Adrien Dorland

korben.info/razor-1911-his… Quelle nostalgie en lisant ce super papier de @Korben 👏

this project stores millions of text chunks inside a video file (mp4) then runs sub-second semantic search on it - no vector DB, no servers - uses 10x less RAM & storage - no internet required it's called Memvid and it just broke my brain

Salut @TraderepublicFR, apparemment, les transferts de PEA vers chez vous sont tous bloqués. il se passe quoi ...? il parait que vous ne satisfaisiez pas les prérequis règlementaires (certificats d'identification notamment)

THREAD : TikTok a mis en place 8 protections pour éviter de fuiter 750GB de données par jour sur leur appli. Je vais vous détailler comment contourner chacune de ces sécurités, et pourquoi j'ai besoin des données de plusieurs millions de créateurs de contenus.

Cc @Stromae Wow 🤯

Deux chercheurs 🇫🇷 du CEA ont créé une nouvelle catégorie de métaux, aux propriétés exceptionnelles, grâce à un procédé révolutionnaire baptisé "Hanetec", qui permet de recréer de la matière sous la forme d’éclairs électriques nanométriques. ▶️ cea.fr/Pages/actualit…

En 2023, mes tweets ont été vus plus de 68 millions de fois. Selon mes analyses : - La monétisation Twitter rapporte environ 1€43 par million de vues. - Twitter redistribue moins de 2% de ses revenus publicitaires. Lien dans le tweet suivant 👇

Et si les villes devenaient des montagnes, collines et plateaux, et les campagnes des mers ou des lacs ? Voici une carte de la France version "topographie humaine", tout juste finie ! 🗻👥🗻

Quick lesson in the dangers of data contamination. Years ago, I came up with an acronym for remembering the periods of the Paleozoic era — “Catastrophic Overthrow Started Different Colder Period”. I was curious if ChatGPT could guess what it stood for. 1/4

Entangled #fxhash

OpenAI leaked Q* so let’s dive into Q-Learning and how it relates to RLHF. Q-learning is a foundational concept in the field of artificial intelligence, particularly in the area of reinforcement learning. It's a model-free reinforcement learning algorithm that aims to learn the value of an action in a particular state. The ultimate goal of Q-learning is to find an optimal policy that defines the best action to take in each state, maximizing the cumulative reward over time. Understanding Q-Learning Basic Concept: Q-learning is based on the notion of a Q-function, also known as the state-action value function. This function takes two inputs: a state and an action. It returns an estimate of the total reward expected, starting from that state, taking that action, and thereafter following the optimal policy. The Q-Table: In simple scenarios, Q-learning maintains a table (known as the Q-table) where each row represents a state and each column represents an action. The entries in this table are the Q-values, which are updated as the agent learns through exploration and exploitation. The Update Rule: The core of Q-learning is the update rule, often expressed as: \[ Q(s,a) \leftarrow Q(s,a) + \alpha [r + \gamma \max_{a'} Q(s', a') - Q(s, a)] \] Here, \( \alpha \) is the learning rate, \( \gamma \) is the discount factor, \( r \) is the reward, \( s \) is the current state, \( a \) is the current action, and \( s' \) is the new state. (See image below). Exploration vs. Exploitation: A key aspect of Q-learning is balancing exploration (trying new things) and exploitation (using known information). This is often managed by strategies like ε-greedy, where the agent explores randomly with probability ε and exploits the best-known action with probability 1-ε. Q-Learning and the Path to AGI Artificial General Intelligence (AGI) refers to the ability of an AI system to understand, learn, and apply its intelligence to a wide variety of problems, akin to human intelligence. Q-learning, while powerful in specific domains, represents a step towards AGI, but there are several challenges to overcome: Scalability: Traditional Q-learning struggles with large state-action spaces, making it impractical for real-world problems that AGI would need to handle. Generalization: AGI requires the ability to generalize from learned experiences to new, unseen scenarios. Q-learning typically requires explicit training for each specific scenario. Adaptability: AGI must be able to adapt to changing environments dynamically. Q-learning algorithms often require a stationary environment where the rules do not change over time. Integration of Multiple Skills: AGI implies the integration of various cognitive skills like reasoning, problem-solving, and learning. Q-learning primarily focuses on the learning aspect, and integrating it with other cognitive functions is an area of ongoing research. Advances and Future Directions Deep Q-Networks (DQN): Combining Q-learning with deep neural networks, DQNs can handle high-dimensional state spaces, making them more suitable for complex tasks. Transfer Learning: Techniques that enable a Q-learning model trained in one domain to apply its knowledge to different but related domains can be a step towards the generalization needed for AGI. Meta-Learning: Implementing meta-learning in Q-learning frameworks could enable AI to learn how to learn, adapting its learning strategy dynamically - a trait crucial for AGI. Q-learning represents a significant methodology in AI, particularly in reinforcement learning. It is not surprising that OpenAI is using Q-learning RLHF to try to achieve the mystical AGI.

Et maintenant que va-t-il se passer ? La question a d'autant plus d'acuité que le récent licenciement de Sam Altman doit nous rappeler que la course technologique est pavée de rebondissements qui sont autant d'opportunités pour ceux qui savent les saisir

Grok grok Grok? X.ai

@fabricekordon @LaMatriceCarree 🤣 j’espère que @LaMatriceCarree à un gros budget rachat d’objets de collection 🤭

@revokiso @LaMatriceCarree Un jolie collection... le label UPMC qui n'existe plus (nous sommes désormais Sorbonne Université depuis la fusion avec Paris IV) va même lui faire prendre de la valeur, comme pour les œuvres de artistes morts 😂😎🤣

Un peu de ménage dans la cave… et un petit poke à @fabricekordon De bon souvenirs à #Jussieu #lip6

@LaMatriceCarree @fabricekordon Hello, faites moi signe en DM si vous voulez qu'on se voit pour faire une "transmission de patrimoine" :-)

@revokiso @fabricekordon Faites moi signe si vous en trouvez d'autres 😔

@LaMatriceCarree @fabricekordon Voilà :-)

@LaMatriceCarree @fabricekordon Je vais me débrouiller pour les recup ;-) Affaire à suivre (dimanche)

J’ai tout recup :-) même celui de POSIX :-)

@LaMatriceCarree @fabricekordon 🫣 il y en avait d’autre en plus …

@revokiso @fabricekordon Un petit bout de l'histoire de Jussieu s'en est allé 😭

@LaMatriceCarree @fabricekordon Je vous l’aurai donné bien volontiers , mais je crois qu’il est parti à la poubelle ce matin 🙄

@revokiso @fabricekordon Je peux vous le racheter ? 👀