Automate Phone Calls with Synthflow AI

• Always-on AI voice assistants to automate your calls.

• Book appointments, transfer calls, and extract valuable info.

• Integrates with your CRM, easy setup, no coding required.

Start with a free trial

Today’s top AI Highlights:

1. Open Mixture-of-experts - New top-performing AI Model

2. Build and deploy powerful AI Agents quickly with LangGraph.js

3. Your spreadsheets now have an army of AI Researchers

4. Ilya Sutskever’s Safe Superintelligence raises $1 billion

5. AI agents that automate your software development

& so much more!

Read time: 3 mins

Latest Developments

Achieve More with Fewer Parameters 🔥

OLMoE is a new, fully open Mixture-of-Experts language model with 7B parameters. It utilizes only a fraction of its parameters per input, making it comparable in inference cost to smaller models while achieving state-of-the-art performance. It outperforms all models with similar active parameters, surpassing even larger ones like Llama2-13B-Chat and DeepSeekMoE-16B. The release includes model weights, training data, code, and a detailed report.

Key Highlights:

1. Efficiency - OLMoE-1B-7B only uses 1.3B of its 7B parameters per input, keeping the inference cost similar to a 1B parameter dense model while outperforming models like Llama2-13B.

2. Performance - OLMoE surpasses open 1B models and even larger, closed models like DeepSeekMoE-16B in benchmarks such as MMLU, GSM8k, and HumanEval.

3. Openness - Fully open-source, including model weights, training data, code, and logs. If you’re building AI models that need top-tier performance without breaking the bank in inference costs, OLMoE is a game-changer.

4. Get Started - Here’s a quick way to load and use the OLMoE model in your own projects. Install the transformers & torch libraries and run (Transformers must be from the source for this PR or until the next release):

# Import the necessary modules
from transformers import OlmoeForCausalLM, AutoTokenizer
import torch

# Set up the device (GPU or CPU)
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

# Load the model and tokenizer from Hugging Face Hub
model = OlmoeForCausalLM.from_pretrained("allenai/OLMoE-1B-7B-0924").to(DEVICE)
tokenizer = AutoTokenizer.from_pretrained("allenai/OLMoE-1B-7B-0924")

# Tokenize the input text
inputs = tokenizer("Bitcoin is", return_tensors="pt")
inputs = {k: v.to(DEVICE) for k, v in inputs.items()}

# Generate output using the model
out = model.generate(**inputs, max_length=64)

# Decode and print the output
print(tokenizer.decode(out[0], skip_special_tokens=True))

Build Reliable AI Agents in JavaScript 🤖

LangChain has released LangGraph.js v0.2.0 giving JavaScript/TypeScript developers new tools to build, debug, and deploy LLM-powered agents quickly. This version includes flexible streaming options for real-time interactions, a built-in checkpointing system for debugging, and human-in-the-loop support for greater control.

Key Highlights:

1. Streaming and Checkpointing - Stream real-time results and use checkpointing to rewind and debug faulty model responses quickly.

2. Human-in-the-loop and Parallel Nodes - Pause and adjust your agents’ internal states during execution or run multiple nodes simultaneously for greater efficiency and flexibility.

3. Deploy and scale seamlessly - LangGraph.js projects can now be used with LangGraph Studio, a desktop agent IDE for visualizing AI agents, and LangGraph Cloud, a scalable deployment platform.

4. Build your first AI Agent - Check out this GitHub repo with an AI agent example built with LangGraph.js. Install the required dependencies, set your API keys, and follow the steps in the repo to build your first AI agent in no time!

Quick Bites

1. Humbling spreadsheets got supercharged by a swarm of AI agents. AI startup Paradigm is introducing intelligent AI agents to find and populate information in your spreadsheets.

   • It is 1000x faster than manual data collection, completing an average of 500 cells per minute.

   • Just tell Paradigm what research you want to do, be it about companies’ information or social media stats. Paradigm's AI agents suggest columns and search, extract, and analyze information from various online sources and databases to create the spreadsheet.

   • Each cell in your Paradigm spreadsheet has a built-in AI engine for further data processing within individual cells. You can join the waitlist here.

2. AI startup Ligo Biosciences has released an opensource implementation of Google DepMind’s AlphaFold 3, under Apache 2.0 license, for efficient protein structure prediction with single-chain protein support and upcoming features for ligands and nucleic acids.
   

3. Anthropic has released Claude Enterprise plan offering an expanded 500K context window, more usage capacity, native GitHub integration, and other enterprise-grade features like managing user permissions and tracing audit logs.
   

4. OpenAI’s co-founder and ex-chief scientist Ilya Sutskever’s new AI startup Safe Superintelligence has raised $1 billion to develop safe artificial intelligence systems. The company plans to use the funds to acquire computing power and hire top talent.
   

5. After facing backlash on the Recall feature in Copilot + PCs, Microsoft had to backpedal and delay its release. They earlier announced that Recall will be made an opt-in feature, not a default one. However, Microsoft has now confirmed that the option of turning off Recall in Windows 11 is a bug and will be fixed in an upcoming update.

Tools of the Trade

1. Honeycomb: Use AI agents to automate tasks in the software development process, like bug fixing, code reviews, and knowledge retrieval. It integrates with platforms like GitHub and Slack.

2. Dot: Opensource app to seamlessly interact with documents and files using local LLMs and RAG. It uses Phi-3 LLM by default, and supports multiple file types.

3. Pipeshift AI: A cloud platform to fine-tune and run opensource LLMs more efficiently. Create specialized models for production with lower latencies, higher accuracy, and have full control over the models you use.

4. Awesome LLM Apps: Build awesome LLM apps using RAG to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos through simple text. These apps will let you retrieve information, engage in chat, and extract insights directly from content on these platforms.

Hot Takes

1. If you think Cursor + Sonnet 3.5 is a "junior engineer," you need to raise your standards. ~
   Santiago
   

2. business school should be a year of v0, cursor, replit, claude prompting, design 101, finance 101, how to find startup ideas and 3 years of how to hack distribution.
   you can't graduate unless one of your products gets 100 customers and you go viral. ~
   Greg Isenberg

Meme of the Day

That’s all for today! See you tomorrow with more such AI-filled content.

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!

Unwind AI - Twitter | LinkedIn | Instagram | Facebook

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one (or 20) of your friends!