Does anyone remember the name of a paper that was floating around a few weeks back, that bypassed the need to organise sequences into batches by length... and instead did something fancy with the attention mechanism and let you pack sequences chronological into a batch?

It's interesting to observe how competitive the chinese AI eco-system is, yet they are also the most regulated and their government aggressively enforce those regulations.

if nvidia doesn't start ramping up the vram on the consumer grade GPUs, they will loose the local/hobby AI/LLM market to Apple and/or AMD.

I see many surprised and overwhelmed by everything in the news recently. Tip: Everything happening makes more sense if you try to understand the interplay between liberal internationalism and great power politics. You don't have to agree with it, it's manifestation requires

Dear EU officials I am co-founder of Klarna one of Europes largest fintech companies. Ursula von der Leyen Please read this about open banking regulation how it needs to be reformed to bring value and innovation. Happy to speak whenever EU ent. help me spread this so she sees it!

🚨 New Publication: "The European Way: A Blueprint for Reclaiming Our Digital Future" Our paper lays out a bold blueprint for strategic autonomy without isolationism. No to dig. dependency. No to protectionism. Yes to trusted int. partnerships. 🔗 Read ft.com/content/2a7812…

Here the link to our policy paper: papers.ssrn.com/sol3/papers.cf…

Terrence Tao is posting videos on how to use AI in maths. This is significant given how skeptical many mathematicians are of LLMs as a useful tool for them. youtu.be/zZr54G7ec7A

One of Visual Studio's biggest losses was deprecating UML designers and modelers (circa 2016). With "vibe coding" on the rise, this would be a pretty good way to architect and orchestrate solutions together with AI models, ensuring strong guidance, steerability and control. Bring

Deep Learning is only useful when a few conditions are met: 1. Modeling the problem is hard. 2. You can collect enough data. 3. Evaluation is 'relatively' easy. If these three conditions cannot be met, then deep learning (i.e. AI) is most likely not the best solution for your

I just read this WSJ article on why Europe's tech scene is so much smaller than the US's and China's. I'm afraid that, like most articles on this topic, it largely misses the mark. Which in itself illustrates a key reason why Europe is lagging behind: when you fail to

🚨 Did you know that the EU has 101 adopted digital laws? These experts are proposing RADICAL changes to the EU's digital policy and advocating for a new approach called "The European Way." Here's what you need to know: The paper "The European Way - A Blueprint for Reclaiming

Okay currently testing out Gemini Diffusion, and it is a marked change in the developer experience… although performance is still lacking, the speed is impressive!

AI benchmarking culture is completely out of control. Tables with dozens of methods, datasets, and bold numbers, trying to answer a question that perhaps no one should be asking anymore.

Hmmm... It's always interesting to watch the pendulum swing back and forth... Watching the AI/LLM space for the last decade has shown me that it's not just that there are alot of people who don't understand the topic.. But there are also many who do, but are willing to

Big news: we’re kicking off the search for ARIA’s next CEO. This is a once-in-a-lifetime opportunity to lead one of the most ambitious orgs in science and tech. 1/8

Dealing with memory across space and time is an open and grand challenge for NN-based models.. particularly in the long-horizon case

Another day, another top job available in the Sovereign AI Unit. We are looking for someone that can sniff out amazing startup partnership opportunities, transformative datasets to invest in, and think through what the UK's most ambitious plays can be Help the UK stand up tall!