Meet the first long-context OCR — PageIndex OCR. It can convert PDFs to Markdown with the document structure preserved. 📄 Upload a PDF ⚡ Get Markdown output (global structure preserved) 🌳 Navigate the hierarchical structure of the document 🆓 Try it now with 200 pages free:

Please do consider submitting to the ICML Workshop on Computer Use Agents. Deadline May 18.

Tired of the poor accuracy of similarity-based retrieval on long documents? You need reasoning-based RAG. We've open-sourced PageIndex — a tool for building indexes optimized for reasoning-based RAG and deep research. https://t.co/Q4KBfnLI4b

Talk by Nicolas Chopin, Prof of Statistics, Paris, hosted by AI Centre/ ELLIS https://t.co/jpqKXP1oHR Date: Monday 17th March 2025 Time: 12:30 - lunch; 1-2 - talk Location: UCL AI Centre, 90 High Holborn For more info and sign up: https://t.co/gBW60JS7AG

For the UK Gen AI community, we are inviting submissions for Gen AI research posters to be displayed during our launch event on 4 March at the Crick Institute in London. Please complete this form if you would like to be considered for the exhibition: https://t.co/diKuta1kZ1

Important article in the FT about the need for the UK to rethink university and PhD funding. PhD students are the lifeblood of research, teaching and innovation.

Anyone can ship AI features quickly, but getting them from OK to great really comes down to the quality of your evaluations. At @GustoHQ, we built our AI workflows around Humanloop, which has been core to shortening our eval cycle time. Congrats on launching, @humanloop!

🚀Mafin2 is based on our new MCTS RAG framework, achieving state-of-the-art accuracy for financial data!

This is a very nice piece of work, showing how to use the optimal form of the denoising covariance to improve generations.

The UCL Venturer Programme is a new way to fund PhD studentships and Postdocs and create a startup. If you're an investor or corporate looking to access world leading AI talent for innovation, please do join us at the UCL AI Centre on 14th November

@radicalvcfund @maxjaderberg @IsomorphicLabs Additionally, @radicalvcfund has launched applications for the first-ever Radical Cloud Cohort, offering selected early-stage AI builders $250K in cloud credits and a year of programming to help scale their businesses. Apply:

Excited to join @radicalvcfund as a Scientific Advisor. I’ll be starting with the Radical AI Founders Masterclass on Oct 30, in a conversation with @maxjaderberg of @IsomorphicLabs. Register here:

https://t.co/duQpMHuuJV Application deadline 5th November. See the pdf in the link for more details.

With funding and mentoring from Google DeepMind @ai_ucl is looking for a Fellow to contribute impactful research in the broad area of sustainable AI. The Fellow will be a proven advocate for increasing participation in AI and an ambassador for inclusion during the Fellowship.

We show how to take a pretrained model (like Gemma) and extend the context significantly with only linear scaling cost. As good as Mamba and other SOTA models for language modelling but with global context. Great work by @ai_ucl @UiPath https://t.co/9ixjIa3koR

I will be in ICML 2024 presenting two conference papers and one workshop paper about diffusion methods, LLM fine-tuning and next-generation network architecture. #ICML2024 #ICML [1/5]

Wonderful anecdote from Turing Award winner Manuel Blum who was told by his teachers that he would be lucky even to finish school. Still going strong at 86. Great to chat with @mmbronstein and make new friends. Thanks #WAIC2024 and unidt for hosting.

Llama3 shows the importance of dataset quality and size during fine-tuning. How to achieve effective fine-tuning while minimizing labeling costs? Check out our paper on active preference learning for LLM, now accepted by #ICML2024 🎉. Stay tuned for the camera-ready version.

Retrieval Augmented Generation (RAG) is crucial for building AI applications that require contextual awareness or specific domain knowledge. Our latest explainer is an introduction to RAG - what it is, how it works and why it's important. Read here: https://t.co/bq4bZ6sP8Y

Latent Attention (Latte) is a new linear time transformer. Unlike sparse methods, all tokens play a role and Latte can be used in causal and non-causal settings. Great work by @raresdolga97 @cobzarenco @ai_ucl and @UiPath. Comments welcome. https://t.co/9ixjIa2Mzj