Magic raised $117m, which is building a superhuman software engineer.
If a copilot generates $10b of revenue, how much is a colleague worth?
Magic has trained a groundbreaking model with many millions of tokens of context that performed far better in our evals than anything we've tried before.
They're using it to build an advanced AI programmer that can reason over your entire codebase and the transitive closure of your dependency tree.
If a copilot generates $10b of revenue, how much is a colleague worth?
Magic has trained a groundbreaking model with many millions of tokens of context that performed far better in our evals than anything we've tried before.
They're using it to build an advanced AI programmer that can reason over your entire codebase and the transitive closure of your dependency tree.
Magic
Magic is an AI company that is working toward building safe AGI to accelerate humanity’s progress on the world’s most important problems.
Reddit has reportedly signed a $60M annual deal granting an AI firm access to its user-generated content for model training.
Notably, the deal is coming prior to the company’s upcoming IPO.
Notably, the deal is coming prior to the company’s upcoming IPO.
Bloomberg.com
Reddit Signs AI Content Licensing Deal Ahead of IPO
Reddit Inc. has signed a contract allowing a company to train its artificial intelligence models on the social media platform’s content, according to people familiar with the matter, as it nears the potential launch of its long-awaited initial public offering.
BioMistral is a new 7B foundation model for medical domains, based on Mistral and further trained PubMed Central.
- top open-source medical Large Language Model (LLM) in its weight class
- Apache License
- includes base models, fine tunes, and quantized versions.
- top open-source medical Large Language Model (LLM) in its weight class
- Apache License
- includes base models, fine tunes, and quantized versions.
huggingface.co
BioMistral/BioMistral-7B · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Andrej Karpathy was leaving OpenAI 4 days ago, and now he released an implementation of the Byte Pair Encoding algorithm behind GPT and most LLMs.
Byte Pair Encoding: "Minimal, clean, educational code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM tokenization."
Byte Pair Encoding: "Minimal, clean, educational code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM tokenization."
GitHub
GitHub - karpathy/minbpe: Minimal, clean code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM tokenization.
Minimal, clean code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM tokenization. - karpathy/minbpe
Cool new idea from DeepMind.
They evaluate LMs by giving them a piece of code, having them describe it, and then asking the LM to rewrite that code given only the denoscription.
The metric is the similarity between the original code and the rewritten code.
They evaluate LMs by giving them a piece of code, having them describe it, and then asking the LM to rewrite that code given only the denoscription.
The metric is the similarity between the original code and the rewritten code.
👍5
Game changer. You can now visualize your RAG Data.
UMAP is dimensionality reduction techniques that transforms complex, high-dimensional data into a clear and interactive 2D map.
It can also be used for debugging and improving the performance of your RAG models.
UMAP is dimensionality reduction techniques that transforms complex, high-dimensional data into a clear and interactive 2D map.
It can also be used for debugging and improving the performance of your RAG models.
Medium
Visualize your RAG Data — EDA for Retrieval-Augmented Generation
How to use UMAP dimensionality reduction for Embeddings to show Questions, Answers and their relationships to source documents with…
👍3🤡1
🚀 Groq is claiming that its technology could replace GPUs in AI tasks with its powerful chip and software
Groq's LPU is faster than Nvidia GPUs, handling requests and responding more quickly.
Groq's LPUs don't need speedy data delivery like Nvidia GPUs do because they don't have HBM in their system. They use SRAM, which is about 20 times faster than what GPUs use. Since inference runs use way less data than model training, Groq's LPU is more energy-efficient. It reads less from external memory and uses less power than a Nvidia GPU for inference tasks.
The LPU works differently from GPUs. It uses a Temporal Instruction Set Computer architecture, so it doesn't have to reload data from memory as often as GPUs do with High Bandwidth Memory (HBM). This helps avoid issues with HBM shortages and keeps costs down.
If Groq's LPU is used in places that do AI processing, you might not need special storage for Nvidia GPUs. The LPU doesn't demand super-fast storage like GPUs do.
Groq's LPU is faster than Nvidia GPUs, handling requests and responding more quickly.
Groq's LPUs don't need speedy data delivery like Nvidia GPUs do because they don't have HBM in their system. They use SRAM, which is about 20 times faster than what GPUs use. Since inference runs use way less data than model training, Groq's LPU is more energy-efficient. It reads less from external memory and uses less power than a Nvidia GPU for inference tasks.
The LPU works differently from GPUs. It uses a Temporal Instruction Set Computer architecture, so it doesn't have to reload data from memory as often as GPUs do with High Bandwidth Memory (HBM). This helps avoid issues with HBM shortages and keeps costs down.
If Groq's LPU is used in places that do AI processing, you might not need special storage for Nvidia GPUs. The LPU doesn't demand super-fast storage like GPUs do.
Sequoia: Scalable, Robust, and Hardware-aware Speculative Decoding
Improves the decoding speed of Vicuna-33B by up to 2.37x and Llama2-70B offloading speed by up to 10.33x
Improves the decoding speed of Vicuna-33B by up to 2.37x and Llama2-70B offloading speed by up to 10.33x
The Gemini 1.5 Pro model guide is live
Gemini 1.5 Pro is among the most powerful long context LLMs available today.
Gemini 1.5 Pro shows impressive capabilities around multimodal reasoning, video understanding, long document question answering, code reasoning on entire codebases, and in-context learning.
One insight from testing this model is that we will have different kinds of LLMs that support different types of use cases. Gemini 1.5 Pro is not meant to be a model to reign among all. The long context LLMs are not meant to cover every use case imaginable, they are meant to unlock complex use cases that were unimaginable before with LLMs.
Gemini 1.5 Pro is among the most powerful long context LLMs available today.
Gemini 1.5 Pro shows impressive capabilities around multimodal reasoning, video understanding, long document question answering, code reasoning on entire codebases, and in-context learning.
One insight from testing this model is that we will have different kinds of LLMs that support different types of use cases. Gemini 1.5 Pro is not meant to be a model to reign among all. The long context LLMs are not meant to cover every use case imaginable, they are meant to unlock complex use cases that were unimaginable before with LLMs.
www.promptingguide.ai
Gemini 1.5 Pro | Prompt Engineering Guide
A Comprehensive Overview of Prompt Engineering
Mixed_Reality_for_Work_Collaboration_Education_more_1708447794.pdf
14.3 MB
Meta released a guide as were approached by individuals & enterprises that want to take full advantage of the metaverse at hand.
Here's what you'll discover inside:
1. Introduction to Mixed Reality: Understand the basics and the immense potential MR holds for businesses.
2. Impactful use cases: Explore how MR is reshaping industries in areas like product design, construction, specialised training, and more.
3. Meta Quest Headsets Overview: A detailed look at Meta’s MR headsets, helping you choose the right one for your needs.
4. Practical Launch Tips: From choosing apps to setting up hardware, get actionable advice for a successful MR implementation.
5. Meta Quest for Business: Learn about the enterprise-grade solutions for managing MR devices and applications effectively.
Here's what you'll discover inside:
1. Introduction to Mixed Reality: Understand the basics and the immense potential MR holds for businesses.
2. Impactful use cases: Explore how MR is reshaping industries in areas like product design, construction, specialised training, and more.
3. Meta Quest Headsets Overview: A detailed look at Meta’s MR headsets, helping you choose the right one for your needs.
4. Practical Launch Tips: From choosing apps to setting up hardware, get actionable advice for a successful MR implementation.
5. Meta Quest for Business: Learn about the enterprise-grade solutions for managing MR devices and applications effectively.
👍1
Hot off the press from Andrej karpathy 🔥
Everything you need to know about LLM tokenization. If you’re a developer using AI, you are going to want to watch this.
Everything you need to know about LLM tokenization. If you’re a developer using AI, you are going to want to watch this.
YouTube
Let's build the GPT Tokenizer
The Tokenizer is a necessary and pervasive component of Large Language Models (LLMs), where it translates between strings and tokens (text chunks). Tokenizers are a completely separate stage of the LLM pipeline: they have their own training sets, training…
🆒3
Exclusive deepdive on how ChatGPT scaled to 100M weekly users: and five engineering challenges the team faced and how they are tackling it.
Pragmaticengineer
Scaling ChatGPT: Five Real-World Engineering Challenges
Just one year after its launch, ChatGPT had more than 100M weekly users. In order to meet this explosive demand, the team at OpenAI had to overcome several scaling challenges. An exclusive deepdive.
❤4
Google released Gemma, the most powerful open LLM yet.
Open for commercial use, it outperforms Mistral AI 7B and LLaMa 2 on Human Eval and MMLU.
It's the first open LLM based on Gemini.
Details:
- Comes in two flavors: 2B and 7B.
- Beats Mistral 7B, DeciLM 7B and Qwen1.5 7B
- Instruction models in 2B and 7B variants.
- 8192 Default context window.
- MMLU score of 64.56, average leaderboard score 63.75 for 7B.
-2B model compatible with mobile phones.
Available on HuggingFace, Kaggle and Vertex AI.
Open for commercial use, it outperforms Mistral AI 7B and LLaMa 2 on Human Eval and MMLU.
It's the first open LLM based on Gemini.
Details:
- Comes in two flavors: 2B and 7B.
- Beats Mistral 7B, DeciLM 7B and Qwen1.5 7B
- Instruction models in 2B and 7B variants.
- 8192 Default context window.
- MMLU score of 64.56, average leaderboard score 63.75 for 7B.
-2B model compatible with mobile phones.
Available on HuggingFace, Kaggle and Vertex AI.
Google AI for Developers
Gemma open models | Google AI for Developers
Открытые модели Gemma созданы на основе тех же исследований и технологий, что и модели Gemini. Gemma 2 выпускается в размерах 2B, 9B и 27B, а Gemma 1 — в размерах 2B и 7B.
❤3
An excellent article on the progression of the academic BCI field.
A must read if you want to know what is going on with BCIs.
A must read if you want to know what is going on with BCIs.
Nature
Mind-reading devices are revealing the brain’s secrets
Nature - Implants and other technologies that decode neural activity can restore people’s abilities to move and speak — and help researchers to understand how the brain works.
Researchers released UMI Gripper, an open source "Universal Manipulation Interface" hardware for collecting demonstration data that is over 3x faster than human teleoperation.
In the video below, you can see data being collected by humans simply holding the gripper. This data was then used to train policies which could autonomously complete the same tasks.
In the video below, you can see data being collected by humans simply holding the gripper. This data was then used to train policies which could autonomously complete the same tasks.
umi-gripper.github.io
Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots
We present Universal Manipulation Interface (UMI) -- a data collection and policy learning framework that allows direct skill transfer from in-the-wild human demonstrations to deployable robot policies. UMI employs hand-held grippers coupled with careful…
Meta presents MobileLLM
Paper addresses the growing need for efficient large language models (LLMs) on mobile devices, driven by increasing cloud costs and latency concerns.
Paper addresses the growing need for efficient large language models (LLMs) on mobile devices, driven by increasing cloud costs and latency concerns.
VC_Industry_Outlook_Insights_Predictions_1708937742.pdf
39.2 MB
This report by Allocate breaks down insights and predictions from top VC managers and limited partners about VC industry in 2024.
Here are key takeaways:
1. Fundraising's going to be tough in 2024, especially for new or average VC funds, so expect consolidation and exits as a result.
2. LPs will get choosy, doubling down on top tier VCs with strong records, while also hunting for more early stage action via new relationships.
3. Many late-stage startups will hit a wall in 2024 as their runway shortens —get ready for shutdowns, acqui-hires, and down rounds.
4. The VC scene will see more layoffs and team shakeups in 2024 as mediocre GPs get exposed by the correction.
5. M&A dealing mid-stage startups will pick up in 2024 as cheaper debt opens up an exit route pre-IPO scale.
6. AI/ML startups will struggle to profit and cover high computing bills, but adoption will accelerate across industries.
7. More founders will look to raise only a Series A, then focus on profitability. Avoiding future dilution and growth rate pressures.
8. With few exits in 2024, VCs and startups will get creative chasing liquidity via secondaries, mini IPOs, and direct listings.
Here are key takeaways:
1. Fundraising's going to be tough in 2024, especially for new or average VC funds, so expect consolidation and exits as a result.
2. LPs will get choosy, doubling down on top tier VCs with strong records, while also hunting for more early stage action via new relationships.
3. Many late-stage startups will hit a wall in 2024 as their runway shortens —get ready for shutdowns, acqui-hires, and down rounds.
4. The VC scene will see more layoffs and team shakeups in 2024 as mediocre GPs get exposed by the correction.
5. M&A dealing mid-stage startups will pick up in 2024 as cheaper debt opens up an exit route pre-IPO scale.
6. AI/ML startups will struggle to profit and cover high computing bills, but adoption will accelerate across industries.
7. More founders will look to raise only a Series A, then focus on profitability. Avoiding future dilution and growth rate pressures.
8. With few exits in 2024, VCs and startups will get creative chasing liquidity via secondaries, mini IPOs, and direct listings.
👍5
Mistral just released a new closed-source LLM available on their Platform or Azure
That's quite a step away from their initial message and position to release an open-source GPT-4 level model in 2024.
Model:
1. Multilingual with English, French, Spanish, German, and Italian
2. 32K tokens context window
3. Supports function calling and JSON mode
4. +80% on MMLU, closely behind GPT-4 or Gemini Ultra
5. Not open, only available on Mistral or Azure
That's quite a step away from their initial message and position to release an open-source GPT-4 level model in 2024.
Model:
1. Multilingual with English, French, Spanish, German, and Italian
2. 32K tokens context window
3. Supports function calling and JSON mode
4. +80% on MMLU, closely behind GPT-4 or Gemini Ultra
5. Not open, only available on Mistral or Azure
mistral.ai
Au Large | Mistral AI
Mistral Large is our flagship model, with top-tier reasoning capacities. It is also available on Azure.
Google DeepMind dropped Genie an AI that can generate interactive video games
Genie is trained on 200,000 hours of unsupervised public internet gaming videos and can generate video games from a single prompt or image.
Despite not being trained on action or text annotations, the foundation model can determine who the main character is and enable a user to control that character in the generated world.
It does this through its Latent Action Model, Video Tokenizer, and Dynamics Model (will go more in-depth on this in tomorrow's newsletter for those interested).
It's research-only and not publicly available (here come the Google memes), and it does come with some limitations, like only currently creating games at 1FPS.
Genie’s model is general and not constrained to 2D.
Google also train a Genie on robotics data (RT-1) without actions, and demonstrate that we can learn an action controllable simulator there too.
Google think this is a promising step towards general world models for AGI.
Paper here.
Genie is trained on 200,000 hours of unsupervised public internet gaming videos and can generate video games from a single prompt or image.
Despite not being trained on action or text annotations, the foundation model can determine who the main character is and enable a user to control that character in the generated world.
It does this through its Latent Action Model, Video Tokenizer, and Dynamics Model (will go more in-depth on this in tomorrow's newsletter for those interested).
It's research-only and not publicly available (here come the Google memes), and it does come with some limitations, like only currently creating games at 1FPS.
Genie’s model is general and not constrained to 2D.
Google also train a Genie on robotics data (RT-1) without actions, and demonstrate that we can learn an action controllable simulator there too.
Google think this is a promising step towards general world models for AGI.
Paper here.
Google
🧞 Genie: Generative Interactive Environments
A Foundation Model for Playable Worlds
🔥5👍3
Lenovo unveiled Project Crystal, a fully functioning transparent laptop.
It uses a transparent micro-LED display that offers a literal clear view of what’s behind it.
It uses a transparent micro-LED display that offers a literal clear view of what’s behind it.
WIRED
Lenovo’s Project Crystal Is a Concept Laptop With a Transparent Display
You can see clearly now through the Project Crystal. But what is it for? And is a transparent phone next?