Good morning. It’s Wednesday, May 1st.

Did you know: 55 years ago today, AMD was founded.

In today’s email:

• Llama’s 1M Token Context

• ChatGPT For CRISPR

• “gpt2 chatbot”

• ChatGPT Memory

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Llama-3 8B Can Now Understand Up to 1 Million Tokens

Gradient AI introduces Llama-3 8B-Instruct-Gradient-1048k, boasting an impressive capability to handle context lengths of up to 1 million tokens. Through sophisticated techniques such as NTK-aware initialization and progressive training, this model is finely tuned for efficient operation on extensive GPU clusters.

Trained on a corpus of 1.4 billion tokens, Llama-3-8B-Instruct-Gradient-1048k marks a significant advancement in contextual understanding.

The expanded context window of this model presents exciting possibilities. It enables tasks like summarizing entire books, answering complex questions with extensive background knowledge, and conducting in-depth analysis across vast textual domains. This development propels natural language processing forward, offering new avenues for exploration and innovation in the field. Read more.

‘ChatGPT for CRISPR’ Creates New Gene Editing Tools

Researchers are utilizing generative AI to engineer novel CRISPR gene-editing systems, potentially transforming genetic manipulation. Through the utilization of protein language models trained on extensive biological datasets, they have developed AI-generated CRISPR proteins and guide RNAs.

Ali Madani's team at Profluent has developed ProGen, a protein language model, which was retrained to generate diverse CRISPR systems. Testing over 200 AI-designed CRISPR–Cas9 proteins in human cells revealed efficacy comparable to natural enzymes, with fewer off-target effects.

Additionally, Brian Hie and Patrick Hsu's team at the Arc Institute trained the EVO model on 80,000 microbial genomes, yielding promising CRISPR designs. These AI-engineered tools offer precision and versatility, holding the potential for breakthroughs in medicine and biotechnology. Read more.

Is OpenAI Testing GPT-4.5? "gpt2-chatbot" Writes Better Code Than GPT-4 and Claude  

The appearance and subsequent removal of the "gpt2-chatbot" on the LMSYS Org platform stirred speculation within the AI community due to its unattributed origin and capabilities similar to leading models such as OpenAI's GPT-4.

It has showcased remarkable coding skills, surpassing even GPT-4 Turbo, according to experts like Professor Ethan Mollick and engineer Chase McCoy. Despite its brief availability, the chatbot's performance in tasks like game emulation and complex problem-solving garnered attention.

While speculation surrounds its connection to OpenAI as a potential precursor to GPT-4.5 or an independent innovation, definitive evidence is lacking, prompting a guarded sense of optimism amid occasional deviations from the reliability of GPT-4 Turbo. Read more.

OpenAI introduces "Memory" feature for ChatGPT Plus users

OpenAI introduces the "Memory" feature for ChatGPT Plus users, allowing the model to retain information across chats, enhancing its conversational capabilities. Users can enable Memory in the settings and instruct ChatGPT to remember specific details, influencing its responses.

Notably, Memory aims to personalize user interactions, learning individual preferences and styles to improve efficiency.

While currently available to ChatGPT Plus users, its rollout excludes Europe and Korea. OpenAI plans to extend Memory to Team and Enterprise customers, with technical implementation and privacy details yet to be disclosed, underscoring ongoing development in AI personalization and user interaction. Read more.

🖇️ Etcetera: Stories you may have missed

> China's S1 robot displays 'human-like' speed and precision 

> Google sued by US artists over AI image generator

> NIST announces new initiative to create systems that can detect AI-generated content

> China’s Vidu Challenges Sora with High-Definition 16-Second AI Video Clips in 1080p

> Microsoft to invest $1.7 billion in cloud, AI in Indonesia, CEO says

> Australia signs deal worth almost $1b with PsiQuantum to build world's first 'useful' quantum computer

> Financial Times signs licensing deal with OpenAI

> Google to pay up to $6M to News Corp for new AI content, The Information reports  

> Newspapers Sue Microsoft, OpenAI for Using Content to Train AI

> Amazon Q, a generative AI-powered assistant for businesses and developers, is now generally available

> Amazon sales soar with boost from artificial intelligence and advertising

> Apple Beefs up AI Talent Pool by Recruiting From Google

> GPT-4 can outperform humans at reframing negative situations, study finds

> The power of persuasion: Google DeepMind researchers explore why gen AI can be so manipulative  

> Smartphones will be obsolete in 10 years, says Meta's AI Chief

PlayAI is a conversational voice AI platform, offering real-time, human-like voice agents, enhancing contextual, fluid conversations with turn-taking, interruption handling, and emotion modulation.

ProfessorAI is a personalized AI tutor, addresses costly tutoring and comprehension issues through custom chatbots and libraries, helping in standardized tests.

Intellecta employs AI for enhanced customer experiences, utilizing GPT models to handle inquiries across chat, social media, and email platforms.

Brandblast is an AI-driven social media automation platform for efficient content creation, scheduling, and posting. Enhances online presence, engagement, and brand visibility.

Happily.ai is an AI-powered platform for workplace engagement, recognition, and performance management. It addresses employee disengagement, turnover, and conflicts, offering real-time insights and analytics.

Outerbase enhances database interaction via AI-driven navigation, SQL/NLP queries, and visualizations. Features include native apps, diverse DB support, and AI Data Studio.

Effie is a versatile writing tool, employs outlining, mind-mapping, and AI support for enhanced creativity. Features include multi-platform support, offline capability, and cloud synchronization.

BlogFox utilizes Anthropic's Claude Opus model to efficiently convert YouTube videos into SEO-optimized blog posts with custom images, internal/external links, and retained writing style.

SecBrain utilizes AI for organizing ideas, meetings, and video content with automatic summarization and advanced search.

Otterly.AI offers AI-powered brand and content monitoring for AI-driven search experiences like GPT, Gemini, and Bing Copilot, aiding SEO optimization.

📄 Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting

📄 Stylus: Automatic Adapter Selection for Diffusion Models ​​

📄 BlenderAlchemy: Editing 3D Graphics with Vision-Language Models

📄 LEGENT: Open Platform for Embodied Agents 

📄 Capabilities of Gemini Models in Medicine

📄 Invisible Stitch: Generating Smooth 3D Scenes with Depth Inpainting

📄 MicroDreamer: Zero-shot 3D Generation in sim20 Seconds by Score-based Iterative Reconstruction 

📄 KAN: Kolmogorov-Arnold Networks  

📄 MotionLCM: Real-time Controllable Motion Generation via Latent Consistency Model

📄 Lightplane: Highly-Scalable Components for Neural 3D Fields

AI Creates Comics

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.