Meta's new model

Plus: OpenAI partners with Scale, Nat Friedman and Daniel Gross interview

August 25, 2023

Hello folks, here’s what we have today;

 Our picks 

1/ Meta has released Code Llama, a large language model optimized for generating code. It is available in three sizes - 7B, 13B, and 34B parameters - each trained on 500B tokens of code data. Benchmark testing showed Code Llama outperforms other publicly available code-specific LLMs like Codex and AlphaCode, and is on par with ChatGPT.

Mysteriously, there’s a secret, unreleased model, “Unnatural Code Llama” mentioned that crushed every other model…

p.s. Code LLaMA is now on Perplexity’s LLaMa Chat

2/ OpenAI partners with Scale to provide support for enterprises fine-tuning models.

We’re working with Scale as a preferred partner to extend the benefits of our fine-tuning capability given their experience helping enterprises securely and effectively leverage data for AI. Building robust enterprise-grade functionality requires rigorous data enrichment and model evaluation. Scale customers can now fine-tune OpenAI models just as they would through OpenAI, while also benefiting from Scale’s enterprise AI expertise and Data Engine.

3/ An Interview with Daniel Gross and Nat Friedman about the AI Hype Cycle (paywalled). Ben Thompson has released his third interview with the duo, I’ve enjoyed each interview enormously.

  • Nat and Daniel discuss the current AI hype cycle: There is more noise but also real traction in enterprise use cases automating human labour.

  • They see the model ecosystem diversifying into generalist, high capability, and specialized expert models suited for different jobs. Enterprises will likely evolve from exploring general models to deploying tailored solutions based on factors like privacy, tuning needs, and performance.

  • This landscape creates infrastructure opportunities in model evaluation, training/deployment, augmentation, and operational tooling.

  • They see local models complementing cloud models, and opportunities in synthetic data and curriculum learning to address data constraints.

from our sponsor

Scale is excited to be OpenAI’s preferred partner for fine-tuning GPT-3.5! Our partnership:

  • Combines GPT-3.5 with Scale's fine-tuning expertise to create custom state-of-the-art models

  • Delivers for customers like Brex, who use a fine-tuned model to generate highly accurate expense memos

  • Enables us to create powerful custom models that increase efficiency and reduce costs for Enterprise customers

Harness the power of your data and supercharge your custom LLMs.

Learn More Today.

 From the community 
 Cool Tools  trending product launches from the last 24 hours

  • Unriddle - a tool to help you read, write and learn faster.

  • Genesis AI - Build your ChatGPT custom plugin without code,

    in less than 2 minutes.

  • Scenario Unity Plugin - Uniting Unity developers and AI-powered game asset generation.

  • Pastors AI - Chat with your Pastor’s sermons.

  • Lilac - An open-source tool that enables AI practitioners to see and quantify their datasets.

  • Wait is that a sticker - AI-generated stickers.

  • AI color - Colorize your old photos with the power of AI.

  • Tweeets - A tool for analyzing and creating tweets to make your life easier as a regular Twitter user.

  • Dicer.ai - Optimize your ad creatives and campaign strategies 10x faster.

  • Katch - An AI Assistant that takes care of scheduling your meetings, right from your email.

  • Altero AI - Access detailed business research on any company.

  • FineShare - Create song covers with your favorite AI vocals in just one click.

  • Watto AI - Transform your product team with AI-powered docs.

  • Fabric browser extension – Rediscover your notes and files as you browse.

 Ben’s Bites News  top posts from the last 24 hours

Ben’s Bites Insights

We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;

  • All 10k+ links we’ve covered, easily filterable (1 referral)

  • 6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)

Reply

or to participate.