Matt Wood

New! Introducing PartyRock - Amazon’s new AI playground that lets you experiment, learn, and build apps with generative AI. Let's rock...

🤘 PartyRock is a new app powered by Bedrock that allows you to build your own app in seconds, share it with others, and learn about generative AI in the process.

🎸 In just a few steps, you can build an app to create the perfect playlist, recommend what to cook based on ingredients in your pantry, or organize your re:invent schedule, all while learning how to write text prompts to generate a desired output — in this case, your own app.

🚀 PartyRock is a blast - and will help build intuition around generative AI techniques and capabilities, like understanding different foundation model strengths and chaining prompts together. You can use text and image models, including Claude.

🥳 PartyRock introduces another layer of fun by allowing you to share your apps with others. Other PartyRockers can try it out, remix it to edit or add new features, leading to even bigger, more creative app concepts.

🌶️ I have spent countless hours playing with PartyRock over the past few months - and it is probably one of the single biggest contributors to how I think about, build with, and explore AI concepts.

🏓 Play is one of the best ways to learn anything: it's fun, motivating, active, exploratory, and low stakes. PartyRock lets you learn by playing with generative AI, building something useful and fun for yourself and others along the way.

🙌 Anyone can access PartyRock through its simple web-based UI, which is separate from the AWS console and does not require an AWS account to get started. So fire it up - share what you build here (and I'll do the same!)...

For those about to PartyRock, we salute you! 🤘

#ai #aws #amazon #generativeai #smallsoftware

New! Claude 3 Opus is now available on Bedrock - the first and only managed service to offer all three Claude 3 models in GA. Let's dive in!

When customers build with generative AI, having access to different models lets them mix and match to find the best match for their workload.

With Bedrock, we provide fully-managed access to the broadest set of models - along with tools to help evaluate and select the best one for your use case - and we're expanding this choice today with a real doozy: Claude 3 Opus.

🏆 Claude 3 Opus, the best-performing foundation model (based on multiple industry benchmarks), is now available on Bedrock.

⚡️ Claude 3 Opus is Anthropic's most intelligent model, with best-in-market performance on highly complex tasks. It can navigate open-ended prompts and sight-unseen scenarios with remarkable fluency and human-like understanding. Opus shows us the outer limits of what’s possible with generative AI.

💬 Claude 3 has higher intelligence than any other model, making it great for task automation (plan and execute complex actions across APIs and databases, interactive coding), R&D (research review, brainstorming and hypothesis generation, drug discovery), and strategy (advanced analysis of charts & graphs, financials and market trends, forecasting).

Combined with Bedrock's security, grounding, alignment, and guard rail capabilities - and currently being the only managed service offering all three Claude 3 models – Opus, Sonnet, and Haiku – in GA - there has never been a better time to prompt happy.

Fire it up!

#genai #ai #aws

New! Llama 3 is here, and available on AWS today. Meta's latest models are a major leap over Llama 2. Let's dive in!

🎸 Llama 3 comes in two parameter sizes — 8B and 70B with 8k context length — that can support a broad range of use cases with improvements in reasoning, code generation, and instruction following.

🤘 Llama 3 uses a decoder-only transformer architecture and new tokenizer that provides improved model performance. What does this mean? A better, faster, lower cost model...

🏎️ The tokenizer offers improved token efficiency, yielding up to 15% fewer tokens compared to Llama 2 (with faster, lower cost responses as a result).

😺 Llama 3 8B is a bigger model, with 1 billion more parameters than Llama 2 7B - encoding more concepts and understanding. Bigger models are usually slower, but with more efficient tokenization and group query attention, Llama 3 has equivalent inference latency as the smaller Llama 2. Impressive.

⏯️ Finally, Meta improved post-training procedures that substantially reduced false refusal rates, improved alignment, and increased diversity in model responses.

These new Llama 3 models are a big step forward, the weights are publicly available, and the models are available today in SageMaker.

#llama3 #aws #genai

Project Kuiper - our broadband satellite network - is making exciting progress: first look at customer terminals and new “Prometheus” chips.

First, the customer terminals. These come in three sizes/speeds: ultra compact (just 7 inches square, 100 Mbps), residential and small businesses (11 inches square, 400 Mbps - enough to stream 4K-quality video), and high bandwidth (19 X 30 inches, 1 Gpbs) .

These are the outdoor antennas used to communicate with the low earth orbit satellites, which puts broadband connectivity into the hands of unserved or underserved communities. The key here has been to innovate in reducing the size of the terminals, which in turn lowers the cost, which in turn makes them more accessible to these communities.

This is hard to do in the wide Ka-band frequency range that these devices operate in - where the range determines a very large physical antenna size (which also makes them expensive). The team did some awesome work here, inventing a phased array antenna which overlaps tiny antenna element structures to reduce size (and cost).

Second, Prometheus! A new custom chip which combines the capabilities of a 5G modem, a base station which can handle thousands of customer connections at once, and the ability of a microwave backhaul antenna to support point to point connections.

Prometheus chips are in the customer terminals, satellites, and ground stations which form the Keiper network.

Just a remarkable amount of invention happening in Kuiper - for a very worthy endevour. Really exciting.