Search
1000 results for “context”
-
context: @poemproducer 🐋 #reconsound #feminisistsonictechnologies #audiocollectifness #pluralistism
talking unscripted and slow about rec-on.org for 5 hours for #radioartzone 2022▶️ https://poemproducer.com/AGaudiodrop/9-14.5_Pluralistism_reconORG.mp3
-
Context: In front of a PIN pad they’ve forgotten the code for.
“Last time, it was the year Freddie Mercury died, wasn’t it?” Ryan nodded. “Before then, it was Whitney Houston,” he muttered. “You might have guessed, he likes to commemorate the deaths of his favourite musicians.”
They need a number. They mention “the year [X] died” (“Maybe famous deaths?”), give us a second mention (“Famous singer deaths!”), and now we can mentally join the dots without the bold text.
-
🧑💻 #Contextualisation / #Formation
✅ La formation « Contextualiser les œuvres avec le numérique », proposée au PAF de l’académie de Versailles, est ouverte aux préinscriptions jusqu’au 10.11.2025.
Au programme, réflexion sur les stratégies de contextualisation en collège et en lycée, avec apport d’outils numériques pertinents dans différentes perspectives pédagogiques
-
[Перевод] Переключение контекста — главный убийца продуктивности разработчика
Новый перевод от команды Spring АйО расскажет вам о том, почему так вредно отвлекать разработчиков от их работы и как избежать большого убытка для компании из-за прерывания рабочего процесса сотрудников.
https://habr.com/ru/companies/spring_aio/articles/890146/
#flow_state #burnout #time_management #productivity #contextswitching
-
[Перевод] Переключение контекста — главный убийца продуктивности разработчика
Новый перевод от команды Spring АйО расскажет вам о том, почему так вредно отвлекать разработчиков от их работы и как избежать большого убытка для компании из-за прерывания рабочего процесса сотрудников.
https://habr.com/ru/companies/spring_aio/articles/890146/
#flow_state #burnout #time_management #productivity #contextswitching
-
The Long Context
In "You Exist In The Long Context," Steven Johnson explores the advancements in large language models (LLMs), particularly the significant impact of long context windows. Johnson illustrates this progress by creating an interactive game based on his book, showcasing the LLM's ability to handle complex narratives and maintain factual accuracy. He draws a parallel between LLMs' short-term memory improvements and the case of Henry Molaison, a patient with severe memory impairment, highlighting how expanded context windows have overcome previous limitations. He ultimately argues that this enhanced contextual understanding allows for more sophisticated applications, including personalised learning and collaborative decision-making. Johnson concludes by discussing the potential for LLMs to become invaluable tools for accessing and integrating expert knowledge.
Limitations of Early Language Models like GPT-3
Early language models like GPT-3, while impressive for their time, exhibited a significant limitation: a limited context window. This meant they had a restricted short-term memory, analogous to the condition of patient H.M., who was unable to form new memories after a specific brain surgery.
GPT-3, introduced in 2019, had a context window of just over 2,000 “tokens”, equivalent to about 1,500 words. This was the maximum amount of new information that could be shared with the model. Exceeding this limit caused the model to "forget" information presented earlier in the conversation. It could follow short instructions based on its vast long-term memory (parametric memory) but struggled with extended narratives or explanations requiring the retention of information over a longer stretch of text. Essentially, interacting with GPT-3 was like having a conversation with someone who had to constantly be reintroduced to the topic because they couldn't retain information beyond a few sentences.
This limited context window resulted in several shortcomings:
- Conversational Incoherence:The inability to remember previous turns in a conversation made interactions with GPT-3 feel disjointed and repetitive. Users had to repeatedly provide context, leading to an unnatural flow.
- Increased Hallucinations: While GPT-3 possessed a vast knowledge base, its limited short-term memory made it prone to fabricating information, especially when the required information was not part of the immediate context.
- Inability to Handle Complex Narratives or Arguments: GPT-3 struggled to follow narratives or arguments that spanned beyond its limited context window. Understanding relationships between events and concepts spread across a large text was impossible, limiting its analytical capabilities.
The subsequent expansion of context windows in models like ChatGPT (which boasts an 8K context window, four times larger than GPT-3) marked a significant advancement in AI capabilities. These larger context windows facilitated more coherent conversations, reduced hallucinations, and allowed for a deeper understanding of complex narratives. However, it's essential to note that even with these advancements, AI models still do not possess human-like consciousness or sentience.
Impacts of Expanding AI Context Windows
The expansion of AI context windows has been a pivotal factor in the advancements of AI capabilities, going beyond simply increasing the size of training data or model parameters. This expansion has led to significant improvements across various aspects of AI functionality:
- Document Summarization and Processing: One prominent application is the processing of extensive documents or text corpora. With larger context windows, LLMs can maintain the coherence and relevance of the generated summary over longer texts. This is particularly beneficial for legal documents, research papers, and books, where context from the entire document is crucial for generating accurate summaries
- Improved Conversational Agents: In the realm of chatbots and conversational agents, long context windows enable the model to maintain the context of the conversation over extended interactions. This means the AI can refer back to previous parts of the dialogue, providing more coherent and contextually relevant responses, leading to more sophisticated and human-like interactions.
- Code Generation and Understanding: For developers using LLMs to assist in code generation, debugging, or understanding, larger context windows allow the model to consider more lines of code at once. This can improve the quality of the generated code and the accuracy of suggestions, as the model can better understand the overall structure and dependencies within the code.
- Historical Data Analysis: In applications involving historical data, such as financial market analysis or historical research, long context windows enable the model to consider larger sequences of events. This can lead to more accurate predictions and insights, as the model can identify patterns and trends over more extended periods (Source [4]).
- Complex Query Processing: When dealing with complex queries that require understanding multiple pieces of information from different parts of a large dataset, extended context windows can significantly enhance the model’s ability to retrieve and synthesize relevant information, providing more accurate and comprehensive responses (Source [9]).
- Creative Writing and Content Generation: For tasks like story writing or content creation, where maintaining narrative coherence and consistency is vital, long context windows allow the model to track character development, plot points, and thematic elements over longer passages of text. This results in more cohesive and engaging content.
Long Context Windows vs. RAG
The advancements in long context windows have sparked a debate on the necessity of techniques like Retrieval Augmented Generation (RAG). While long context windows allow models to process and utilize vast amounts of context directly, RAG combines the retrieval of relevant information from external sources with the generative capabilities of LLMs. Here are some key applications and advantages of RAG:
- Real-Time Information Retrieval: One of the primary advantages of RAG is its ability to retrieve up-to-date information from external databases or documents, ensuring that the generated content is current and accurate. Traditional language models, even with large context windows, rely heavily on their pre-existing training data, which can become outdated over time. RAG addresses this by accessing real-world data as needed, enhancing the model’s ability to answer complex and timely questions effectively.
- Enhanced Enterprise AI Capabilities: RAG's ability to access specific, relevant external data enhances the model’s precision and utility. This combination is crucial for various enterprise applications, such as legal document analysis, financial reporting, and customer support, where accuracy and relevancy are paramount.
- Augmented Retrieval and Agent Capabilities: RAG is particularly useful in applications where detailed and context-specific information retrieval is necessary. For example, in customer support systems, RAG can retrieve specific answers from a company’s knowledge base, providing more precise and contextually appropriate responses to user queries. This contrasts with long context window models that might struggle to identify the most relevant information from a vast pool of data.
The choice between long context windows and RAG significantly influences the overall performance of deep learning models in various real-world applications. RAG is significantly more scalable and cost-effective than long context windows because it only retrieves and processes the most relevant pieces of information, reducing the number of tokens that need to be processed. This approach minimizes computational costs and latency, making it suitable for high-volume queries and real-time applications.
Summary
In summary, long context windows improve LLM performance by allowing the model to process and retain more internal context without external retrieval. In contrast, RAG is an algorithmic retrieval technique that enhances LLMs by fetching relevant information from external sources. While long context windows cannot replicate the exact functionality of RAG, they can be used in conjunction with RAG to create a more powerful system. This combination allows the model to leverage the strengths of both approaches: the ability to process extensive internal context and the efficiency of selective external information retrieval.
Unlock the Future of Business with AI
Dive into our immersive workshops and equip your team with the tools and knowledge to lead in the AI era.
Get in touch with us -
Context matters. Ministerin #Reiche berät Reiche, wie diese Steuersenkungen erREICHen. Während Ärmere mit härteren Sanktionen belegt werden, die gegen die Regeln der Jobcenter verstoßen. Das ist #Klassenkampf von oben! Diese #AbschaumElite wurde gewählt! #Politik www.tagesschau.de/inland/innen...
RE: https://bsky.app/profile/did:plc:foc4lht37ziujsrhbz6jhyk7/post/3m2u7a3up4c2j
Die Regeln beim Bürgergeld sol... -
Context.ro: Vedenia de la chilia sponsorizată de #Becali – Cine a văzut “fiara mare” din spatele lui #NicușorDan, cu două zile înainte de vot.
-
CW: The Cosmic Zoom #AmReading thread
p.173:
"The parameterization of data collection (what we might call its thickness or depth) may expand, but parameters must pass through disciplinary optics and become operationalized before they are articulated in
any way."Or, data don't "speak for themselves", but only gain their meaning when incorporated into disciplinary practices.
The definition I learned as undergrad: "Data are numbers in context." #ContextChangesEverything
-
context always helps :awesome: #cactus #cactiofMastodon
-
Context: Cine s-a ocupat de operațiunea “🇺🇸#America susține Călin #Georgescu președinte”.
Un 🇺🇸#American care locuiește în 🇷🇴#România și se laudă cu relații în 🇷🇺#GuvernulRusiei, un influencer crypto arestat pentru înșelăciune și un furnizor al STS (📡#ServiciulDeTelecomunicațiiSpeciale) au manipulat opinia publică. -
switch_to_blog is a handy function to switch sites (“blogs”) in a multisite. However, beware that you might have problems with certain data in such context switches.
[…]
https://epiph.yt/en/blog/2025/beware-when-using-switch_to_blog/
-
Contexte financier général de l’intervention culturelle des collectivités https://www-cd.org/contexte-financier-general-de-lintervention-culturelle-des-collectivites/ #Veilleinformative
-
Contexte politique, syndical et technologique de l’implantation d’usines de fabrication de puces en Arizona par l’entreprise chinoise TSMC
https://arstechnica.com/tech-policy/2024/01/bidens-plan-to-dominate-chips-market-may-face-delays-downgrades-at-tsmc-fabs/
#TSMC #Chine #USA #puces #protectionisme #économie #commerce #technologie -
Contexte géomorphologique du site archéologique de Chuchuwayha (Colombie Britannique, Canada). Apports de l’analyse sédimentologique
▶️ L’objectif est ici d’identifier les signatures sédimentaires des différentes formations superficielles environnant le site […] et de les discuter par rapport aux grandes phases climatiques et aux périodes de fréquentation humaine.
https://journals.openedition.org/geomorphologie/17496#archéologie #Autochtones #Canada #géomorphologie #géologie #climat #sédiments
-
Heute im #Sonntagssketchen war das Thema „Sags mit Tieren“. Da mir der Talk von Gilles Demarty am World IA Day Zürich so gut gefallen hat, habe ich eine weitere #Sketchnote davon angefertigt und die präsentierten Prinzipien aus dem Improvisieren mit Tieren illustriert. #WIAD #WIAD24 #WorldIADay #Context
-
Kontext war das Überthema des gestrigen #WIAD24 Der Kontext gibt der Information erst die Bedeutung. #WIAD #WorldIADay #Context
-
# Context Engineering: Nâng Cấp AI Coding Agents Với DSPy GEPA
Bài viết mới về kỹ thuật Context Engineering giúp cải thiện hiệu suất AI Coding Agents bằng phương pháp DSPy GEPA. Một hướng tiếp cận thú vị để tối ưu hóa chất lượng code AI.
#AI #MachineLearning #AIcoding #DSPy #GEPA #KỹthuậtAI #PháttriệnAI
-
(context: #ThreePines, Season 1, Episode 3)
How did humans become such #devils? Even worse, how can some of our most #monstrous deeds come potentially from places of seeming good will? And how can we stop it as it is happening now? And how can we prevent it in the future? Are we damned to these #horrors forever?
-
PRODUCTHEAD: Product-led sales / product value / context & intent
» Switching to a product-led sales movement should not be a hard pivot but a blended transition
» The product must demonstrate its own value regardless of whether sales people are involved
» Intent adds nuance to the classic “strategic, execution, and planning” cascade
#prodmgmt #context #intent #productLed #sales
📖 Read more: https://imanageproducts.com/producthead-product-led-sales-product-value-context-intent/
-
ConTeXt’s sourcecode is slowly moving from github to codeberg!
(ATM the cb repo is secondary and needs to get updated manually, but it will become the main one over time.) -
It’s this time of the year when everyone asks for donations.
Here’s how you make an EPC QR code for SEPA transfers in #ConTeXt #LMTX:\startluacode
local sum = 50.0
local subject = "Donation"
local qrtext=string.format([[BCD
002
1
SCT
RABONL2U
ConTeXt Group
NL04RABO0163223726
EUR%.02f
CHAR
%s
]], sum, subject)
figures.qrcode(qrtext,"blue",tex.sp("5cm"))
\stopluacodeAccording to
https://en.wikipedia.org/wiki/EPC_QR_code
it works with banks in Austria, Belgium, Finland, Germany, and The Netherlands. -
-
context: current #worldcup match
fuck
fuck
fucfk
shit
fuck
#worldcup2022 #worldcup22 #worldcupqatar2022 #brasil #brazil #soccer #football #futebol -
Contextualize Me – The Case for Context in Reinforcement Learning
Carolin Benjamins, Theresa Eimer, Frederik Schubert et al.
Action editor: Adam White.
-
Contextualize Me – The Case for Context in Reinforcement Learning
Carolin Benjamins, Theresa Eimer, Frederik Schubert et al.
Action editor: Adam White.
