Dmitrii Khizbullin

writehere by ruibin xiong, @Beastlyprime @dmitrii_tech @mingchen @SchmidhuberAI is fantastic. I use it with gpt-4o-mini and it's a valid replacement for most of my openai deep searches.

@sameed_ahmad12 @bentossell Thanks for spotting. The website is back online.

@dmitrii_tech @bentossell the site is down.

this writing model is goooood

@urosn @SchmidhuberAI Good to know, thank you. We will keep multilingual support in mind.

@dmitrii_tech @SchmidhuberAI Serbian (Latin+Cyrillic) for example 😀

What if AI could write creative stories & insightful #DeepResearch reports like an expert? Our heterogeneous recursive planning [1] enables this via adaptive subgoals [2] & dynamic execution. Agents dynamically replan & weave retrieval, reasoning, & composition mid-flow. Explore & try: writehere.site [1] Ruibin Xiong, Yimeng Chen, Dmitrii Khizbullin, Mingchen Zhuge, Juergen Schmidhuber. Beyond Outlining: Heterogeneous Recursive Planning for Adaptive Long-form Writing with Language Models. arxiv.org/abs/2503.08275 [2] J. Schmidhuber & R. Wahnsiedler. Planning simple trajectories using neural subgoal generators. In Proc. SAB'1992, p 196–202, MIT Press, 1992. Based on TR FKI-129-90, TU Munich, 1990.

@SchmidhuberAI For those who are interested in an open-source alternative to OpenAI's Deep Research feature, please check out the Report Generation tab of our demo. You will need to provide your Serp API key to have your report grounded in the latest web information.

@RealAIGuys @SchmidhuberAI If you are more interested in highly grounded report generation, please check out the Report Generation tab of our demo. It is a competitive open-source replacement for OpenAI's Deep Research.

@SchmidhuberAI Let creative writing stay with humans, we don't need AI for that.

@urosn @SchmidhuberAI So far, the engine has been extensively validated for English. Please let us know what languages you are interested in.

@SchmidhuberAI What languages are available and will be available in future?

@2xehpa @SchmidhuberAI Practically, a story expends $0.1-0.5 with gpt-4o.

@SchmidhuberAI Looks nice! Can you estimate the number of token needed before entering my API key? Im a bit worried...

🚀 #WriteHERE is LIVE! Experience adaptive AI writing with our open-source framework—no rigid outlines, just human-like creativity! ✨ What it can do: ✅ Creative Fiction: Craft mystery/sci-fi stories with real-time plot adjustments ✅ Deep Research Reports: Generate technical docs with seamless knowledge integration ✅ Live Demo: Test it yourself—transparent and fully open-sourced! 🔥 What makes it unique: ⚡️ Mid-writing reasoning & task-switching ⚡️ Heterogeneous recursive task decomposition 📲 Try the demo: writehere.site ⭐️ Star us on GitHub: github.com/principia-ai/W… Join the future of AI writing—where plans adapt, not constrain. 🤖✍️ #AI #OpenSource #WriteHERE

We just hit 10K ⭐️ on GitHub! Many interesting updates are coming: workforce for orchestrations, MCP integration for tools, memory toolkits for knowledge management and reinforcement learning for evolution!!! I am grateful for every community members’ efforts to make this possible!! 🦉github.com/camel-ai/owl

Ever noticed how one scene seamlessly transitions into another in films? That’s a match-cut—a subtle yet powerful cinematic trick. Our MatchDiffusion, generates two videos from text prompts, designed to form a seamless match-cut—effortlessly and training-free. 🎥✨ 1/n

Are you a rising star in AI? 🌟 Join us as a speaker for the 4th edition of the KAUST Rising Stars in AI Symposium. In the past 2 years co-organizing this event, I've met incredible researchers now in top industrial and academic positions worldwide. More info: 📅 Event date: April 7-10, 2025 ⏳ Application deadline: December 18 🔗 Apply here: kaust.edu.sa/en/news/rising…

Error bars are very important, but please let's calculate valid ones! TL;DR Calculate valid and tight confidence intervals using our state-of-the-art approach, now with Python code too: github.com/bremen79/preci… This recent post by @AnthropicAI gives me the perfect excuse to advertise the Python code I just published of our state-of-the-art method to calculate confidence intervals. Why would you need something different from what they propose? Well, the approach suggested in this paper will give you wrong confidence intervals. Example: consider the sequence of random variables 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 The approach in the paper will give a confidence interval of 1±0, with probability of 95%. This is clearly wrong! The problem is that the Gaussian approximation they use holds only asymptotically, so using it with a small sample size is a risky business. In alternative, you can use the approach @kwangsungjun and I recently developed, that gives - always valid intervals - never vacuous, even with 1 sample - holds uniformly over time, so you can look at the intervals and decide if you want to add more samples without ruining the guarantee - with Matlab and Python code Paper: ieeexplore.ieee.org/document/10315… Code: github.com/bremen79/preci…

#مجموعة_بيانات مجموعة بيانات مترجمة آليًا لدعم النماذج اللغوية العربية الصغيرة (FineWeb-Edu-Ar) البيانات هي نسخة مترجمة آليًا من مجموعة البيانات الشهيرة FineWeb-Edu المتاحة عبر منصة HuggingFace، والمُنتقاة بعناية بعد إزالة البيانات المكررة. لذا يُعدّ FineWeb-Edu-Ar أكبر مجموعة بيانات مترجمة آليًا للغة العربية متاحة للجمهور، حيث تحتوي على ما يصل إلى 202 مليار رمز (tokens) باستخدام أداة تحليل بيانات مُدربة على اللغة العربية. تشكل هذه المجموعة خطوة مهمة في تزويد النماذج اللغوية المتعددة اللغات ببيانات عالية الجودة باللغة العربية، مما يدعم تحسين أداء النماذج اللغوية ويعزز دقتها في سياق #اللغة_العربية، خصوصًا في ظل قلة البيانات الأصلية المتاحة باللغة العربية على الإنترنت. للحصول على البيانات: huggingface.co/datasets/kaust… لقراءة الورقة البحثية: arxiv.org/abs/2411.06402

Have you ever wondered what happens when you reduce a story to a low-dimensional latent representation? Well my new IEEE TPAMI work with @idivinci, Zachary Friggstad, and @SchmidhuberAI shows that it has some pretty cool applications 🎉 Check it out: doi.org/10.1109/TPAMI.… (1/n)

This is the best paper written so far about the impact of AI on scientific discovery

Meta presents MarDini Masked Autoregressive Diffusion for Video Generation at Scale

Agent-as-a-Judge: An Advanced AI Framework for Scalable and Accurate Evaluation of AI Systems Through Continuous Feedback and Human-level Judgments Meta AI and King Abdullah University of Science and Technology (KAUST) researchers introduced a novel evaluation framework called Agent-as-a-Judge. This innovative approach uses agentic systems to evaluate other agentic systems, providing detailed feedback throughout the task-solving process. The researchers developed a new benchmark called DevAI, which includes 55 realistic AI development tasks, such as code generation and software engineering. DevAI features 365 hierarchical user requirements and 125 preferences, offering a comprehensive testbed for evaluating agentic systems in dynamic tasks. The introduction of Agent-as-a-Judge enables continuous feedback, helping to optimize the decision-making process and significantly reducing the reliance on human judgment. The Agent-as-a-Judge framework assesses agentic systems at each task stage rather than just evaluating the outcome. This approach is an extension of LLM-as-a-Judge but is tailored to the unique characteristics of agentic systems, allowing them to judge their performance while solving complex problems. The research team tested the framework on three leading open-source agentic systems: MetaGPT, GPT-Pilot, and OpenHands. These systems were benchmarked against the 55 tasks in DevAI. MetaGPT was the most cost-effective, with an average cost of $1.19 per task, while OpenHands was the most expensive at $6.38. Regarding development time, OpenHands was the fastest, completing tasks in an average of 362.41 seconds, whereas GPT-Pilot took the longest at 1622.38 seconds.... Read the full article: marktechpost.com/2024/10/18/age… Paper: arxiv.org/abs/2410.10934… Dataset: huggingface.co/DEVAI-benchmark Listen to the podcast as well on 'Agent-as-a-Judge': youtube.com/watch?v=ctasuN… @MingchenZhuge @SchmidhuberAI @tydsh @zechunliu @vikasc @YoungXiong1 @vikasc @Obs01ete @erniecyc @oneDylanAshley