AI News Now – UAE’s Falcon 40B Goes Royalty-Free, Gorilla Goes Ape for APIs, New Superbug Killer Discovered by AI, QLoRA Brings Mem Efficient Finetune
AI Infrastructure Alliance
We’re dedicated to bringing together the essential building blocks for the AI/ML applications of today and tomorrow.
Every Monday we deliver the latest and greatest news in AI. We sift through all the stories so you don't have to and bring you the best breaking news, papers, articles and more.
The UAE’s Technology Innovation Institute (TII) announced that their acclaimed Falcon 40B open source model is now royalty-free for commercial and research use. Originally, the creators wanted to charge 10% on commercial ventures over $1M but after pushback from the community the research institute wisely shifted focus.?Now we’ve got a powerful new model that anyone can use for research and for commercial purposes for free, without any restrictions or fees and it’s a major milestone for open source LLMs, with the model competitive to GPT-3.5 and GPT 4 on multiple benchmarks.
Falcon 40B stands at #1 globally on Hugging Face’s leaderboard for large language models. This is a significant achievement, as Hugging Face is probably the most well-respected platform for natural language processing, starting with their brilliant transformers library and racing forward to today where they stand as the top open source model hub. Falcon 40B’s high ranking on the Hugging Face benchmarks matters because the open tests are repeatable by other researchers, rather than just claims in a paper that nobody can verify.
The newly minted Falcon 40B is now available under the well understood Apache 2.0 software license, instead of a tangled and modified version of it. This license is widely used in the industry and is known for its flexibility and compatibility with other licenses. By providing unrestricted access to this language model, TII hopes to cultivate a thriving ecosystem of collaboration, innovation, and knowledge sharing.
The open-source, royalty-free deployment of Falcon 40B could empower public and private sector entities to develop new applications and solutions that leverage the power of natural language processing in ways they just can’t do with closed source models.?
Overall, TII’s decision to make Falcon 40B royalty-free is a positive development for the AI industry and for open source machine learning. Open-source software is the driving force behind much of the tech industry today. Everything from the cloud, to SaaS services, supercomputers and edge routers runs on open source but much of the power of AI has stayed closed until now. ?We look forward to seeing how Falcon 40B will weave its way into commercial applications and hope it drives other people to push forward with more openness and transparency.?
?“AI allows paralyzed person to ‘handwrite’ with his mind” - this headline might sound like something out of a sci-fi movie, but it’s actually a reality! Thanks to advances in artificial intelligence, a paralyzed man was able to “write” with his mind. By using a brain-computer interface and an algorithm that translated his thoughts into text, he was able to communicate with others in a way that was previously impossible. The researchers hope this technology can eventually be used to help paralyzed individuals communicate more easily and efficiently.
Scientists have discovered a new antibiotic, abaucin, which is effective against Acinetobacter baumannii, using artificial intelligence (AI). The AI was used to screen millions of potential compounds for antibiotic discovery, which accelerated and expanded the search for novel antibiotics. Abaucin is precise and could lead to fewer side-effects, making it harder for drug-resistance to emerge. This breakthrough is a “big game-changer” and will save countless lives, as other antibiotics are facing huge surge in killer bug resistance and antibiotics are expensive to research and hold little of the blockbuster potential of other commercial drugs. AI hold the potential to speed up discovery of these crucial compounds that make everything from surgery to surviving minor cuts a reality in the modern world.
领英推荐
QLoRA is a new approach to finetuning quantized LLMs that reduces memory usage while still achieving state-of-the-art results. It introduces innovative techniques that save memory without sacrificing performance. QLoRA introduces a number of innovations to save memory without sacrificing performance.?The researchers used QLoRA to finetune more than 1,000 models, providing a detailed analysis of instruction following and chatbot performance across 8 instruction datasets, multiple model types (LLaMA, T5), and model scales that would be infeasible to run with regular finetuning (e.g. 33B and 65B parameter models). Their results show that QLoRA finetuning on a small high-quality dataset leads to state-of-the-art results, even when using smaller models than the previous SoTA.
Meet Voyager, an embodied agent that uses Large Language Models to learn and explore Minecraft . This lifelong learning agent leverages GPT-4 to develop increasingly sophisticated skills and outperforms baselines in discovering novel items, unlocking tech trees, traversing terrains, and applying skills to unseen tasks. The best part? Voyager does not require model parameter fine-tuning and is built on existing frameworks such as NeRFies, CLIPort, and VIMA. Get ready to embark on an exciting journey with Voyager!
英伟达 has announced the release of its latest breakthrough in GPU-accelerated computing, the NVIDIA DGX GH200. This system boasts 256 GPUs and 128 TBps bi-section bandwidth, and is powered by the NVIDIA Grace Hopper Superchip, which combines Grace and Hopper architectures with NVLink-C2C for CPU + GPU coherent memory model. The NVLink Switch System forms a two-level, non-blocking, fat-tree NVLink fabric to connect 256 GPUs. NVIDIA Modulus, a physics-ML platform, is now open source, and version 23.1 of NVIDIA HPC SDK introduces CUDA 12 support.
In a bizarre turn of events, a lawyer has used ChatGPT for a legal filing and didn’t realize that the model was hallucinating fake cases for the filing. The judge was not amused and has issued an ORDER TO SHOW CAUSE why counsel should not be sanctioned. This highlights the potential dangers of relying on AI language models without proper supervision and raises questions about the ethical use of such technology in the legal profession. Simon Willison has created a timeline of the events surrounding the case, which makes for an interesting read.
Also this week:
Thanks for supporting "AI News Now". See you next week!?