AI Edition - It's AI Time - 75 [28-06-2024]
AI Edition - It's AI Time - 75
Snap Lense Studio 5.0
Snap has released the GenAI suite in Lens Studio 5.0 - this is a great advancement and is massively useful for developing AR applications.
Instagram Launching An AI Studio
Instagram's new "AI Studio" lets creators develop AI chatbot versions of themselves. It is currently rolling out as an early test in the US.
Apple, Microsoft Shrink AI Models to Improve Them
Tech giants are shifting focus from large language models to more efficient small language models (SLMs), with Apple and Microsoft introducing models with substantially fewer parameters yet comparable, or even superior, performance in benchmarks. OpenAI's CEO suggests we're moving beyond the era of LLMs, as SLMs offer advantages like local device operation, increased accessibility for smaller entities, and potential insights into human language acquisition. Training SLMs on high-quality or "textbook-quality" data contributes to their effectiveness despite their smaller scale.
LLM Compiler
Meta has released two language models that can compile code to assembly and decompile to LLVM IR. They were trained on 546 Billion tokens of high-quality data then further instruction tuned. They are able to achieve 77% of optimized assembly performance and have 45% round trip disassembly performance.
OpenAI has trained a model based on GPT-4 that can catch GPT-4 mistakes and help humans spot mistakes during the RLHF process.
Gemma 2 is now available to researchers and developers
Google DeepMind has released Gemma 2, a family of high-performing and efficient open models for various AI tasks. Gemma 2 is designed to run efficiently on a single GPU or TPU host, and it is compatible with major AI frameworks like Hugging Face Transformers, JAX, PyTorch, and TensorFlow. It is available in 9 billion and 27 billion parameter sizes.
Comments
Post a Comment