information analyst
43.7K views | +1 today
Follow
 
Rescooped by michel verstrepen from 21st Century Learning and Teaching
onto information analyst
Scoop.it!

Infographie sur la gouvernance de l’information

Infographie sur la gouvernance de l’information | information analyst | Scoop.it

Dans le cadre de la production de la communauté qui s’est crée autour de cet observatoire, nous avons décidé de publier une infographie qui présente des grandes lignes autour d’un point de vue sur la gouvernance de l’information, et d’une partie des résultats.

 

Si le sujet de la gouvernance de l’information, ou du pilotage de l’information, des référentiels, des politique de gestion de l’information, .. vous intéresse, voici donc cette infographie.


Via Gust MEES
No comment yet.
information analyst
km, ged / edms, workflow, collaboratif
Your new post is loading...
Your new post is loading...
Rescooped by michel verstrepen from Les outils du Web 2.0
Scoop.it!

Mymind. La prise de notes avec IA qui change tout

Mymind. La prise de notes avec IA qui change tout | information analyst | Scoop.it
Mymind est une application de prise de notes IA qui se souvient de tout à votre place. Un deuxième cerveau pour booster votre productivité.

Via Fidel NAVAMUEL
No comment yet.
Rescooped by michel verstrepen from Time to Learn
Scoop.it!

Civitai: The Home of Open-Source Generative AI

Civitai: The Home of Open-Source Generative AI | information analyst | Scoop.it
Explore thousands of high-quality Stable Diffusion models, share your AI-generated art, and engage with a vibrant community of creators

Via Frédéric DEBAILLEUL
No comment yet.
Rescooped by michel verstrepen from Veille Usages numériques
Scoop.it!

Intelligences artificielles génératives

Intelligences artificielles génératives | information analyst | Scoop.it



L’Université de Bordeaux « met à disposition de la communauté enseignante une série de ressources pédagogiques permettant de comprendre le fonctionnement des IA génératives comme ChatGPT, leurs avantages, leurs limites, mais aussi de connaître les différentes utilisations qu'un étudiant, enseignant, ou tuteur peuvent en faire ».


Via Tice Doc Reims
No comment yet.
Rescooped by michel verstrepen from Time to Learn
Scoop.it!

ThinkAny - Moteur de Recherche IA

ThinkAny - Moteur de Recherche IA | information analyst | Scoop.it
ThinkAny est un moteur de recherche IA d'une nouvelle ère qui utilise la technologie RAG pour récupérer et agréger du contenu de haute qualité, combiné aux fonctionnalités de réponse intelligente de l'IA, répondant efficacement aux questions des utilisateurs.

Via Frédéric DEBAILLEUL
No comment yet.
Rescooped by michel verstrepen from TIC et TICE mais... en français
Scoop.it!

Où trouver des outils libres ? –

Où trouver des outils libres ? – | information analyst | Scoop.it
Framalibre est un projet de l'association Framasoft qui vise à recenser et promouvoir les logiciels libres et open source. Sur ce site sont répertorié

Via Elena Pérez
No comment yet.
Rescooped by michel verstrepen from E-Learning-Inclusivo (Mashup)
Scoop.it!

Large language models use a surprisingly simple mechanism to retrieve some stored knowledge

Large language models use a surprisingly simple mechanism to retrieve some stored knowledge | information analyst | Scoop.it

Researchers find large language models use a simple mechanism to retrieve stored knowledge when they respond to a user prompt. These mechanisms can be leveraged to see what the model knows about different subjects and possibly to correct false information it has stored.

Researchers demonstrate a technique that can be used to probe a model to see what it knows about new subjects.
 
Adam Zewe | MIT News
Publication Date:
March 25, 2024
Researchers from MIT and elsewhere found that complex large language machine-learning models use a simple mechanism to retrieve stored knowledge when they respond to a user prompt. The researchers can leverage these simple mechanisms to see what the model knows about different subjects, and also possibly correct false information that it has stored.

 

Large language models, such as those that power popular artificial intelligence chatbots like ChatGPT, are incredibly complex. Even though these models are being used as tools in many areas, such as customer support, code generation, and language translation, scientists still don’t fully grasp how they work.

In an effort to better understand what is going on under the hood, researchers at MIT and elsewhere studied the mechanisms at work when these enormous machine-learning models retrieve stored knowledge.

They found a surprising result: Large language models (LLMs) often use a very simple linear function to recover and decode stored facts. Moreover, the model uses the same decoding function for similar types of facts. Linear functions, equations with only two variables and no exponents, capture the straightforward, straight-line relationship between two variables.

The researchers showed that, by identifying linear functions for different facts, they can probe the model to see what it knows about new subjects, and where within the model that knowledge is stored.

Using a technique they developed to estimate these simple functions, the researchers found that even when a model answers a prompt incorrectly, it has often stored the correct information. In the future, scientists could use such an approach to find and correct falsehoods inside the model, which could reduce a model’s tendency to sometimes give incorrect or nonsensical answers.

“Even though these models are really complicated, nonlinear functions that are trained on lots of data and are very hard to understand, there are sometimes really simple mechanisms working inside them. This is one instance of that,” says Evan Hernandez, an electrical engineering and computer science (EECS) graduate student and co-lead author of a paper detailing these findings.

Hernandez wrote the paper with co-lead author Arnab Sharma, a computer science graduate student at Northeastern University; his advisor, Jacob Andreas, an associate professor in EECS and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL); senior author David Bau, an assistant professor of computer science at Northeastern; and others at MIT, Harvard University, and the Israeli Institute of Technology. The research will be presented at the International Conference on Learning Representations.

Finding facts

Most large language models, also called transformer models, are neural networks. Loosely based on the human brain, neural networks contain billions of interconnected nodes, or neurons, that are grouped into many layers, and which encode and process data.

Much of the knowledge stored in a transformer can be represented as relations that connect subjects and objects. For instance, “Miles Davis plays the trumpet” is a relation that connects the subject, Miles Davis, to the object, trumpet.

As a transformer gains more knowledge, it stores additional facts about a certain subject across multiple layers. If a user asks about that subject, the model must decode the most relevant fact to respond to the query.

If someone prompts a transformer by saying “Miles Davis plays the. . .” the model should respond with “trumpet” and not “Illinois” (the state where Miles Davis was born).

“Somewhere in the network’s computation, there has to be a mechanism that goes and looks for the fact that Miles Davis plays the trumpet, and then pulls that information out and helps generate the next word. We wanted to understand what that mechanism was,” Hernandez says.

The researchers set up a series of experiments to probe LLMs, and found that, even though they are extremely complex, the models decode relational information using a simple linear function. Each function is specific to the type of fact being retrieved.

For example, the transformer would use one decoding function any time it wants to output the instrument a person plays and a different function each time it wants to output the state where a person was born.

The researchers developed a method to estimate these simple functions, and then computed functions for 47 different relations, such as “capital city of a country” and “lead singer of a band.”

While there could be an infinite number of possible relations, the researchers chose to study this specific subset because they are representative of the kinds of facts that can be written in this way.

They tested each function by changing the subject to see if it could recover the correct object information. For instance, the function for “capital city of a country” should retrieve Oslo if the subject is Norway and London if the subject is England.

Functions retrieved the correct information more than 60 percent of the time, showing that some information in a transformer is encoded and retrieved in this way.

“But not everything is linearly encoded. For some facts, even though the model knows them and will predict text that is consistent with these facts, we can’t find linear functions for them. This suggests that the model is doing something more intricate to store that information,” he says.

Visualizing a model’s knowledge

They also used the functions to determine what a model believes is true about different subjects.

In one experiment, they started with the prompt “Bill Bradley was a” and used the decoding functions for “plays sports” and “attended university” to see if the model knows that Sen. Bradley was a basketball player who attended Princeton.

“We can show that, even though the model may choose to focus on different information when it produces text, it does encode all that information,” Hernandez says.

They used this probing technique to produce what they call an “attribute lens,” a grid that visualizes where specific information about a particular relation is stored within the transformer’s many layers.

Attribute lenses can be generated automatically, providing a streamlined method to help researchers understand more about a model. This visualization tool could enable scientists and engineers to correct stored knowledge and help prevent an AI chatbot from giving false information.

In the future, Hernandez and his collaborators want to better understand what happens in cases where facts are not stored linearly. They would also like to run experiments with larger models, as well as study the precision of linear decoding functions.

“This is an exciting work that reveals a missing piece in our understanding of how large language models recall factual knowledge during inference. Previous work showed that LLMs build information-rich representations of given subjects, from which specific attributes are being extracted during inference. This work shows that the complex nonlinear computation of LLMs for attribute extraction can be well-approximated with a simple linear function,” says Mor Geva Pipek, an assistant professor in the School of Computer Science at Tel Aviv University, who was not involved with this work.

This research was supported, in part, by Open Philanthropy, the Israeli Science Foundation, and an Azrieli Foundation Early Career Faculty Fellowship.


Via Charles Tiayon, juandoming
Charles Tiayon's curator insight, March 25, 10:31 PM

"Researchers find large language models use a simple mechanism to retrieve stored knowledge when they respond to a user prompt. These mechanisms can be leveraged to see what the model knows about different subjects and possibly to correct false information it has stored.

Researchers demonstrate a technique that can be used to probe a model to see what it knows about new subjects.
 
Adam Zewe | MIT News
Publication Date:
March 25, 2024
Researchers from MIT and elsewhere found that complex large language machine-learning models use a simple mechanism to retrieve stored knowledge when they respond to a user prompt. The researchers can leverage these simple mechanisms to see what the model knows about different subjects, and also possibly correct false information that it has stored.

 

Large language models, such as those that power popular artificial intelligence chatbots like ChatGPT, are incredibly complex. Even though these models are being used as tools in many areas, such as customer support, code generation, and language translation, scientists still don’t fully grasp how they work.

In an effort to better understand what is going on under the hood, researchers at MIT and elsewhere studied the mechanisms at work when these enormous machine-learning models retrieve stored knowledge.

They found a surprising result: Large language models (LLMs) often use a very simple linear function to recover and decode stored facts. Moreover, the model uses the same decoding function for similar types of facts. Linear functions, equations with only two variables and no exponents, capture the straightforward, straight-line relationship between two variables.

The researchers showed that, by identifying linear functions for different facts, they can probe the model to see what it knows about new subjects, and where within the model that knowledge is stored.

Using a technique they developed to estimate these simple functions, the researchers found that even when a model answers a prompt incorrectly, it has often stored the correct information. In the future, scientists could use such an approach to find and correct falsehoods inside the model, which could reduce a model’s tendency to sometimes give incorrect or nonsensical answers.

“Even though these models are really complicated, nonlinear functions that are trained on lots of data and are very hard to understand, there are sometimes really simple mechanisms working inside them. This is one instance of that,” says Evan Hernandez, an electrical engineering and computer science (EECS) graduate student and co-lead author of a paper detailing these findings.

Hernandez wrote the paper with co-lead author Arnab Sharma, a computer science graduate student at Northeastern University; his advisor, Jacob Andreas, an associate professor in EECS and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL); senior author David Bau, an assistant professor of computer science at Northeastern; and others at MIT, Harvard University, and the Israeli Institute of Technology. The research will be presented at the International Conference on Learning Representations.

Finding facts

Most large language models, also called transformer models, are neural networks. Loosely based on the human brain, neural networks contain billions of interconnected nodes, or neurons, that are grouped into many layers, and which encode and process data.

Much of the knowledge stored in a transformer can be represented as relations that connect subjects and objects. For instance, “Miles Davis plays the trumpet” is a relation that connects the subject, Miles Davis, to the object, trumpet.

As a transformer gains more knowledge, it stores additional facts about a certain subject across multiple layers. If a user asks about that subject, the model must decode the most relevant fact to respond to the query.

If someone prompts a transformer by saying “Miles Davis plays the. . .” the model should respond with “trumpet” and not “Illinois” (the state where Miles Davis was born).

“Somewhere in the network’s computation, there has to be a mechanism that goes and looks for the fact that Miles Davis plays the trumpet, and then pulls that information out and helps generate the next word. We wanted to understand what that mechanism was,” Hernandez says.

The researchers set up a series of experiments to probe LLMs, and found that, even though they are extremely complex, the models decode relational information using a simple linear function. Each function is specific to the type of fact being retrieved.

For example, the transformer would use one decoding function any time it wants to output the instrument a person plays and a different function each time it wants to output the state where a person was born.

The researchers developed a method to estimate these simple functions, and then computed functions for 47 different relations, such as “capital city of a country” and “lead singer of a band.”

While there could be an infinite number of possible relations, the researchers chose to study this specific subset because they are representative of the kinds of facts that can be written in this way.

They tested each function by changing the subject to see if it could recover the correct object information. For instance, the function for “capital city of a country” should retrieve Oslo if the subject is Norway and London if the subject is England.

Functions retrieved the correct information more than 60 percent of the time, showing that some information in a transformer is encoded and retrieved in this way.

“But not everything is linearly encoded. For some facts, even though the model knows them and will predict text that is consistent with these facts, we can’t find linear functions for them. This suggests that the model is doing something more intricate to store that information,” he says.

Visualizing a model’s knowledge

They also used the functions to determine what a model believes is true about different subjects.

In one experiment, they started with the prompt “Bill Bradley was a” and used the decoding functions for “plays sports” and “attended university” to see if the model knows that Sen. Bradley was a basketball player who attended Princeton.

“We can show that, even though the model may choose to focus on different information when it produces text, it does encode all that information,” Hernandez says.

They used this probing technique to produce what they call an “attribute lens,” a grid that visualizes where specific information about a particular relation is stored within the transformer’s many layers.

Attribute lenses can be generated automatically, providing a streamlined method to help researchers understand more about a model. This visualization tool could enable scientists and engineers to correct stored knowledge and help prevent an AI chatbot from giving false information.

In the future, Hernandez and his collaborators want to better understand what happens in cases where facts are not stored linearly. They would also like to run experiments with larger models, as well as study the precision of linear decoding functions.

“This is an exciting work that reveals a missing piece in our understanding of how large language models recall factual knowledge during inference. Previous work showed that LLMs build information-rich representations of given subjects, from which specific attributes are being extracted during inference. This work shows that the complex nonlinear computation of LLMs for attribute extraction can be well-approximated with a simple linear function,” says Mor Geva Pipek, an assistant professor in the School of Computer Science at Tel Aviv University, who was not involved with this work.

This research was supported, in part, by Open Philanthropy, the Israeli Science Foundation, and an Azrieli Foundation Early Career Faculty Fellowship."

#metaglossia_mundus

Rescooped by michel verstrepen from Planète DDurable
Scoop.it!

L’ère de l’Anthropocène a-t-elle commencé ? Les experts ont (enfin) tranché

L’ère de l’Anthropocène a-t-elle commencé ? Les experts ont (enfin) tranché | information analyst | Scoop.it
Après moult discussions et 15 années de délibérations, les scientifiques de l’IUGS, l'International Union of Geological Sciences, ont finalement rendu leur verdict.

Via Hubert MESSMER
No comment yet.
Rescooped by michel verstrepen from Media, Business & Tech
Scoop.it!

Why is AI so bad at spelling? Because image generators aren't actually reading text

Why is AI so bad at spelling? Because image generators aren't actually reading text | information analyst | Scoop.it
AIs are easily acing the SAT, defeating chess grandmasters and debugging code like it’s nothing. But put an AI up against some middle schoolers at the

Via Bruno Renkin
No comment yet.
Rescooped by michel verstrepen from Learning & Technology News
Scoop.it!

AI Thoughts - AI Misinformation

AI Thoughts - AI Misinformation | information analyst | Scoop.it

Four waves in computing have directly touched consumers and reshaped the world: desktop, the web, smartphone, and now AI. At this inflection point, it's imperative to contextualize facts and objectively analyze the costs and benefits of AI. Poor understanding produces poor policy, which limits progress and harms society.


Via Nik Peachey
Nik Peachey's curator insight, March 23, 3:45 AM

This is a very interesting read about AI misinformation - Well worth a few moments to read through to get a balanced view of the real and unreal threats from AI https://hotpot.ai/blog/ai-thoughts

Rescooped by michel verstrepen from Educational Technology News
Scoop.it!

LLMs Are More Than Generative AI

"While large language models (LLMs) and generative AI have been all the rage over the past year, the most attention has been given to their intersection — the text generation capabilities of LLMs."


Via EDTECH@UTRGV
EDTECH@UTRGV's curator insight, March 21, 12:38 PM

"There is no doubt that the ability to generate answers to questions is a major value proposition of LLMs. However, there are other uses of LLMs that are both common and valuable. This blog will discuss a few primary uses of LLMs to ensure that you don’t fall into the trap of considering them exclusively for generative purposes."

Rescooped by michel verstrepen from Educación a Distancia y TIC
Scoop.it!

Is ChatGPT making scientists hyper-productive? The highs and lows of using AI

Is ChatGPT making scientists hyper-productive? The highs and lows of using AI | information analyst | Scoop.it
Large language models are transforming scientific writing and publishing. But the productivity boost that these tools bring could have a downside.
Via LGA
No comment yet.
Rescooped by michel verstrepen from Media, Business & Tech
Scoop.it!

Databricks vs Snowflake : deux approches différentes, mais convergentes

Databricks vs Snowflake : deux approches différentes, mais convergentes | information analyst | Scoop.it
Les deux data platform ont tendance à voir leurs fonctionnalités se rapprocher. Elles n'en gardent pas moins des spécificités.

Via Bruno Renkin
No comment yet.
Scooped by michel verstrepen
Scoop.it!

Introducing Devin, the first AI software engineer

Introducing Devin, the first AI software engineer | information analyst | Scoop.it
With our advances in long-term reasoning and planning, Devin can plan and execute complex engineering tasks requiring thousands of decisions. Devin can recall relevant context at every step, learn over time, and fix mistakes.We've also equipped Devin with common developer tools including the...
michel verstrepen's insight:

Qui SARA ...SORA ;-)

No comment yet.
Rescooped by michel verstrepen from Time to Learn
Scoop.it!

Bypass AI Detection - Undetectable AI & AI Bypasser

Bypass AI Detection - Undetectable AI & AI Bypasser | information analyst | Scoop.it
Bypass AI detection effortlessly with our undetectable AI tool. We're an AI bypasser that makes AI text undetectable, allowing it to seamlessly get past AI detectors.

Via Frédéric DEBAILLEUL
No comment yet.
Rescooped by michel verstrepen from Veille Usages numériques
Scoop.it!

Guide de la sécurité des données personnelles : nouvelle édition 2024 | CNIL

Guide de la sécurité des données personnelles : nouvelle édition 2024 | CNIL | information analyst | Scoop.it

Une édition 2024 du guide de la @CNIL restructurée, avec de nouvelles fiches sur l’IntelligenceArtificielle, les applications mobiles, l’informatique en nuage, etc ...


Via Tice Doc Reims
No comment yet.
Rescooped by michel verstrepen from Veille Usages numériques
Scoop.it!

IA : les Français sont-ils capables d’identifier les deepfakes ?

IA : les Français sont-ils capables d’identifier les deepfakes ? | information analyst | Scoop.it

Les Français se sentent-ils capables de déterminer si une vidéo est réelle ou générée par une IA ? Sont-ils en mesure d'identifier les deepfakes ? Un sondage à retrouver sur le Blog Du Modérateur.


Via Tice Doc Reims
No comment yet.
Rescooped by michel verstrepen from Media, Business & Tech
Scoop.it!

Inside the Creation of DBRX, the World's Most Powerful Open Source AI Model

Inside the Creation of DBRX, the World's Most Powerful Open Source AI Model | information analyst | Scoop.it
Startup Databricks just released DBRX, the most powerful open source large language model yet—eclipsing Meta's Llama 2.

Via Bruno Renkin
No comment yet.
Scooped by michel verstrepen
Scoop.it!

Nvidia vient juste de révolutionner l'I.A ?

Nvidia vient d'annoncer les GPU Blackwell GB200 avec 30x les performances en inference comparé à la génération précédente.

Formations: https://cours.cocadmin.com
Newsletter: https://mailchi.mp/118f9876cefe/cocadmin
Rejoignez le discord: https://discord.gg/4h2xBFh
Twitter: https://twitter.com/cocadmin
Instagram: https://instagram.com/cocadmin
Tiktok: https://www.tiktok.com/@cocadmin
No comment yet.
Rescooped by michel verstrepen from Trucs et astuces du net
Scoop.it!

Top 50 des IA génératives à découvrir absolument en 2024

Top 50 des IA génératives à découvrir absolument en 2024 | information analyst | Scoop.it
Découvrez le top 50 des applications IA génératives les plus innovantes et populaires en 2024, classées selon leur trafic web et mobile. Des assistants généraux aux outils de productivité en passant par les compagnons IA, explorez les tendances clés et les entreprises émergentes dans le domaine de l'IA générative grand public.

Via Philou49
No comment yet.
Rescooped by michel verstrepen from Media, Business & Tech
Scoop.it!

iLobby: Apple, like the rest of big tech, wants to change the rules

iLobby: Apple, like the rest of big tech, wants to change the rules | information analyst | Scoop.it
Visual insights that go beyond the headlines, from Chartr's newsletter published on Mar 20, 2024. iLobby: Apple, like the rest of big tech, wants to change the rules.

Via Bruno Renkin
No comment yet.
Rescooped by michel verstrepen from Educación a Distancia y TIC
Scoop.it!

Beyond Burnout: AI as an Academic Ally in the “Publish or Perish” Culture

Beyond Burnout: AI as an Academic Ally in the “Publish or Perish” Culture | information analyst | Scoop.it
The potential for AI assistance to enhance traditional database searches is apparent, signifying a transformative shift in how researchers can retrieve information. 

Via Vladimir Kukharenko, LGA
No comment yet.
Rescooped by michel verstrepen from Veille CDI
Scoop.it!

Comment un « monologue intérieur » a considérablement amélioré les performances d'une IA

Comment un « monologue intérieur » a considérablement amélioré les performances d'une IA | information analyst | Scoop.it
Donner aux systèmes d'intelligence artificielle (IA) un "monologue intérieur" les rend beaucoup plus efficaces dans leur raisonnement.

Via cdi
No comment yet.
Rescooped by michel verstrepen from Educational Technology News
Scoop.it!

Generative AI and the Question of Open Source

"LLMs generally open up new frontiers for constructive as well as malicious uses."


Via EDTECH@UTRGV
EDTECH@UTRGV's curator insight, March 21, 12:34 PM

"Open release LLMs should not be given a free pass given the realities that they can, have been and will be used to perpetrate malicious use at a broader scale than closed source LLMs....

 

As much as we should be wary of premature, naive regulation that impedes the societal benefits of new technologies, so should we be wary of a hands off approach that assumes the marketplace will address all possibilities for societal harm and any regulation is bad."

Rescooped by michel verstrepen from Educación a Distancia y TIC
Scoop.it!

20 emerging technologies that will change the world 

This video explores 20 emerging technologies and their future. Watch this next video about the 10 stages of AI.


Via Edumorfosis, LGA
No comment yet.
Rescooped by michel verstrepen from Media, Business & Tech
Scoop.it!

Which AI should I use? Superpowers and the State of Play

Which AI should I use? Superpowers and the State of Play | information analyst | Scoop.it
And then there wer

Via Bruno Renkin
No comment yet.