur Chat Control le scan de vos messages privés. Pas le vote final : jeudi. La manœuvre est plus intéressante que le résultat.
Le Parlement avait déjà refusé ce texte. La dérogation a expiré en avril. Alors le Conseil l'a ressuscité en seconde lecture la procédure où il faut réunir une majorité absolue de 361 voix pour le bloquer, où les abstentions comptent comme des oui, calée pile sur la dernière séance avant l'été quand la moitié de l'hémicycle est déjà en vacances. Le service juridique du Conseil lui-même dit que ça viole l'article 7 de la Charte. Ils le savent. Ils le font quand même.
Mais voici ce que personne ne veut voir.
Le problème n'est pas que ces gens soient malveillants. Ils ne le sont probablement pas. La technocratie de Bruxelles n'utilisera sans doute jamais cet outil pour vous écraser elle est trop molle, trop procédurière, trop occupée à se justifier.
Le problème est qu'on ne construit jamais une infrastructure pour le pouvoir qu'on a. On la construit pour le pouvoir qui vient.
Et ce qu'ils assemblent, brique par brique, dérogation « volontaire » par dérogation « volontaire », c'est l'arsenal parfait. Le jour où une vraie crise économique portera au pouvoir des gens qui, eux, n'auront aucun scrupule, ils n'auront rien à inventer. Tout sera déjà là. Le scan généralisé, normalisé, légal, opérationnel. Une arme chargée posée sur la table, qu'il suffira de ramasser.
Hayek l'avait compris : la route de la servitude est pavée par des administrateurs pleins de bonnes intentions. On ne cède pas sa liberté d'un coup. On la cède ligne par ligne, « pour protéger les enfants », « pour lutter contre le terrorisme », toujours pour une raison que personne n'ose contester.
L'histoire ne retiendra pas ces gens comme les protecteurs de quoi que ce soit. Elle les retiendra comme les architectes de la cage. Ceux qui ont bâti l'appareil, et l'ont tendu, souriants, à ceux qui savaient s'en servir.
Jeudi, 361 voix. C'est le dernier verrou.
Giels Brouwer retweeted
🚨 Folarin Balogun is eligible to play against Belgium.
FIFA sources had insisted US Soccer had no appeal. Sources say US Soccer did not attempt one as a result.
And Balogun had spoken on Monday here in Seattle about missing the game telling us, “I have to accept it…”
Yet FIFA have decided to hand Balogun a suspended one-game ban, which is in their powers as per Article 27. The red card stands but Balogun is free to play.
Cristiano Ronaldo also given a suspended ban before the tournament allowing him to play Portugal’s opening World Cup games.
Giels Brouwer retweeted
Btw, this is the red card Balogun was shown...
Come on, tell me he didn't deserve it and that the game against Belgium should have been called off.
Trump's conflict of interest here is huge, and his close relationship with Infantino... I don't know, something smells really fishy here
Writer: Samuel
This media is unavailable
🚨 BREAKING:
🇺🇸🇧🇪 USA’s Balogun is free to play in the next World Cup match against Belgium after his red card suspension was lifted by FIFA.
Trump:
"Thank you to FIFA for doing what was right, and reversing a great injustice!"
Trump said it was a terrible decision, and now they're removing the red card, just as the host country's president said that on social media... Coincidence or corruption?
Source: Fox News, @realDonaldTrump / Writer: Samuel
Giels Brouwer retweeted
Naar achteren graag!
Tijdens het concert van Sophie Straat werden witte mannen gevraagd om naar achteren te gaan zodat anderen vooraan meer ruimte kregen. Dit werd gepresenteerd als inclusie.
Rosa Parks weigerde in 1955 haar plaats af te staan omdat mensen niet op basis van huidskleur behandeld mogen worden.
Vroeger was “naar achteren in de bus” racisme. Vandaag wordt “naar achteren in de zaal” gezien als progressief.
Iedereen dezelfde kansen en rechten is niet meer voldoende, het draait het nu om equity: gelijke uitkomsten tussen groepen. Beleid moet actief ingrijpen op basis van ras, geslacht en achtergrond om de gewenste resultaten te bereiken.
In Nederland zie je dat terug in vrouwenquota en streefcijfers in besturen en organisaties.
Selectie gebeurt niet alleen op basis van individuele kwaliteiten, maar ook op geslacht en herkomst om de groepssamenstelling te corrigeren.
Mensen worden daardoor steeds meer beoordeeld als lid van een groep in plaats van als individu. Het klassieke ideaal van kleurblinde gelijke behandeling maakt plaats voor beleid dat onderscheid maakt om gelijke uitkomsten af te dwingen.
Positieve discriminatie is ook discriminatie!
Discriminatie verdwijnt niet. Ze verandert alleen van richting.
Mijn oproep: Als je echt tegen discriminatie bent moet je geen beweging creëren die juist discrimineert. En al zeker niet de overheid die macht geven. Die zal in de toekomst ook zich tegen jou kunnen keren.
AI FOOTBALL ANALYSIS. A FULL COMPUTER VISION SYSTEM. BUILT ON YOLO, OPENCV, AND PYTHON.
You upload a regular match video. No sensors, no GPS trackers, just camera footage.
The neural network finds every player, referee, and ball on its own. Every frame, in real time.
KMeans clustering breaks down jersey colors pixel by pixel. The system splits players into teams automatically. Without a single manual hint.
Optical Flow tracks camera movement. Separates it from player movement. Perspective Transformation converts pixels into real meters.
Speed of every player. Distance covered. Ball possession percentage. All calculated automatically.
Four hours of tutorial from zero to a working system.
The model is trained on real Bundesliga matches. Runs on a regular GPU. Python code - take it and run.
Sports analytics is no longer behind closed doors. AI leveled the playing field.







































