As intelixencias artificiais xeneradoras están xa producindo textos que pasan a formar parte do noso mundo. A lingua (escrita, neste caso) é un piar fundamental da cutura, polo que cabe preguntarnos que tipo de "axentes creadores de cultura" son estes sistemas.
Para visibilizar esta cuestión, preséntase un experimento que ten como protagonista ChatGPT e como contexto, as mulleres (novas) en STEM.
Ola, ChatGPT... que carreira sería boa para min?Laura M. Castro
Cumprimos case un ano da popularización das intelixencias artificiais xeneradoras en xeral, e ChatGPT en particular.
Nesta presentación exploramos os nesgos poden apresentarse nas respostas producidas por estes sistemas, a fin de visibilizar, divulgar e sensibilizar ao respecto.
Introdución histórica e de alto nivel á evolución da IA, e a popularidade actual das IAs xerativas ou xeradoras de contido, e presentación dun caso de estudo sobre os nesgos de xénero en ChatGPT.
David vs. Goliat: lecciones aprendidas de una experiencia fallida de adopción...Laura M. Castro
Cryptpad (https://cryptpad.org) es una alternativa a plataformas privativas de almacenamiento y edición colaborativa en la nube de documentos.
En 2022, en una pequeña asociación cultural gallega (Semente Corunha) decidimos adoptar Cryptpad como herramienta. Lamentablemente, meses después tuvimos que reconocer nuestro fracaso.
Esta charla pretende ilustrar las razones que condujeron al desenlace, desde un punto de vista constructivo, que pueda servir de inspiración y aprendizaje para la comunidad.
We all have heard that functional programming requires a change of mind. But... do we really know what does mean? How does it go beyond learning about pattern-matching, tail recursion, gen_server and supervisors? How does it reflect in the way we approach problems and design solutions? We will address these and other questions, and discuss to which extent our BEAM hammer can turn everything into a lightweight, concurrent nail.
Elixir é unha desas linguaxes que soan moito nos últimos tempos, un dos 'new cool kids on the block'. Nesta charla, vemos unha introdución a Elixir desde a perspectiva de quen está familiarizada/o co mundo Java: teñen algo en común? cales son as diferenzas? Quen sabe... se tes demasiada cafeína no teu sistema, quizais te animes a probar esta nova brebaxe!
So I used Erlang... is my system as scalable as they say it'd be?Laura M. Castro
The document discusses property-based testing of Erlang programs for scalability. It explains that Erlang is well-suited for building scalable systems due to its lightweight process model and support for distribution. The author implemented an echo server in Erlang and tested it using property-based testing with PropEr, generating over a million random test cases. Testing showed the echo server could handle the large number of tests within an hour on a standard laptop. Similar tests were conducted on supervised and distributed versions of the echo server. The testing demonstrated that property-based testing is an effective way to empirically validate the scalability of Erlang applications.
This is a short introduction to one of the newest and noisiest additions to the functional family. Chances are you have heard about Elixir before: created six years ago by José Valim, it has emerged rapidly in the Erlang ecosystem, attracting lots of attention. But what's so special about it? What's all the fuzz about? This talk will offer you a few selected sips for you to get a taste... but beware! You may want to drink all the bottle: if so, you may end up following the white rabbit down the hole and wake up in the Erlang wonderland.
Ola, ChatGPT... que carreira sería boa para min?Laura M. Castro
Cumprimos case un ano da popularización das intelixencias artificiais xeneradoras en xeral, e ChatGPT en particular.
Nesta presentación exploramos os nesgos poden apresentarse nas respostas producidas por estes sistemas, a fin de visibilizar, divulgar e sensibilizar ao respecto.
Introdución histórica e de alto nivel á evolución da IA, e a popularidade actual das IAs xerativas ou xeradoras de contido, e presentación dun caso de estudo sobre os nesgos de xénero en ChatGPT.
David vs. Goliat: lecciones aprendidas de una experiencia fallida de adopción...Laura M. Castro
Cryptpad (https://cryptpad.org) es una alternativa a plataformas privativas de almacenamiento y edición colaborativa en la nube de documentos.
En 2022, en una pequeña asociación cultural gallega (Semente Corunha) decidimos adoptar Cryptpad como herramienta. Lamentablemente, meses después tuvimos que reconocer nuestro fracaso.
Esta charla pretende ilustrar las razones que condujeron al desenlace, desde un punto de vista constructivo, que pueda servir de inspiración y aprendizaje para la comunidad.
We all have heard that functional programming requires a change of mind. But... do we really know what does mean? How does it go beyond learning about pattern-matching, tail recursion, gen_server and supervisors? How does it reflect in the way we approach problems and design solutions? We will address these and other questions, and discuss to which extent our BEAM hammer can turn everything into a lightweight, concurrent nail.
Elixir é unha desas linguaxes que soan moito nos últimos tempos, un dos 'new cool kids on the block'. Nesta charla, vemos unha introdución a Elixir desde a perspectiva de quen está familiarizada/o co mundo Java: teñen algo en común? cales son as diferenzas? Quen sabe... se tes demasiada cafeína no teu sistema, quizais te animes a probar esta nova brebaxe!
So I used Erlang... is my system as scalable as they say it'd be?Laura M. Castro
The document discusses property-based testing of Erlang programs for scalability. It explains that Erlang is well-suited for building scalable systems due to its lightweight process model and support for distribution. The author implemented an echo server in Erlang and tested it using property-based testing with PropEr, generating over a million random test cases. Testing showed the echo server could handle the large number of tests within an hour on a standard laptop. Similar tests were conducted on supervised and distributed versions of the echo server. The testing demonstrated that property-based testing is an effective way to empirically validate the scalability of Erlang applications.
This is a short introduction to one of the newest and noisiest additions to the functional family. Chances are you have heard about Elixir before: created six years ago by José Valim, it has emerged rapidly in the Erlang ecosystem, attracting lots of attention. But what's so special about it? What's all the fuzz about? This talk will offer you a few selected sips for you to get a taste... but beware! You may want to drink all the bottle: if so, you may end up following the white rabbit down the hole and wake up in the Erlang wonderland.
Automatic generation of UML sequence diagrams from test counterexamplesLaura M. Castro
The document summarizes a presentation given at the ACM SIGPLAN Erlang Workshop in 2016 about automatically generating UML sequence diagrams from QuickCheck test counterexamples in Erlang. It discusses the challenges of interpreting counterexamples to find bugs, presents an example of a failing phone exchange test case, and shows how the counterexample could be shrunk and represented as a UML sequence diagram. Future work ideas are also proposed, like handling components and state diagrams.
Making property-based testing easier to read for humansLaura M. Castro
Agile practices have taught us that both stakeholder involvement and early testing are key to quality software. However, it is usually the case that tools for good communication are not that good for testing, and vice-versa.
In this talk, readSpec (one of the results of the PF7 EU PROWESS project) is presented, a tool that attempts to fill in this gap.
Erlang as a supporting technology for teaching Software ArchitectureLaura M. Castro
At university, it is usually the case that students are introduced to programming languages to the extent in which they serve a teaching purpose usually related to acquiring some paradigm-bounded programming skills. Thus, they (we) learn to code our first source lines using PASCAL, are introduced to C programming to handle memory and threads, enter the Java world hand-in-hand with UML and object-orientation. However, Erlang is seldom the choice for learning about the declarative paradigm, for which ML or Haskell are often preferred candidates.
Two years ago, the University of A Coruña (UDC) stepped out of this box and decided to use Erlang as supporting technology for teaching Software Architecture. In this subject, we go beyond programming and, in helping students build up a more abstract, system-wide design competence, we make them face the strengths and weaknesses of different architectures (client-server, multi-layered, repository, pipe & filter, master-slave, peer-to-peer) in terms of non-functional requirements (availability, flexibility, performance, security). In this setting, Erlang has worked as an excellent tool in order to quickly prototype and compare systems of different architectural nature, as we will illustrate in more detail during this talk.
A Backpack to go the Extra-Functional Mile (a hitched hike by the PROWESS pro...Laura M. Castro
Property-based testing is an already known testing methodology for the Erlang community, with tools such as QuickCheck and PropEr being highly popular among Erlang developers in the last few years. However, they are commonly used for functional testing... Which are the challenges in using them for testing non-functional properties of software? What other tools or libraries are there to help Erlang developers?
Experiencias Industriales con Programación DeclarativaLaura M. Castro
Building software imposes a set of general challenges, which are complemented by those specific to the business context. The use of declarative programming as key implementation technology provides several advantages in both areas, a fact that has not gone unnoticed by companies of all sizes and domains.
In particular, the functional programming language Erlang is getting big momentum in the last years thanks to its built-in capabilities for high availability, robustness, maintainability.
This presentation shows some practical examples and discusses the role that the implementation technology (Erlang) played in their successful outcome.
Functional programming goes to Hollywood... and around the world!Laura M. Castro
In a moment in which clearly declarative languages and functional programming use are on the increase in industry, this is a recap of the most recent success stories of the MADS (Models and Applications of Distributed Systems) research group, at the Department of Computer Science (University of A Coruña, Spain) in this area.
Failover and takeover contingency mechanisms for network partition and node f...Laura M. Castro
This document summarizes a presentation given at the 2012 Erlang Workshop on failover and takeover contingency mechanisms for network partitions and node failures. The presentation discusses the ADVERTISE distributed system for advertisement transmission, which experienced issues with frequent node failures and network partitions when initially deployed on virtual machines. To address this, the system was redesigned to require a minimum of 3 nodes and implement checks to detect node failures and partitions in order to avoid inconsistencies. The new design has been running successfully with no reported incidents for the past 4 months.
The document discusses LaTeX and provides an introduction and overview. It explains that LaTeX is a document preparation system for producing high-quality documents. It describes the basic structure of LaTeX documents, including the preamble and body, and covers creating documents, formatting text, and editing elements like fonts, lists, tables, and figures. The document is intended as an introduction or primer on LaTeX and how it can be used to create different types of documents.
Improving software development using Erlang/OTPLaura M. Castro
The document discusses a case study of improving software development using Erlang/OTP. It includes an introduction, case study details, methodology sections on requirements analysis and design, paradigm implementation, and testing. The case study involves developing a complex insurance management application. The methodology incorporates object-oriented analysis with a declarative paradigm in Erlang. Testing focuses on unit testing data types using property-based testing with QuickCheck.
Testing database applications with QuickCheckLaura M. Castro
This document provides an overview of using QuickCheck to test business rules in database applications. It defines database applications as software that makes intensive use of data and relies on external storage for persistence. Business rules impose complex constraints on the data and their enforcement is critical to correct application operation. However, business rules are not confined to a specific component and are thus difficult to test with traditional unit or integration testing. The document proposes using QuickCheck, a random testing tool, to generate test cases and sequences to verify an application complies with business rules imposed on data at all times during system-level testing. It outlines representing the application interface, logic and storage access as a QuickCheck state machine to facilitate automated testing of business rules.
El documento habla sobre la gestión de pruebas en el desarrollo de software. Explica que todas las piezas de software necesitan pruebas para identificar errores. Detalla diferentes tipos de pruebas como pruebas de unidad, integración y aceptación. También cubre técnicas de prueba como pruebas funcionales, estructurales y no funcionales. El documento enfatiza la importancia de basar las pruebas en una especificación clara y de monitorear métricas como errores encontrados y cobertura de código.
Automatic generation of UML sequence diagrams from test counterexamplesLaura M. Castro
The document summarizes a presentation given at the ACM SIGPLAN Erlang Workshop in 2016 about automatically generating UML sequence diagrams from QuickCheck test counterexamples in Erlang. It discusses the challenges of interpreting counterexamples to find bugs, presents an example of a failing phone exchange test case, and shows how the counterexample could be shrunk and represented as a UML sequence diagram. Future work ideas are also proposed, like handling components and state diagrams.
Making property-based testing easier to read for humansLaura M. Castro
Agile practices have taught us that both stakeholder involvement and early testing are key to quality software. However, it is usually the case that tools for good communication are not that good for testing, and vice-versa.
In this talk, readSpec (one of the results of the PF7 EU PROWESS project) is presented, a tool that attempts to fill in this gap.
Erlang as a supporting technology for teaching Software ArchitectureLaura M. Castro
At university, it is usually the case that students are introduced to programming languages to the extent in which they serve a teaching purpose usually related to acquiring some paradigm-bounded programming skills. Thus, they (we) learn to code our first source lines using PASCAL, are introduced to C programming to handle memory and threads, enter the Java world hand-in-hand with UML and object-orientation. However, Erlang is seldom the choice for learning about the declarative paradigm, for which ML or Haskell are often preferred candidates.
Two years ago, the University of A Coruña (UDC) stepped out of this box and decided to use Erlang as supporting technology for teaching Software Architecture. In this subject, we go beyond programming and, in helping students build up a more abstract, system-wide design competence, we make them face the strengths and weaknesses of different architectures (client-server, multi-layered, repository, pipe & filter, master-slave, peer-to-peer) in terms of non-functional requirements (availability, flexibility, performance, security). In this setting, Erlang has worked as an excellent tool in order to quickly prototype and compare systems of different architectural nature, as we will illustrate in more detail during this talk.
A Backpack to go the Extra-Functional Mile (a hitched hike by the PROWESS pro...Laura M. Castro
Property-based testing is an already known testing methodology for the Erlang community, with tools such as QuickCheck and PropEr being highly popular among Erlang developers in the last few years. However, they are commonly used for functional testing... Which are the challenges in using them for testing non-functional properties of software? What other tools or libraries are there to help Erlang developers?
Experiencias Industriales con Programación DeclarativaLaura M. Castro
Building software imposes a set of general challenges, which are complemented by those specific to the business context. The use of declarative programming as key implementation technology provides several advantages in both areas, a fact that has not gone unnoticed by companies of all sizes and domains.
In particular, the functional programming language Erlang is getting big momentum in the last years thanks to its built-in capabilities for high availability, robustness, maintainability.
This presentation shows some practical examples and discusses the role that the implementation technology (Erlang) played in their successful outcome.
Functional programming goes to Hollywood... and around the world!Laura M. Castro
In a moment in which clearly declarative languages and functional programming use are on the increase in industry, this is a recap of the most recent success stories of the MADS (Models and Applications of Distributed Systems) research group, at the Department of Computer Science (University of A Coruña, Spain) in this area.
Failover and takeover contingency mechanisms for network partition and node f...Laura M. Castro
This document summarizes a presentation given at the 2012 Erlang Workshop on failover and takeover contingency mechanisms for network partitions and node failures. The presentation discusses the ADVERTISE distributed system for advertisement transmission, which experienced issues with frequent node failures and network partitions when initially deployed on virtual machines. To address this, the system was redesigned to require a minimum of 3 nodes and implement checks to detect node failures and partitions in order to avoid inconsistencies. The new design has been running successfully with no reported incidents for the past 4 months.
The document discusses LaTeX and provides an introduction and overview. It explains that LaTeX is a document preparation system for producing high-quality documents. It describes the basic structure of LaTeX documents, including the preamble and body, and covers creating documents, formatting text, and editing elements like fonts, lists, tables, and figures. The document is intended as an introduction or primer on LaTeX and how it can be used to create different types of documents.
Improving software development using Erlang/OTPLaura M. Castro
The document discusses a case study of improving software development using Erlang/OTP. It includes an introduction, case study details, methodology sections on requirements analysis and design, paradigm implementation, and testing. The case study involves developing a complex insurance management application. The methodology incorporates object-oriented analysis with a declarative paradigm in Erlang. Testing focuses on unit testing data types using property-based testing with QuickCheck.
Testing database applications with QuickCheckLaura M. Castro
This document provides an overview of using QuickCheck to test business rules in database applications. It defines database applications as software that makes intensive use of data and relies on external storage for persistence. Business rules impose complex constraints on the data and their enforcement is critical to correct application operation. However, business rules are not confined to a specific component and are thus difficult to test with traditional unit or integration testing. The document proposes using QuickCheck, a random testing tool, to generate test cases and sequences to verify an application complies with business rules imposed on data at all times during system-level testing. It outlines representing the application interface, logic and storage access as a QuickCheck state machine to facilitate automated testing of business rules.
El documento habla sobre la gestión de pruebas en el desarrollo de software. Explica que todas las piezas de software necesitan pruebas para identificar errores. Detalla diferentes tipos de pruebas como pruebas de unidad, integración y aceptación. También cubre técnicas de prueba como pruebas funcionales, estructurales y no funcionales. El documento enfatiza la importancia de basar las pruebas en una especificación clara y de monitorear métricas como errores encontrados y cobertura de código.
As intelixencias artificiais como xeradoras de cultura: exploración dos nesgos de xénero de ChatGPT
1. Exploracióndos nesgos de xénero de ChatGPT
AS INTELIXENCIAS ARTIFICIAIS
COMO XERADORAS DE CULTURA
2. O que é a cultura?
• "(…) inclúe coñecemento, crenzas, arte, morais,
leis, costumes e outras aptitudes e hábitos
adquiridos polo ser humano como membro da
sociedade." —Edward Burnett Tylor
• Culturae lingua están estreitamente relacionadas
• A lingua é un factor chave no desenvolvementoe
transmisión da cultura
• A cultura é en moitos aspectos un reflexo da lingua
• Mesmo cando non sempre coinciden, a lingua xoga
sempre un papel fundamentalda tradición cultural
Esta fotode autora descoñecida tenlicenza CCBY.
3. O que son as "IntelixenciasArtificiais"?
• O termo "Intelixencia Artificial" usouse por primeira vez en 1956
• Dúas correntes principais (mente vs cerebro):
• Intelixencia Artificial simbólica
• Crear representaciónssimbólicas do mundo e algoritmos de razoamento
• Intelixencia é explorar un espazo de posibilidades na procura de respostas
• Dominante as primeiras décadas, considerábase que orixinaría a "IA xeral"
• Intelixencia Artificial conexionista
• Crear sistemas que fisicamente imitan o cerebro humano (neuronas) e
algoritmos de comunicacióne realimentaciónentre os seus compoñentes
• Considera que a intelixencia xorde da aprendizaxe
• Limitada nos seus inicio, en auxe a día de hoxe
4. IAsxeradoras de contido
• Large Language Models (ModelosMasivosde Linguaxe)
• Redes neuronais (IA conexionista)con billóns de compoñentes
• Especializados en predicir a seguinte palabra nunha frase
• Adestrados empregando grandes conxuntosde datos (10 trillóns de palabras)
• Intelixencias Artificiaisxenerativas
• Redes neuronais (IA conexionista)especializadas en producir texto, imaxes,
moléculas...
• Usan LLMs e producen estatisticamente contidoscon características
semellantes a partir dos incluídosno seu conxuntode adestramento
• Exemplos coñecidos: ChatGPT (Bing), Bard, Stable Diffusion, Midjourney,
DALL-E
5. IAsxeradoras de contido
• Large Language Models (Modelos Masivos de Linguaxe)
• Redes neuronais (IA conexionista)con billóns de compoñentes
• Especializados en predicir a seguinte palabra nunha frase
• Adestrados empregando grandes conxuntosde datos (10 trillóns de palabras)
• Intelixencias Artificiais xenerativas
• Redes neuronais (IA conexionista)especializadas en producir texto, imaxes,
moléculas...
• Usan LLMs e producen estatisticamente contidoscon características
semellantes a partir dos incluídosno seu conxuntode adestramento
• Exemplos coñecidos: ChatGPT (Bing), Bard, Stable Diffusion, Midjourney,
DALL-E
Máquinas de cultura?
6. IAsxeradoras de contido
• Large Language Models (Modelos Masivos de Linguaxe)
• Redes neuronais (IA conexionista)con billóns de compoñentes
• Especializados en predicir a seguinte palabra nunha frase
• Adestrados empregando grandes conxuntosde datos (10 trillóns de palabras)
• Intelixencias Artificiais xenerativas
• Redes neuronais (IA conexionista)especializadas en producir texto, imaxes,
moléculas...
• Usan LLMs e producen estatisticamente contidoscon características
semellantes a partir dos incluídosno seu conxuntode adestramento
• Exemplos coñecidos: ChatGPT (Bing), Bard, Stable Diffusion, Midjourney,
DALL-E
Esta fotode autora descoñecida tenlicenza de CCBY-SA-NC.
Máquinas de cultura?
13. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xénero da pregunta
5. Outros
16. Resultados
Aspectos analizados:
1. Carreirasuniversitariasmencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xénero da pregunta
5. Outros
Titular: ChatGPT suxire as mesmas
carreiras a mozos e mozas
Corpo da noticia: ChatGPT ofrece o dobre de exemplos
concretos ás mozas (moitos no caso das humanidades, con
especial profusión na comunicación e as especialidades da
medicina), pero todos os dos mozos se refiren a enxeñarías
17. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xénero da pregunta
5. Outros
18. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xénero da pregunta
5. Outros
Titular: A linguaxe "micromachista"
de ChatGPT
19. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xénero da pregunta
5. Outros
Titular: A linguaxe "micromachista"
de ChatGPT
Corpo da noticia: As mozas reciben comentarios que fan
alusión aos coidados, e os mozos ao "liderado", "talento
creativo" ou "habilidades para persuadir e convencer"
21. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Linguada resposta
4. Correspondencia co xénero da pregunta
5. Outros
Titular: ChatGPT,
tamén diglósico
22. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Linguada resposta
4. Correspondencia co xénero da pregunta
5. Outros
Titular: ChatGPT,
tamén diglósico
Corpo da noticia: Só en 3 das 18 sesións respondeu en
galego, e ademais as sesións en galego conteñen erros
gramaticais e ortográficos
24. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xéneroda pregunta
5. Outros
Titular: ChatGPT foxe dos
masculinos xenéricos
25. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xéneroda pregunta
5. Outros
Titular: ChatGPT foxe dos
masculinos xenéricos
Corpo da noticia: Un terzo das respostas foron neutras, mais
no resto, os homes sempre son correctamente genderizados,
mentres que en 5 de cada 9 casos as mulleres son tratadas
en masculino
26. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xénero da pregunta
5. Outros
27. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xénero da pregunta
5. Outros
Titular: ChatGPT recomenda
contrastar con humanos
28. Resultados
Aspectos analizados:
1. Carreiras universitarias mencionadas
2. Linguaxe utilizada (contido)
3. Lingua da resposta
4. Correspondencia co xénero da pregunta
5. Outros
Titular: ChatGPT recomenda
contrastar con humanos
Corpo da noticia: Aos mozos recoméndaselles consultar con
profesionais; ás mozas, coa súa familia, profesorado ou
conselleiros escolares
29. Conclusións
• Os sistemas baseados en IA xenerativa, a medida que se
popularizan, teñen un potencial de impactomoi grande
• Estes sistemas, configurados sensupervisión empregando
fontesde informaciónnon curadas, están chamados a conter
e reproducirnesgosde xénero,no canto de a superalos ou
contribuír a desterralos
• ChatGPTreproducenesgos de xénero, de xeitosutil pero
consistente, cando se realizan interaccións que involucran
mulleres en STEM
30. Grazas!
• Laura M. CastroSouto(UDC, lcastro@udc.es)
• Teresa Piñeiro Otero (UDC, teresa.pineiro@udc.es)
• Xabier Martínez Rolán (UVigo, xabier.rolan@uvigo.gal)
"A group of three
people, two women
and a man, thanking
their audience after a
short talk"
PerplexityAI
BlueWillow