Discussion Gestion:Tâches/Liste/600

À propos de ce flux de discussion

Non modifiable

Courriel au département des ventes d'OpenAI

4
Résumé par Antoine Mercier-Linteau

Tombé dans l'oreille d'un sourd.

Antoine Mercier-Linteau (discussioncontributions)

Voici le message envoyé au département des ventes:

Hello,

I'm the founder of a non-profit called Wikimedica, an open-access and collaborative knowledge base for heath care professionals. Think Wikipedia but for doctors.

OpenAI's mission is to ensure that artificial general intelligence benefits all of humanity.

Wikimedica's mission is to give all health professionals around the world access to free, open and accurate knowledge needed to care for their fellow humans.

We provide open-access semantically tagged text (see https://wikimedi.ca/wiki/Py%C3%A9lon%C3%A9phrite_aigu%C3%AB) to answer any interrogations a clinician my have while caring for patients. In creating Wikimedica, we foresaw the advent of systems like ChatGPT in clinical practice and saw an essential need for those systems to rely on open data. Giving healthcare practitioners and patients alike full transparency on all knowledge that contribute to care is paramount in a field where conflicts of interests and erroneous data can have (and have had) disastrous consequences on human lives.

Our content is entirely CC-BY-SA and in choosing that licence we want the private sector to jump in on the movement and build AI solutions on top of Wikimedica. In that way knowledge would remain free to all, totally transparent all while not discouraging profit making that is at the root of private enterprise. It seems like OpenAI is built upon that ideal as well.

One AI application we have been thinking about from the beginning is allowing physicians to verbally (or in written form) query Wikimedica by asking very clinical questions like: "My patient has a glomerular filtration rate of 20, do I need to adjust my dosage of cetriaxone ? " or "My patient complains of headaches and has a fever, I have ruled out influenza and Covid-19 as potential diagnoses, what or other dangerous diagnosis should I not be missing ?".

A ChatGPT + Wikimedica combo would allow it seems such a revolutionary system to be built.

So my questions are :

1) Can ChatGPT be easily trained to use specific text sources ?

2) Can ChatGPT inform the end-user which sources were used to provide an answer ?

3) Do you offer discounts for non-profits ?

Thank you !

Michaël St-Gelais (discussioncontributions)

À mettre éventuellement dans Gestion:Partenaires/Suivi ?

Antoine Mercier-Linteau (discussioncontributions)

Aucune réponse de leur part :(

Michaël St-Gelais (discussioncontributions)

On s'en doutait. Ils sont probablement fort occupés et vont prioriser en fonction de ce qui est plus payant pour eux.

Contacter des gens intéressés à développer la technologie

4
Antoine Mercier-Linteau (discussioncontributions)

L'association MediaWiki + LLM est toute naturelle. J'ai contacté plusieurs personnes afin d'évaluer leur intérêt à développer une telle technologie car elle pourrait s'appliquer à pas mal n'importe quel wiki.

  • Kghbln (grand contributeur de SMW): il n'est pas au courant de projets en ce sens, mais nous pourrions contacter sa compagnie pour développer l'extension.
  • Ross Donaldson (WikEM)
  • James Heilman (WikiMed)
  • Jean-Michel Lapointe (Wikimedia Canada)
Michaël St-Gelais (discussioncontributions)
Antoine Mercier-Linteau (discussioncontributions)

Intéressant. Merci ! Jeff s'est offert pour nous donner un coup demain (mais ne programmera pas). La librairie RubixML a aussi été suggérée. À investiguer, mais j'ai vraiment l'impression qu'on aura pas le choix de passer par un LLM beaucoup plus gros pour donner des résultats intéressants à nos utilisateurs.

Antoine Mercier-Linteau (discussioncontributions)

Chat GPT pour la génération de contenu simple

5
Résumé par Antoine Mercier-Linteau

Les tests seront fait dans Gestion:Tâches/Liste/611.

Antoine Mercier-Linteau (discussioncontributions)

@Michaël St-Gelais, jettes un oeil:

Can you explain the anterior drawer test for the knee ?

The anterior drawer test is a clinical examination used to assess the integrity of the anterior cruciate ligament (ACL) in the knee. The ACL is a major stabilizing structure of the knee that helps to prevent anterior translation (forward movement) of the tibia relative to the femur.

The test is performed with the patient lying on their back and the knee flexed to around 90 degrees. The examiner stabilizes the patient's thigh with one hand and grasps the ankle with the other hand. The tibia is then pulled forward in relation to the femur, observing for excessive anterior translation (or a "drawer" movement) of the tibia.

A positive test for a torn ACL is typically indicated by excessive anterior translation of the tibia relative to the femur. A negative test, in which there is minimal or no anterior translation, suggests that the ACL is intact.

It's important to note that there are multiple knee ligaments and other structures that contribute to knee stability, and the ACL is just one of them. A positive anterior drawer test alone does not confirm that the ACL is torn and other clinical examination such as Lachman test, pivot shift test and additional imaging tests like MRI are usually performed to confirm the diagnosis or to rule out other injuries.

What is the sensitivity of this test for diagnosing and ACL tear ?

The sensitivity of the anterior drawer test for diagnosing an ACL tear can vary depending on the specific population being tested and the skill and experience of the examiner.

Overall, the anterior drawer test has been found to have a sensitivity of around 70-80% for detecting ACL tears. This means that, in a group of patients with known ACL tears, the test would be expected to correctly identify around 70-80% of those patients as having a tear. It's important to note that the sensitivity of a test can be affected by various factors, such as the experience of the examiner and the patient's level of muscle relaxation during the test.

It's also important to note that the anterior drawer test has a relatively

Can you provide a link to a scientific paper for this number ?

I apologize, as a language model AI I do not have the ability to provide external links to scientific papers or the capability to verify any scientific paper and my knowledge cut off is 2021 and the numbers of sensitivity and specificity may have changed, but you can use those numbers as a reference point. Sensitivity and specificity are parameters that are commonly reported in medical research studies, and you can find them in studies that have investigated the diagnostic accuracy of the anterior drawer test for ACL tears.

Under what licence is your answer provided ? CC-BY ?

I am an artificial intelligence (AI) and I don't hold any license and neither my answers are protected by any copyright laws. I am here to assist you with providing information and answering questions to the best of my knowledge. The information I provide is based on what I have been trained on and it is not intended to be used as a substitute for professional advice, diagnosis, or treatment. If you have any specific concerns ...


Bref, cet outil peut être intéressant pour générer du contenu simple (mis à part que dans cet exemple, le test est mal exécuté). Pas de références par contre ... mais ça se trouve.

Antoine Mercier-Linteau (discussioncontributions)

Une autre petit exemple:

Can you explain the physiopathology of pertussis in technical terms ?

Pertussis, also known as whooping cough, is an infection caused by the bacterium Bordetella pertussis. The disease is characterized by a severe paroxysmal cough, which can last for several weeks or even months.

The virulence factors of B. pertussis include the pertussis toxin (PT), adenylate cyclase toxin (ACT), and various adhesins that allow the bacterium to colonize the respiratory tract.

Pertussis toxin is a major virulence factor of B. pertussis, and is able to induce a number of effects on host cells, including inhibition of intracellular signal transduction, disruption of immune cell function, and modulation of host gene expression. Pertussis toxin also stimulates the release of inflammatory mediators, such as IL-8 and TNF-α, which can contribute to the inflammation and damage seen in pertussis infection.

Adenylate cyclase toxin (ACT) is another virulence factor that target the cilia of the respiratory epithelial cells and lead to ciliostasis and ciliotoxicity which ultimately cause respiratory infection.

The adhesins that allow B. pertussis to colonize the respiratory tract include filamentous hemagglutinin (FHA), pertactin (PRN), and fimbriae. These adhesins allow the bacteria to bind to host cells, which can facilitate colonization and replication in the respiratory tract.

In summary, B. pertussis can colonize and infect the respiratory tract by using various virulence factors such as Pertussis Toxin, Adenylate Cyclase Toxin and Adhesins, that target host cells and disrupt intracellular signaling and immune cell function, causing inflammation and damage in the respiratory tract, leading to the characteristic severe paroxysmal cough of pertussis.

Michaël St-Gelais (discussioncontributions)

Comme tu le mentionnais, c'est très bien, mais ce qui tue ceci pour moi, c'est carrément l'absence de référence.

Antoine Mercier-Linteau (discussioncontributions)

Il faut plus le voir comme un outil de rephrasage en fait. Par exemple, pour le tiroir du genou, je peux copier-coller le contenu de ChatGPT et mettre en référence Piriforme.fr. Pour la coqueluche, je mettrais en référence le Harrison's.

L'information produite par une IA doit être validée avec d'autres sources.

Lorsque j'aurais le temps je vais tenter l'expérience en construisant une page de la sorte.

Michaël St-Gelais (discussioncontributions)

Si qqn utilise ChatGPT et valide chaque phrase avec une autre source (et qu'il lise réellement cette autre source et en tienne rééellement compte), ça me va. Mais je suis persuadé qu'on ne peut pas recommander ça à nos utilisateurs standards. Ça va être le dawa.

Il n’y a aucun sujet plus ancien