Real users hit ML endpoints with unpredictable inputs. InferProbe lets you simulate that locally — no sanitized tests. What's one unpredictable scenario you wish you could test more easily?
machinelearning
I built a tool to find problems hiding in my training data.
LabelLens analyzes labeled text classification datasets for duplicates, mislabels, and class imbalance. Ran it on my own 26K sample dataset — found 5,664 exact duplicates I had no idea about.
Try it: https://huggingface.co/spaces/mikenoe/label-lens
Blog post: https://mikenoe.com/posts/i-built-a-tool-to-find-the-problems-in-my-training-data/
⏳ Only 1 month until #ArcOfAI!
If you want to build AI like the pros, now’s the time to act.
Regular registration ends tomorrow, March 14.
Bring your team, learn together, and stay ahead in AI development. Tickets currently BOGOF 🚀
#AI #GenAI #AIEngineering #AgenticAI #MachineLearning #LLM #TechConference #DeveloperTools #Austin #DevCommunity #AustinTech
LLMs are changing development—but do you really understand how they work? 🧠
At #ArcOfAI, Luke VanderHart explores the history, philosophy & linguistics behind modern language models and what it means for developers.
https://www.arcofai.com/speaker/59ce30d57920486c808a24605882d04a
🎟 Tickets: https://arcofai.com
#AI #LLM #GenerativeAI #MachineLearning #DeveloperTools #AIEngineering #NaturalLanguageProcessing #TechConference #SoftwareDevelopment #AgenticAI
Guess I should post a new introduction since I moved servers and this account has no posts at all 😛
I'm a developer/coder. I used to be very much into mobile development (both #iOS and #Android) using #Swift and #Kotlin. Now my main area of interest is digital creativity using #AI to generate art, text, and music.
I don't post a lot generally, but do daily postings of interesting papers from the cs.CV category on arXiv.org. So first thing in the morning, there will be a barrage of posts from me and then probably nothing much for the rest of the day 🙂
I also post a daily set of images based on my #StableDiffusion prompt of the day — I generate a bunch of images through the day based on a single prompt, pick the four best images (according to me) and post them the next day.
Other than that, I do boost a lot — #MachineLearning and #DeepLearning, #Writing, #AIArt, #Photographs, #Humour are probably the main topics of interest for me. But if I find anything interesting in the areas of #Books, #TV, #Movies, I'd boost those too 🙂
#introduction #introductions
I'm a developer/coder. I used to be very much into mobile development (both #iOS and #Android) using #Swift and #Kotlin. Now my main area of interest is digital creativity using #AI to generate art, text, and music.
I don't post a lot generally, but do daily postings of interesting papers from the cs.CV category on arXiv.org. So first thing in the morning, there will be a barrage of posts from me and then probably nothing much for the rest of the day 🙂
I also post a daily set of images based on my #StableDiffusion prompt of the day — I generate a bunch of images through the day based on a single prompt, pick the four best images (according to me) and post them the next day.
Other than that, I do boost a lot — #MachineLearning and #DeepLearning, #Writing, #AIArt, #Photographs, #Humour are probably the main topics of interest for me. But if I find anything interesting in the areas of #Books, #TV, #Movies, I'd boost those too 🙂
#introduction #introductions
ML endpoints in the wild see messy inputs: noise, typos, adversarial tricks. InferProbe simulates that mess locally. What's the most chaotic real input you've had to test?
InferProbe wants to let ML engineers test endpoints without fear — local, private, cheap. Perturbations + explanations on your machine. What one barrier is holding back your testing right now?
Nun noch Mal eine #introduction. Ich bin Philip, mag #opensource und #opendata, programmiere nur noch in der Freizeit etwas #Python, interessiert an #machinelearning, Kochen und Selbermachen. Lebe im Raum Düsseldorf und trage gerne zu #OpenStreetMap und anderen Projekten bei. #neuHier
Meine Pronomen sind er/ihm.
Edited 2y ago
How do you build AI systems that can actually research? 🔎
At #ArcofAI, Vanya Seth & Sarang Kulkarni from Thoughtworks share how multi-agent architectures, context engineering, and reasoning workflows power deep research assistants.
https://www.arcofai.com/speaker/f0de9de7e37a4159939ba080d7cd1979
🎟 Get tickets: https://arcofai.com
#AI #AgenticAI #GenAI #AustinTech #MachineLearning #AIEngineering #Architecture #SoftwareEngineering #Developer #Conference #TechConference
Many ML models look great in dev but drift and fail in production. InferProbe lets you test for drift locally before it becomes a problem. Have you ever had a model break after going live?
AI workshops designed for builders, not spectators. 🧠
At Arc of AI, our workshops focus on how AI is actually designed, built, and
operated in real systems. From RAG pipelines and agent workflows to AI-driven
API design, you’ll get hands-on guidance you can apply immediately.
https://www.arcofai.com/schedule
Learn from industry experts actively building production AI.
🎟️ Secure tickets: https://arcofai.com
#ArcOfAI #AIEngineering #GenAI #MachineLearning #SoftwareEngineering #LLM #AustinTech
Improving Turbulence Models
Calculating turbulent flows like those found in the ocean and atmosphere is extremely expensive computationally. That’s why forecasting models use techniques like Large Eddy Simulation (LES), where large physical scales are calculated according to the governing physical equations while smaller scales are approximated with mathematical models. Researchers are always looking for ways to improve these models–making them more physically accurate, easier to compute, and more computationally stable.
In a new study, researchers used an equation-discovery tool to find new improvements to these models for the smaller turbulent scales. They started by doing a full, computationally expensive calculation of the turbulent flow. The equation-discovery tool then analyzed these results, looking to match them to a library of over 900 possible equations. When it found a form that fit the data, the researchers were then able to show analytically how to derive that equation from the underlying physics. The result is a new equation that models these smaller scales in a way that’s physically accurate and computationally stable, offering possibilities for better LES. (Image credit: CasSa Paintings; research credit: K. Jakhar et al.; via APS)
#CFD #computationalFluidDynamics #fluidDynamics #geophysics #largeEddySimulation #machineLearning #mathematics #numericalSimulation #physics #science #turbulence
ML endpoints in production see messy inputs: noise, adversarial attacks, typos. InferProbe simulates all of it locally. What's the most chaotic real input you've had to debug? #MachineLearning #ArtificialIntelligence #DevTools #ML #AI
Great AI results start with great prompts 🚀
At #ArcofAI, Craig Walls explores how to craft effective prompts, apply prompt engineering techniques, and use emerging Agentic AI patterns to turn LLMs into powerful development tools.
https://www.arcofai.com/speaker/c57969c9cacc4475a111aa8edf02969b
🎟 Get tickets: https://arcofai.com
#AI #GenAI #PromptEngineering #AgenticAI #TechConference #AustinTech #LLM #MachineLearning
What are people exploring in AI right now?
At #ArcofAI, sessions dive into AI-enabled apps, multimodal systems, AI-powered workflows, and responsible AI.
Take a look at some of the topics in this year’s program: https://www.arcofai.com/program
🎟 Tickets: https://arcofai.com
#AI #TechConference #MachineLearning #Developers #EnterpriseAI #Innovation #ArcOfAI #LLM #Security #ResponsibleAI #Arhitecture #AustinTech
InferProbe wants to move ML debugging from reactive to proactive: local aggressive testing + clear anomaly explanations. What one thing would help you ship ML features faster and safer?
If you're not stress-testing ML endpoints for hallucinations, you're shipping with unknown risk. InferProbe will provoke and explain them locally. What's the most surprising hallucination you've seen?
ML engineers often limit test coverage to control cloud costs or protect data. InferProbe removes that limit — aggressive perturbation testing happens locally. What edge case do you under-test right now?
AI is moving fast — don’t try to figure it out alone 🤖
Right now #ArcOfAI tickets are BOGO, so you can bring a colleague for free & learn together.
⏳ Regular registration ends March 14
Join the developers & engineers building real, production-ready AI systems.
🎟️ Get tickets: https://arcofai.com
#AI #GenAI #ArtificialIntelligence #MachineLearning #Developer #SoftwareEngineering #AIEngineering #TechConference #DevCommunity
heise+ | KI für mehr Effizienz? Forscher warnt vor Wissensverlust in Unternehmen
In einem Interview erklärt Prof. Jin Gerlach, warum das Versprechen von KI-Tools einer radikalen Effizienzsteigerung fatale Folgen haben könnte.
#Interviews #KünstlicheIntelligenz #MachineLearning #Wissenschaft #news
I guess I’m getting old, but I really miss the days when a computer and other consumer electronics were just useful tools, instead of the hideout of privacy-invading, overzealous, often superfluous personal assistants that connect to data-hungry corporations. 🫤
#tech #technology #BigTech #AI #ArtificialIntelligence #LLM #LLMs #ML #MachineLearning #GenAI #generativeAI #AIAgent #AISlop #Fuck_AI #Microsoft #copilot #Meta #Google #NVIDIA #gemini #OpenAI #ChatGPT #anthropic #claude #computers #IT
Mais bien sûr, je pensais que lire la documentation allait m'aider... 💀
#machinelearning
Peut-être que je devrais d'abord commencer par réviser mes maths avant d'essayer de la mettre en pratique 😥
(C'est pour mon cours de Machine learning, et la documentatino c'est la documentation de la régression logistique https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression)
RE: https://mathstodon.xyz/@tao/115722360006034040
On the Measure of Intelligence (2019) https://arxiv.org/abs/1911.01547 is more relevant every day.
Two views of intelligence: Intelligence as a collection of task-specific 𝐬𝐤𝐢𝐥𝐥𝐬, or intelligence as 𝐬𝐤𝐢𝐥𝐥-𝐚𝐜𝐪𝐮𝐢𝐬𝐢𝐭𝐢𝐨𝐧 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐜𝐲, relative to priors, experience, and generalization difficulty.
Isaac Newton’s skill in doing calculus was low compared to that of the average mathematician today. However, Newton was able to develop the entirely new skill of calculus, starting only with the priors and experience available in the 17th century.
Newton had a dual capacity:
1. The ability to learn how to stand on the shoulders of giants.
2. The ability to see further than those who came before him.
"AI community still gravitates towards benchmarking intelligence by comparing the skill exhibited by AIs and humans at specific tasks [...] solely measuring skill at any given task falls short of measuring intelligence, because skill is heavily modulated by prior knowledge and experience: unlimited priors or unlimited training data allow experimenters to "buy" arbitrary levels of skills for a system, in a way that masks the system's own generalization power." – Francois Chollet (2019)
Edited 83d ago
If your AI feels unpredictable, the model might not be the problem.🤔
At #ArcOfAI, Brent Laster introduces Context Engineering — practical ways to design smarter, more reliable AI systems without fine-tuning.
https://www.arcofai.com/speaker/590f34c7b6364bd8bfe7e6068f316fdb
🎟️ Tickets: https://arcofai.com
#AI #GenAI #MachineLearning #SoftwareEngineering #PromptEngineering #AIArchitecture #LLM #TechConference #AILeadership
[zvec - 초경량·초고속 인프로세스 벡터 DB
Zvec는 초경량·초고속 인프로세스 벡터 DB로, Alibaba의 Proxima 엔진 기반으로 구축되어 대규모 유사도 검색을 최소 설정으로 수행하도록 설계되었습니다. 밀집 및 희소 벡터를 모두 지원하며, 하이브리드 검색 기능을 통해 의미적 유사도와 구조적 필터링을 결합한 정밀 검색을 제공합니다. C++ 기반 핵심 엔진과 SWIG·Python 바인딩 구조로 구성되어 고성능 연산과 다양한 언어 통합을 지원하며, Apache-2.0 라이선스로 제공됩니다.
https://news.hada.io/topic?id=27147
#vectordatabase #algorithms #machinelearning #opensource #search