Issue 54 28th May 2024

Highlights

Big tech has distracted world from existential risk of AI, says top scientist

Max Tegmark argues that the downplaying is not accidental and threatens to delay, until it’s too late, the strict regulations needed Big tech has succeeded in distracting the world from the existential risk to humanity that artificial intelligence still poses, a leading scientist and AI campaigner has warned. Speaking with the Guardian at the AI Summit in Seoul, South Korea, Max Tegmark said the shift in focus from the extinction of life to a broader conception of safety of artificial intelligence risked an unacceptable delay in imposing strict regulation on the creators of the most powerful programs.

Microsoft intros a Copilot for teams

Microsoft wants to make its brand of generative AI more useful for teams — specifically teams across corporations and large enterprise organizations. This morning at its annual Build dev conference, Microsoft announced Team Copilot, the latest expansion of its Copilot family of generative AI tech. Unlike Microsoft’s previous Copilot-branded products, Team Copilot isn’t so much […]

This Week in AI: OpenAI and publishers are partners of convenience

Keeping up with an industry as fast-moving as AI is a tall order. So until an AI can do it for you, here’s a handy roundup of recent stories in the world of machine learning, along with notable research and experiments we didn’t cover on their own. By the way, TechCrunch plans to launch an AI newsletter […]

Microsoft Build 2024: All the AI and hardware products Microsoft announced

Copilot, Microsoft's brand of generative AI, will soon be far more deeply integrated into the Windows 11 experience.

Paper of the week

Understanding Diffusion Models

Generative AI is a highly complex subject. LLMs are probably the best-understood models among all of the generative models in AI, and most of us will still have difficulty understanding the maths behind these models. This week's paper of the week tries to clear up another type of generative model, the diffusion model. It's used for image generation, but how does it work? Prepare to be amazed!

Video

AI Show On demand | AI Integration with DataRobot

Join us as we talk with our special guest, Venky Veeraraghavan - Chief Product Officer of DataRobot, and learn more about how DataRobot partnered with Micros...

Articles

Profiling CUDA using Nsight Systems: A Numba Example

Following my initial series CUDA by Numba Examples (see parts 1, 2, 3, and 4 ), we will study a comparison between unoptimized, single-stream code and a slightly better version which uses stream concurrency and other optimizations. In this example we are following the "reduce" pattern introduced in article CUDA by Numba Examples Part 3: Streams and Events to compute the sum of an array.

An Introduction to Reinforcement Learning

Dynamical programming solves general stochastic optimal control problems (afflicted by the curse of dimensionality — meaning that computational requirements grow exponentially with the number of state variables) by decomposing them into smaller sub-problems and computing the value function. As we demonstrate the rudiments of reinforcement learning, we will delve into the heart of dynamic programming: the recursive relationship between the state and value functions of the agent.

Quantize Llama 3 8B with Bitsandbytes to Preserve Its Accuracy

Llama 2 vs. Llama 3 vs. Mistral 7B, quantized with GPTQ and Bitsandbytes.

Elon Musk’s xAI raises $6bn in bid to take on OpenAI

Funding round values artificial intelligence startup at $18bn before investment, says multibillionaire Elon Musk’s artificial intelligence company xAI has closed a $6bn (£4.7bn) investment round that will make it among the best-funded challengers to OpenAI. The startup is only a year old, but it has rapidly built its own large language model (LLM), the technology underpinning many of the recent advances in generative artificial intelligence capable of creating human-like text, pictures, video, and voices. Continue reading...

Why Google’s AI might recommend you mix glue into your pizza

I got “distrust and verify” as advice on using LLMs into this Washington Post piece by Shira Ovide.

Maximizing Training Throughput Using PyTorch FSDP and Torch.compile

Recently, we demonstrated how FSDP and selective activation checkpointing can be used to achieve 57% MFU (Model Flops Utilization) for training a 7B model on A100 GPUs. We also demonstrated how it can train a high quality model, which we open sourced as Granite 7B base model on Hugging Face Hub under the Apache v2.0 license.

Turning your relational database into a graph database

Leveraging LLMs to augment your data.

Navigating the Maze: A Guide to Data Architectures

Data is the lifeblood of any organization today. But how do you ensure your data is organized, secure, and accessible for all the needs of…