Welcome insideAI News AI News Briefs, our timely new feature bringing you the latest industry insights and perspectives surrounding the field of AI including deep learning, large language models, generative AI, and transformers. We’re working tirelessly to dig up the most timely and curious tidbits underlying the day’s most popular technologies. We know this field is advancing rapidly and we want to bring you a regular resource to keep you informed and state-of-the-art. Enjoy!
AI21 Labs, a GenAI tech startup is gaining adoption for enterprise AI infrastructure and used by eBay, Monday.com, Carrefour and Ubisoft. The Israeli company that develops advanced AI technologies to solve complex problems across various industries has raised $155 million.
A few days ago, Gartner released the 2023 Hype Cycle for emerging technologies as shown below. No surprises here.
Supercharge AI training and inference workloads with Latitude.sh Accelerate. Using NVIDIA H100 GPUs, Latitude Accelerate speeds up AI and machine learning tasks, making both training and running models faster and more efficient. With dedicated instances, 32-cores/GPU and hourly billing, Accelerate offers unmatched performance and flexibility, all at the best cost per GPU on the market.
GPT Pilot is a research project to see how can GPT-4 be utilized to generate fully working, production-ready, apps. The main idea is that AI can write most of the code for an app (maybe 95%) but for the rest 5%, a developer is and will be needed until we get full AGI. Here are the steps GPT Pilot takes to create an app:
Prompt2Model – Generate Deployable Models from Instructions – Open-source project which allows developers to train deployable, special-purpose NLP models using natural language task descriptions. The method combines dataset retrieval, LLM-based dataset generation, and supervised fine-tuning.
WizardLM aims to improve large language models (LLMs) by generating complex instruction data using LLMs rather than manual human input. The model uses a method called Evol-Instruct to evolve simpler instructions into more complex ones for fine-tuning.
At last week’s Google Cloud Next event in San Francisco, Google made a surprise announcement that the company is offering Llama 2 as well as Falcon LLM on Google Cloud’s Vertex AI. This move was not expected since Google was the sole cloud provider that hadn’t partnered with rival institutions to host Llama 2 or any other open-source LLM models before this. It appears that this decision by Google has taken into account enterprises that are looking for more options. Following this trend, after GPT-4, Llama 2 is the most sought-after large language model, considering it is open-sourced and commercially available. In the case of Llama 2, Google said that it is the only cloud provider offering both adapter tuning and RLHF.
Additional info! SeamlessM4T (Massive Multilingual Multimodal Machine Translation) by Meta is the a multimodal model representing a significant breakthrough in speech-to-speech and speech-to-text translation and transcription. Publicly-released under a CC BY-NC 4.0 license, the model supports nearly 100 languages for input (speech + text), 100 languages for text output and 35 languages (plus English) for speech output. It aims to eliminate reliance on multiple models by unifying capabilities into a single one. It can handle:
- 101 languages for speech input
- 96 Languages for text input/output
- 35 languages for speech output
The model achieves state-of-the-art results by leveraging Fairseq2, the largest open dataset for multimodal translation, and other advancements. It reduces toxicity and bias compared to previous models. This unified model enables multiple tasks without relying on multiple separate models:
- Speech-to-speech translation (S2ST)
- Speech-to-text translation (S2TT)
- Text-to-speech translation (T2ST)
- Text-to-text translation (T2TT)
- Automatic speech recognition (ASR)
OpenAI partners with Scale for GPT-3.5 fine-tuning and advanced data labeling, allowing you to unlock the full potential of GPT by adapting the model to your own data. Companies like Brex are already using the platform to optimize their business and model performance. Scale’s high-quality Data Engine and custom LLM platform will help:
- Build custom LLMs that fits your business needs
- Create powerful custom models that increase efficiency and reduce costs
- Benefit from Scale’s fine-tuning and data preparation platform
- Optimize your AI investment
- Make your AI work for you, not the other way around
LlamaGPT is a self-hosted, offline chatbot that offers a private, ChatGPT-like experience.. The project is a culmination of open-source contributions from various developers.
AI2 releases the largest open source text dataset for LLM pretraining – Dolma is a 3 trillion-token dataset that sets a new standard for openness in language model research.
Hugging Face raises $235M series D at $4.5B valuation – The round received contributions from major players including Google, Amazon, NVIDIA, Salesforce, AMD, Intel, IBM, and Qualcomm. The funds will be allocated towards talent acquisition.
Chip Huyen Outlines 10 Open Challenges for LLMs – Chip Huyen, a prominent figure in AI research, shed light on the top 10 challenges facing Large Language Model (LLM) development in her recent blog post that gained significant attention.
- Hallucinations: Minimizing AI’s creation of inaccurate data.
- Context Mastery: Enhancing LLMs’ understanding of context.
- Data Modalities: Incorporating diverse data types, like text and images.
- Efficiency: Boosting LLMs’ speed and affordability.
- Architecture Evolution: Innovating beyond current model designs.
- Beyond GPUs: Seeking alternatives to dominant AI training hardware.
- AI Agents: Crafting LLMs for real-world tasks.
- Human Preferences: Refining models based on human feedback.
- Chat Interfaces: Streamlining user interactions with LLMs.
- Multilingualism: Expanding LLMs to non-English languages.
Addressing these challenges is crucial for the next generation of LLMs. As AI becomes an integral part of various sectors, solving these issues will determine its future utility and impact. Chip Huyen’s insights provide a roadmap for researchers and industry professionals in the AI domain.
Sign up for the free insideAI News newsletter.
Join us on Twitter: https://twitter.com/InsideBigData1
Join us on LinkedIn: https://www.linkedin.com/company/insidebigdata/
Join us on Facebook: https://www.facebook.com/insideAI NewsNOW