. . AI: Survey on AI adoption, faster, bigger, smarter, more open AI tools (11.13.23) Post

Friends, here are the 5 news stories rising to the top of the AI community conversation over the weekend and Monday.

Once again, the day’s news is like a fractal snapshot of the same larger trends and interesting new developments that unfold day after day, week after week. Pick one of these stories and imagine where it will lead in 5 years. I think they’re all pretty illustrative examples.

Thanks to everyone helping spread the word about this newsletter. I hope it’s useful to you.

Cheers,

Marshall Kirkpatrick, Editor

Survey: Tech Pros Say AI is Overrated, but Still Valuable and Important

First impacted: Professionals broadly
Time to impact: Short

Software development platform Retool surveyed 1500 people in tech, across industries, about their companies' AI adoption and use cases. More than half of those surveyed consider AI to be overrated, but most also said internal use cases are delivering value (36.8% described them as “extremely useful,” 42.9% said “somewhat useful,” and 16.8% said “a little useful.” Just 3.4% said “not very useful” or worse) and most said AI would impact their role significantly (8/10) over the next 5 years. [Retool State of AI Report] Share by email

Together AI Launches High-Speed Inference Engine

First impacted: AI developers, Cloud computing engineers
Time to impact: Short

Together AI ("the fastest cloud for gen AI") has launched the Together Inference Engine, which they claim is the world's fastest inference stack, reportedly up to 3X faster than TGI or vLLM on the same hardware. The company also announced new features, including Serverless Endpoints hosting over 50 leading open-source models and Dedicated Instances, and a price reduction for Serverless Endpoints for 70B models, attributing these changes to the improved performance of their inference engine. [Announcing Together Inference Engine – the fastest inference available] Share by email

LangChain Announces OpenGPT Platform Updates

First impacted: AI developers, OpenGPTs platform users
Time to impact: Short

OpenGPT, LangChain's open-source competitor to OpenAI's new make-your-own GPTs platform, has launched several updates, including the ability to upload files to a retrieval tool with full configurability. The updates also allow users to share public bots they've created and there's a free hosted preview anyone can try here. https://opengpts-example-vz4y4ooboq-uc.a.run.app/ [via @LangChainAI] Share by email

Researchers Launch JARVIS-1, a Multitasking Minecraft Agent

First impacted: AI agent developers
Time to impact: Medium

A team of researchers has introduced JARVIS-1, an AI agent designed for the Minecraft universe, which they claim can progressively improve task completion, achieving a 12.5% completion rate in the long-horizon diamond pickaxe task, a five-fold increase compared to previous records. Built on pre-trained multimodal language models, JARVIS-1 maps visual observations and textual instructions to plans, demonstrating nearly perfect performances across over 200 varying tasks from the Minecraft Universe Benchmark, and is said to have the ability to self-improve following a life-long learning paradigm, thanks to its multimodal memory. [JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Models] Share by email

NVIDIA Announces Supercharged AI Computing Platform

First impacted: Cloud service providers, AI model trainers
Time to impact: Short to medium

NVIDIA has announced the launch of its HGX H200, an upgraded version of its AI computing platform, Hopper, which is expected to be available from global system manufacturers and cloud service providers starting in Q2 2024. The H200, the first GPU to offer HBM3e, a larger, faster memory designed to accelerate AI and large language models, boasts 141GB of memory at 4.8 terabytes per second, nearly double the capacity and 2.4x more bandwidth than its predecessor, the NVIDIA A100. [NVIDIA Supercharges Hopper, the World’s Leading AI Computing Platform] Share by email

That’s it! More AI news tomorrow!