Have you ever wanted to do an experiment on LLMs and found that none of the existing model suites met your needs?

Eleuther AI designed a model suite that centers enabling scientific research as its primary goal.

Does One Large Model Rule Them All?

Predictions on the Future AI Ecosystem

Does the rise of large, general AI models means the future AI ecosystem is dominated by a single general model?

Singapore central bank and police have been helping banks to fine-tune their vetting approach when opening accounts for service providers in all types of digital assets.

The project has been ongoing for about six months, and an industry report outlining best practices in areas like due diligence and risk management may be published in the next two months.

Researchers at University of the Witwatersrand in Johannesburg, South Africa used an EMOTIV device to link a brain to the internet similar to IoT.

Dubbed Brainternet, the BCI project aims to "simplify a person's understanding of their own brain and the brains of others" through machine learning and interactivity.

Anthropic’s has a $5B, 4-year plan to take on OpenAI.

It plans to build a “frontier model” called “Claude-Next” 10 times more capable than GPT, but will require billions in spending over the next 18 months.

Google owns a 10% stake in the startup.

Global unicorn club by Cb Insights

The striking thing about this list, to me, is the preponderance of software companies.

MetaMask announced the launch of fiat currency purchase cryptocurrency function in its Portfolio Dapp, users can use debit or credit card, PayPal, Bank Transfer, or Instant ACH to buy crypto.

Bitcoin has had $3T in trading volume over the last 90 days, more than 3x Ethereum.

LLMs just hit a major milestone with the release of the new "Generative agents" paper.

By using LLMs, generative agents were able to simulate human-like behavior in an interactive sandbox inspired by The Sims.

The agent architecture extends Language Models to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior.

There are three components to it:
1. The memory stream, which records a comprehensive list of the agent's experiences

2. Reflection, which synthesizes memories into higher-level inferences over time

3. Planning, which translates those conclusions and the current environment into high-level action plans.

Paper: https://arxiv.org/abs/2304.03442
Demo: https://reverie.herokuapp.com/arXiv_Demo/

OpenAGI: When LLM Meets Domain Experts

OpenAGI is an open-source AGI research platform designed to offer complex, multi-step tasks.

OpenAGI formulates complex tasks as natural language queries serving as input to the Large Language Models.

The LLM subsequently selects, synthesizes, and executes models provided by Open AGI to address the task.

Paper https://arxiv.org/abs/2304.04370

GitHub https://github.com/agiresearch/OpenAGI

A16Z crypto today realized the 2023 State of Crypto Report and introduce the State of Crypto Index.

One of the key insights that you’ll find from the report is that blockchains have more active users, and more ways to engage than ever before.

More monthly active addresses – unique addresses sending on-chain transactions each month – than ever.

Last month 15 million addresses, more than twice as many as two years ago when prices were still elevated.

One explanation: there are increasingly more ways to engage with blockchains and web3 applications.

From DeFi to web3 games – more than 700 of which launched last year – a variety of new applications create addresses for their users to interact with, without having to download or connect a wallet.

AutoGPT might be the next big step in AI

Here's why Karpathy, OpenAI, recently said "AutoGPT is the next frontier of prompt engineering".

AutoGPT is the equivalent of giving GPT-based models a memory and a body. You can now give a task to an AI agent and have it autonomously come up with a plan, execute on it, browse the web, and use new data to revise the strategy until the task is completed.

It can analyze the market and come up with a trading strategy, customer service, marketing, finance, or other tasks that requires continuous updates.

There are three components to it:

1. Architecture: It leverages GPT-4 and GPT-3.5 via API.

2. Autonomous Iterations: AutoGPT can refine its outputs by self-critical review, building on its previous work and integrating prompt history for more accurate results.

3. Memory Management: Integration with Pinecone allows for long-term memory storage, enabling context preservation and improved decision-making.

4. Multi-functionality: Capabilities include file manipulation, web browsing, and data retrieval, distinguishing AutoGPT from previous AI advancements by broadening its application scope.

Baby-AGI is taking the world by storm

Here's an implementation within the LangChain framework, allowing you to easily substitute in other vectorstores and other LLMs.

DeepSpeed Chat. Impressive open-source effort by Microsoft!

DeepSpeed Chat offers an end-to-end RLHF pipeline to train ChatGPT-like models. This is the missing piece from other efforts like Alpaca and Vicuna. The RLHF pipeline is replicated from the InstructGPT paper.

The other challenges are cost and efficiency. DeepSpeed Chat aims to make this process more accessible and affordable through a unified hybrid engine (DeepSpeed-HE) for RLHF. For example, DeepSpeed-HE can train an OPT-66B model in 2.1 days for $1620.

With further scalability (e.g., using multi-node multi-GPU systems), DeepSpeed-HE can train an OPT-13B model in 1.25 hours for $320 and an OPT-175B model in under a day for $5120. That's a big deal!

Elon Musk Makes Big Investment in Generative-AI Project at Twitter

He recently purchased roughly 10,000 GPUs for the platform

Two AI engineers from Alphabet's DeepMind joined Twitter last month.

Intel plans to make graphics processors (GPUs) that meet US rules for export to China, media report, adding that Intel CEO Pat Gelsinger is in China now meeting government officials.

Current architecture of research agent enables learning through Google Search, Pinecone, multiple agent feedback, and incremental draft improvement.

Based on babyAGI