Forward Future AI
Posts
🧑‍🚀OpenAI Launches GPT-4o Fine-Tuning, Microsoft Unveils Phi-3.5-mini AI Model

🧑‍🚀OpenAI Launches GPT-4o Fine-Tuning, Microsoft Unveils Phi-3.5-mini AI Model

OpenAI now offers fine-tuning for GPT-4o, allowing tailored AI outputs, while Microsoft introduces the compact Phi-3.5-mini model for high-quality data tasks. Waymo doubles robotaxi rides to 100K weekly, and AWS envisions AI transforming software development. Stay updated on key AI developments.

Matthew Berman
August 23, 2024

Good morning, it’s Friday! Buckle up, AI enthusiasts – this week's newsletter has it all. We're talking self-driving cars, the future of coding, major tech launches, and groundbreaking research. Let's jump in!

Your Daily Roundup:

Fine-tuning for GPT-4o is now available, allowing developers to enhance model performance and customize outputs for specific applications.
The Microsoft Phi-3.5-mini model is a compact yet advanced AI designed primarily for tasks requiring strong reasoning.
Waymo has doubled its paid robotaxi rides to over 100,000 per week in the US since May.
Amazon Web Services' chief suggests a future where the role of software developers shifts away writing code.
We finally have a definition for open-source AI.

👉️ Top Stories

OpenAI Launches Fine Tuning for GPT-4o

Fine-tuning for GPT-4o is now available, allowing developers to enhance model performance and customize outputs for specific applications. Through September 23, organizations can access 1M free training tokens per day, with the option to fine-tune both GPT-4o and GPT-4o mini models. Fine-tuning helps tailor responses, optimize tone, and meet domain-specific requirements with just a few examples. Continue reading here.

Microsoft Introduces Phi-3.5-mini Model

The Phi-3.5-mini model is a compact yet advanced AI designed primarily for tasks requiring strong reasoning and operating in memory-constrained environments. It extends the Phi-3 family with a focus on high-quality data and boasts a 128K token context length, making it suitable for long document processing. Continue reading here.

Waymo has doubled its weekly paid robotaxi trips to 100,000 since May

Waymo, an Alphabet subsidiary, has doubled its paid robotaxi rides to over 100,000 per week in the US, with its latest "generation 6" self-driving system poised to enhance operation in various weather conditions; it has expanded its service area and claims superior safety performance compared to human drivers. Continue reading here.

Amazon cloud chief tells employees that most developers could stop coding soon as AI takes over

Amazon Web Services' CEO, Matt Garman, suggests a future where the role of software developers shifts from writing code to focusing on innovation and building products aligned with customer needs, as AI and new tools could handle more programming tasks. AWS emphasizes the importance of upskilling developers to harness AI in their workflow, framing these technological advancements as opportunities for developers to engage in more creative and high-value work rather than a decrease in their significance. Continue reading here.

What margins? AI's business model is changing fast, says Cohere founder

Cohere CEO Aidan Gomez highlights the challenges AI startups face as selling model access becomes a "zero margin business" due to competitive price dumping. Companies like OpenAI and Google are cutting prices, while open-source models from Meta are free, creating a race to the bottom. Gomez suggests that while selling AI models remains crucial, focusing on applications and products may be a more sustainable revenue path in the short term. Continue reading here.

We finally have a definition for open-source AI

The group advocates for an open-source AI system that allows unrestricted use, inspection, modification, and sharing, while also promoting transparency in the system's training data, source code, and weights. Despite AI models from companies like Meta and Google being accessible, some experts argue they aren't truly open source due to restrictive licenses and non-public training data sets, with companies potentially misusing the term "open source" for marketing, leading to a perception of trustworthiness without verifiable openness. Continue reading here.

👾 Forward Future Original

Former Google CEO Eric Schmidt's Controversial Stanford Interview: A Deep Dive into AI's Future

This post summarizes key points from a recent interview at Stanford with former Google CEO Eric Schmidt. We'll explore his insights on AI, Google's culture, and the future of technology, including some controversial takes that led to the video's removal from YouTube.

The Future of AI: Large Context Windows, Agents, and Text to Action

Schmidt sees a future where AI models have massive context windows, allowing them to process and understand vast amounts of information, acting like a powerful short-term memory. He predicts that this, coupled with the development of AI agents (LLMs with additional functionality, like the ability to conduct experiments) and text to action capabilities (converting natural language to executable code), will have a profound impact on the world.

"In the next year, you're going to see very large context windows, agents, and text to action. When they are delivered at scale, it's going to have an impact on the world at a scale that no one understands yet, much bigger than the horrific impact we've had by social media."

Google's Culture: Work-Life Balance Over Winning?

One of the most controversial statements from the interview was Schmidt's critique of Google's culture:

"Google decided that work-life balance and going home early and working from home was more important than winning."

He contrasted this with the intense work ethic often seen in successful startups and emphasized the importance of founders being in charge and pushing their teams hard.

Continue reading… The AI Race: US vs. China

✍️ Editor Picks

Research

Transformers in music recommendation

YouTube Music has implemented transformers within their recommendation system to better interpret user actions, such as skips and likes, within the context of their current activity, thus tailoring music suggestions more accurately. Through incorporating user feedback and contextual information, the enhanced system has demonstrated improved user engagement, indicated by reduced skip rates and increased listening times, with potential future enhancements including integration with nonsequential features for even more refined recommendations. Continue reading.

Robotics

$16,000 humanoid robot ready to leap into mass production

Unitree Robotics has unveiled the mass production variant of its G1 humanoid robot, designed for general purposes and featuring enhanced capabilities such as improved performance in climbing, leaping, and real-time gait adjustment. The versatile G1 stands 1.32 meters tall, offers 23 degrees of freedom in its joints, includes features like 3D LiDAR, RealSense depth camera, and can operate for approximately 2 hours on a single charge. Continue reading.

Models

Mistral-NeMo-Minitron 8B Foundation Model Delivers Unparalleled Accuracy

The newly introduced Mistral-NeMo-Minitron 8B model by NVIDIA and Mistral AI sets a new benchmark in accuracy for models in its size category. Developed by width-pruning the Mistral NeMo 12B model and applying knowledge distillation, the 8B model achieves leading performance across nine popular benchmarks. Continue reading.

🚀 AI Launches

Luma Drops Dream Machine 1.5 - Luma Labs has updated its AI video creation tool, Dream Machine, to version 1.5, enhancing realism, motion tracking, and prompt comprehension.As
Astribot S1: Launch - Meet Astribot S1: the Next-Gen AI Robot Assistant with unparalleled agility, dexterity and accuracy.
Daily Launches Bots Platform for Ultra-Low Latency Voice, Vision, and Video AI - Daily has introduced Daily Bots, an open-source cloud platform designed for ultra-low latency voice, vision, and video AI, enabling conversational interactions with latencies as low as 500ms.
Launch of Hotshot - Hotshot is a large-scale diffusion transformer text-to-video model providing superior prompt alignment, consistency, and motion, outperforming similar models in user preference by 70%.
Introducing LM Studio 0.3.0 - Enhanced features for running local large language models (LLMs), including Retrieval Augmented Generation for document interaction, OpenAI-compatible JSON outputs, customizable UI themes, automated load parameters with advanced configuration options, network server capabilities, folder-based chat organization, and multiple regeneration of chat responses
Introducing Ideogram 2.0 - Ideogram has released Ideogram 2.0, enhancing its image generation capabilities with a focus on superior quality and customization, including a beta API for developers, providing high-quality image outputs at lower costs.

📽️ New Videos

🧰 Toolbox

Prompt Gallery in Google AI Studio - Google AI Studio has introduced a native prompt gallery, enabling users to explore new capabilities like long context, native multi-modal outputs (image, video, and audio), and structured outputs.
Gemini 1.5 Pro Removes RPD Limit for Paid Users - Gemini 1.5 Pro has lifted its RPD (requests per day) limit for paid tier users, increasing the capacity from 10,000 to 518,400 RPD.
Anthropic Interactive Prompt Engineering Tutorials - An in-depth course designed to teach individuals how to craft effective prompts for Claude, featuring hands-on practice, nine chapters with exercises, and a focus on prompt engineering techniques from basic construction to avoiding common errors and creating complex examples for specific industries.