Thảo luận: Giá LLM nên dựa trên “zero‑shot intelligence”. Hiện nhiều công ty tính phí token đầu ra gấp 4‑5× token đầu vào, ngầm cho rằng đầu ra chứa “trí tuệ”. Nếu vậy, mức giá nền tảng cần phản ánh chất lượng trí tuệ – mô hình zero‑shot mạnh hơn nên có mức phí cao hơn. Bạn nghĩ sao? #LLM #AI #Pricing #TríTuệ #ZeroShot #AIVietnam
https://www.reddit.com/r/LocalLLaMA/comments/1qivp84/llms_value/
🤖 #Zeroshot creates autonomous engineering teams for #ClaudeCode #AI #DevTools #opensource
🔧 Point at a #GitHub issue, walk away, and return to verified, production-ready PRs
🛡️ Multiple isolated agents check each other's work - validators didn't write the code, so they can't lie about tests
🔄 Smart task classification: TRIVIAL (1 agent) → CRITICAL (7 agents with security, adversarial testing)
🧵 👇
Sopro TTS: A 169M model with zero-shot voice cloning that runs on the CPU
https://github.com/samuel-vitorino/sopro
#HackerNews #SoproTTS #ZeroShot #VoiceCloning #CPU #Model #AI #Technology
Mô hình TTS mã nguồn mở Sopro (169M tham số) vừa ra mắt! Sopro hỗ trợ chuyển văn bản thành giọng nói thời gian thực và nhân bản giọng nói zero-shot (chỉ cần 3-12s mẫu). Dù chỉ hỗ trợ tiếng Anh và chưa phải SOTA, mô hình đạt tốc độ nhanh 0.25 RTF trên CPU, giúp tạo 30s audio chỉ trong 7.5s.
#Sopro #TTS #AI #ZeroShot #VoiceCloning #MãNguồnMở #TríTuệNhânTạo #CôngNghệ
https://www.reddit.com/r/LocalLLaMA/comments/1q6sp4b/sopro_a_169m_parameter_realtime_tts_model_with/
Чистая правда: как компьютерное зрение помогает делать мир чище
Компьютерное зрение развивается стремительно: задачи, которые ещё недавно требовали собственных датасетов и долгого обучения моделей, теперь решаются готовыми инструментами. Всем привет! Меня зовут Алина, я инженер‑исследователь в Центре искусственного интеллекта СФУ. В этой статье расскажу, как мы применяем методы компьютерного зрения для анализа фотографий уборок в проекте «Чистые игры», как подошли к задаче автоматической оценки качества уборок и какие технические решения легли в основу MVP. В конце поделюсь выводами и тем, что удалось сделать, а что ещё предстоит улучшить.
I do not want to hear another word from anyone how useless #AI is...
... after a 5 hour (compute burn not RT), at the bottom of a very very deep rabbit hole...there was this magnificent conclusion that made me laugh out loud.
"The universe is just really efficient at disappointing itself." 🤡
~ Claude
I guess its one of those "You had to be there moments".
MultiADS: Defect-aware Supervision for Multi-type Anomaly Detection and Segmentation in Zero-Shot Learning
In manufacturing, quality control remains a critical yet complex task, especially when multiple defect types are involved. MultiADS introduces a system capable of detecting and segmenting a wide range of anomalies (e.g., scratches, bends, holes), even in zero-shot settings.
By combining visual analysis with descriptive textual input and using a curated Knowledge Base for Anomalies, MultiADS generalizes to unseen defect types without requiring prior visual examples and consistently outperforms state-of-the-art models across several benchmarks, offering a robust and scalable solution for industrial inspection tasks.
Sadikaj, Y., Zhou, H., Halilaj, L., Schmid, S., Staab, S., & Plant, C. MultiADS: Defect-aware Supervision for Multi-type Anomaly Detection and Segmentation in Zero-Shot Learning. International Conference on Computer Vision, ICCV 2025, Hawai, Oct 19-23, 2025, #ICCV2025. https://arxiv.org/abs/2504.06740.
Do you want to Master Prompt Engineering? Here’s my overview on how to Prompt like a Pro!
#promptEngineering #ConversationalAI #GenerativeAI #AIPrompts #PromptDesign #ZeroShot #FewShot #ChainOfThought #AIForBusiness #AIEdge #AIProductivity #WorkSmarter #DigitalTransformation #TechLeadership #FutureOfWork #Innovation #LinkedInCreators #AIInsights #TechTips #CareerGrowth
Zero-shot и Few-shot Learning в NLP
Zero-shot Learning (ZSL) - это способность модели выполнять задачи без каких-либо примеров обучения. Она делает это за счёт обобщённых знаний, полученных во время предобучения. Few-shot Learning (FSL) - это метод, при котором модели предоставляется всего несколько примеров (обычно от 1 до 5), чтобы лучше понять структуру задачи.
https://habr.com/ru/articles/897604/
#NLP #Zeroshot #Fewshot #Finetuning #Zeroshot_и_Fewshot_Learning_в_NLP #Zeroshot_или_Fewshot
Segment Anything Model Can Not Segment Anything - Assessing AI Foundation Model’s Generalizability In Permafrost Mapping
--
https://doi.org/10.3390/rs16050797 <-- shared paper
--
#GIS #spatial #mapping #remotesensing #foundationmodel #AI #artificialintelligence #zeroshot #segmentation #GeoAI #spatialanalysis #LargeLanguageModel #LLM #SAM #performance #metrics #permafrost #visionmodel #icewedge #Arctic #warming #climatechange #thawslumps #landform #terrainmapping #EuroCrops #agriculture
![maps/images - Results from knowledge-embedded learning with SAM. The results are those after fine-tuning. The images are the same as those in [other figure.]](https://files.mastodon.social/cache/media_attachments/files/113/953/101/638/955/634/original/5c838b30a20e5b84.png)
InstantID: Создание персонализированных изображений по одному фото. И лучший бесплатный генератор нейро-аватарок
Друзья, всем привет, вышла в свет новая удивительная технология, на гите она подписана как InstantID : Zero-shot Identity-Preserving Generation in Seconds . Но по факту это самый крутой генератор аватарок в мире, который по одной лишь вашей фотографии может создать персонализированное изображение с вашим лицом в любом стиле по текстовому запросу используя SDXL модели Stable Diffusion . Так что к концу этой статьи, у вас так же как и у меня будет десяток новых аватарок и совершенно бесплатно. В статье много красивых и тяжелых картинок.
https://habr.com/ru/articles/789028/
#InstantID #zeroshot #stablediffusion #аватарки #персонализация #controlnet #нейросети #генерация #sdxl
InstantID: Создание персонализированных изображений по одному фото. И лучший бесплатный генератор нейро-аватарок
Друзья, всем привет, вышла в свет новая удивительная технология, на гите она подписана как InstantID : Zero-shot Identity-Preserving Generation in Seconds . Но по факту это самый крутой генератор аватарок в мире, который по одной лишь вашей фотографии может создать персонализированное изображение с вашим лицом в любом стиле по текстовому запросу используя SDXL модели Stable Diffusion . Так что к концу этой статьи, у вас так же как и у меня будет десяток новых аватарок и совершенно бесплатно. В статье много красивых и тяжелых картинок.
https://habr.com/ru/articles/789028/
#InstantID #zeroshot #stablediffusion #аватарки #персонализация #controlnet #нейросети #генерация #sdxl
Bytedance DiffPortrait3D: Eine bahnbrechende KI-Methode zur Synthese neuer Ansichten
#DiffPortrait3D,#Bytedance,#KI,#3DBildsynthese,#Diffusionsmodelle,#ComputerVision,#Fotorealismus,#ZeroShot,#KünstlicheIntelligenz,#Technologieinnovation
Yesterday at #TPDL2023 David Pride presented “CORE-GPT: Combining Open Access research and large language models for credible, trustworthy question answering”
Rather than #ZeroShot question/answering, Pride’s team combines the #CORE #OpenAccess dataset with #ElasticSearch to create #FewShot prompts that leverage the strength of combining #search results with the #LLM’s (#GPT) #summarization abilities to produce an answer to a user’s question including citations.
Proud to announce our new paper, "Relevant Entity Selection: Knowledge Graph Bootstrapping via Zero-Shot Analogical Pruning" with Lucas Jarnac and Miguel Couceiro, accepted in #cikm2023
#knowledgeGraph #machineLearning #graphEmbedding #transferLearning #zeroShot #openScience
I have a #LLM that has 100% accuracy and #ZeroShot transfer on all blocks world problems. It is 'better than humans' who only have 80% accuracy. I call it the PDDLModel. With some prompt engineering (also called model writing), it can #zeroshot transfer to other problems like tower-of-hanoi, 8-puzzle, eaters, and what not. It even fits on a raspberry pi.
We can only imagine what this new #AGI would enable.
https://github.com/gerryai/PDDL4J/blob/master/pddl/blockworld/blocksworld.pddl
Let's all play this game!!
#Stanford Introduces #DetectGPT to Help #Educators Fight Back Against #ChatGPT Generated Papers
"This method, called ' #ZeroShot,' allows DetectGPT to detect machine-written text without any knowledge of the #AI that was used to generate it. It operates in stark contrast to other methods which require training #classifiers and datasets of real and generated passages."
#ArtificialIntelligence #Education #Teaching #OpenAI #TechNews #TechnologyNews #BreakingNews
@dx One of the papers in the #zeroshot space is "Language Models are Zero-Shot Reasoners" (https://arxiv.org/abs/2205.11916)
Followup work ("Impact of Pretraining Term Frequencies on Few-Shot Reasoning", https://arxiv.org/abs/2202.07206) found that the model only does well on arithmetic test items that appear at least 10^7 times in the training set (for the test item "what is 2 times 24?", the problem and answer appears 10^7 times in training set).
Hard to think of this as “zero” or “few” shot learning.
@cmaclell I think that's an important point. I like framing it as #transfer learning, because you have a network that has settled pretty solidly into a particular basin of the parameter space, based on the data it was trained with. This is VERY different than a system that has abstracted generalizable, relevant concepts, which is what I think people want to be true in #zeroshot learning.
I dislike how people frame the use of pre-trained models as engaging in #zeroshot or #fewshot learning.
There is rarely discussion about the data used for pre-training. I suspect the “generic” data used for pre-training probably includes examples similar to “task-specific” targets.
Without knowing the scope of the pre-training data, how can we evaluate the claim of few- or zero-shot learning?
It would be better to describe most of this work as investigating #transfer.
