About This Week’s Covers
Over the past year of creating cover images, I began with DALL-E/GPT and later switched to using MidJourney. In the past month, several strong new competitors have emerged. A few weeks ago, I tried BlackForest Labs’ Flux. Last week, I tested the Grok tool (a white-label version of Flux). Flux is stronger than MidJourney when it comes to complex compositions.
This week, I experimented with Ideogram.ai 2.0, and it blew me away. The image quality is very close to MidJourney, but it also has the same ability to create complex images, such as multiple layers and nested elements, while offering incredibly strong text capability. When it comes to newsletters or publishing needs, I can’t see myself returning to MidJourney unless there’s an overwhelming chorus praising new text support.
Here are all of my Ideogram images over time (another nice feature): https://ideogram.ai/u/ethanh/generated
Ideogram has a nice feature called “Magic Prompt,” which I left on for all of this week’s covers. The main cover prompt I used was “an amazon delivery truck with the words ‘AI NEWS 47: 2024/08/23’ on the side. Magic hour. San Francisco.” The Magic Prompt generated: “A medium shot of an Amazon delivery truck parked on the road. The truck has the words ‘AI NEWS 47: 2024/08/23’ written on the side. The background reveals a city with tall buildings. The sky has an orange hue.” I then upscaled it using Magnific.ai.
Some of the category covers are simply remarkable. Here are few highlights:
Executive Summaries:
Amazon’s GenAI Coding Assistant Saves 4,500 Developer-Years of Effort
Amazon CEO Andy Jassy shared insights on how the company’s generative AI assistant, Amazon Q, is revolutionizing software development. A new feature drastically reduced the time required to upgrade applications to Java 17—from an average of 50 developer-days to just a few hours. This has saved Amazon approximately 4,500 developer-years of work, enabling the company to modernize over 50% of its Java systems in under six months. Developers accepted 79% of the auto-generated code reviews without further changes, boosting productivity. The upgrades have also improved security and cut infrastructure costs, resulting in an estimated $260 million in annual efficiency gains. Jassy emphasized that Amazon plans to further leverage and expand these capabilities across more transformations.
Ajassy | emollick
Will Websites and Apps Matter in an AI-Driven Future?
Jeremiah Owyang argues that as AI evolves, it will transform how people access information, potentially diminishing the role of traditional websites and apps. In this future, AI would gather, filter, and even respond to information requests on users’ behalf, delivering content directly through preferred formats like text, audio, or video. Despite this shift, Owyang believes websites and apps will still play a role by offering an “agent API” that allows them to integrate seamlessly with AI agents, ensuring they remain relevant and useful in an increasingly AI-centric landscape.
jowyang
Waymo Doubles Weekly Robotaxi Rides, Eyes Further Expansion
Waymo, the Alphabet-owned autonomous vehicle company, announced that it has doubled its weekly paid robotaxi rides in the U.S. to over 100,000 since May. The increase comes amid a $5 billion investment from Alphabet and recent service expansions in Phoenix and San Francisco. Waymo’s latest “generation 6” self-driving system promises improved performance in diverse weather conditions and reduced hardware costs. San Francisco now serves the most trips among Waymo’s service areas, which include Phoenix, Austin, and Los Angeles. With around 700 vehicles in its fleet, Waymo is currently the only commercial robotaxi service operating in the U.S., as competitors like GM’s Cruise face setbacks and Tesla’s self-driving ambitions remain unfulfilled.
cnbc
California’s AI Regulation Bill Weakened After Amendments from Anthropic and Industry Pressure
California’s bill aimed at regulating AI to prevent large-scale disasters (SB 1047), has softened after pushback from AI firms, including Anthropic. For example, the Attorney General can no longer sue companies for unsafe practices before a catastrophe, and AI developers are required only to submit safety statements instead of legally binding certifications. The bill eliminates plans for a new government agency but expands the Board of Frontier Models’ role within the existing Government Operations Agency. Despite these changes, the bill still holds AI developers liable for catastrophic damages. The revised bill will face a final vote in California’s Assembly before potentially reaching Governor Newsom’s desk for approval. Here’s a summary of Anthropic’s key points:
- Pros: Encourages safety protocols, deters downstream harm through liability clarification, advances the science of AI risk reduction.
- Cons: Pre-harm enforcement through auditing creates potential overreach, the attorney general can enforce the bill before harm occurs creating more overreach, notice periods for incidents reports are too short.
- Suggestions: include flexibility and avoid being overly prescriptive, prioritize preventing catastrophe, avoid restrictions which inadvertently hinder safety.
Google’s New “Reimagine” Feature on Pixel 9 Raises Concerns Over AI Image Manipulation
The Verge’s Chris Welch tests Google’s “Reimagine” feature on the Pixel 9 and highlights its capability to generate incredibly realistic AI modifications directly within the phone’s photo editor (see images of the week, below). Powered by Google’s Gemini AI, the tool can add objects and animals to images with remarkable precision, adjusting lighting and angles seamlessly to blend new elements into the scene. Welch demonstrates how the feature can create scenarios that never occurred, such as accidents, new pets, or bizarre settings. While the technology is powerful and intuitive, it raises ethical questions about AI’s potential for creating misleading images and challenges our perception of authenticity in digital media.
Chriswelch
Ideogram 2.0 Launches with New Features and Expanded Access
Ideogram has rolled out its latest version, Ideogram 2.0, introducing significant upgrades to its text-to-image capabilities. Rowan Cheung calls it “the new gold standard for anyone who uses AI image generation for thumbnails, posters, newsletter graphics, memes, etc.” Users can generate images across five distinct styles: General, Realistic, Design, 3D, and Anime. The Realistic style offers lifelike human features, while the Design style enhances text rendering for graphics such as posters and t-shirts. Ideogram 2.0 also supports custom and pre-set color palettes, allowing for precise control over image aesthetics. Leaderboard evaluations show Ideogram 2.0 outperforms both Flux Pro and DALL·E 3. Available for free to all users, the release includes an iOS app, a beta API, and a new Ideogram Search feature.
Ideogram_ai
Top AI Visuals and Charts
The “Reimagine” feature on Google’s new Pixel 9 lineup is incredible. It’s so impressive that testing it has left me feeling uneasy on multiple occasions.
With a simple prompt, you can add things to photos that were never there. And the company’s Gemini AI makes it look astonishingly realistic. This all happens right from the phone’s default photo editor app. In about five seconds.
“I’ve been beta-testing Ideogram 2.0 for the past couple of days. I have to say: it’s REALLY good at generating text, and a lot better at realism. This is the new gold standard for anyone who uses AI image generation for thumbnails, posters, newsletter graphics, memes, etc.
“Atlas doing a quick warm up before work.
“4. Here’s an example of text-to-video. The prompt was “Inside shot, close-up, a Chinese child is eating dumplings”
“9. This one is a classic. Prompt: “A Chinese man sitting at a table, eating noodles with chopsticks”
“images: @midjourney style: –p rixmhgc music: @suno_ai_ video: @Kling_ai upscale: @topazlabs soundfx: @elevenlabsio
“1. Here’s an example of Kling’s image-to-video. The prompt here was “clean the debris off the old woman”
Top 44 Links of The Week – Organized by Topic
Artificial General Intelligence (AGI)
AGI Safety and Alignment at Google DeepMind: A Summary of Recent Work — AI Alignment Forum
Audio
ElevenLabs
ElevenLabs’ text-to-speech app Reader is now available globally | TechCrunch
Chips, Hardware, and Infrastructure
NVIDIA
NVIDIA Releases Small Language Model With State-of-the-Art Accuracy | NVIDIA Blog
“Nvidia and Mistral just released Mistral-NeMo-Minitron 8B, a small language model that can run on laptops and PCs It outperforms Mistral-7B and Meta-LLama 3.1-8B on the Open LLM leaderboard Small models are improving at an insane rate 👀
Imagery
Google (Imagen)
Google’s upgraded AI image generator is now available – The Verge
MidJourney
“Here are all of my Midjourney personalization codes These have been tuned on ~3800 image ratings They tend to push things towards a more analog film look. To use, just add them to the end of your prompt (left) MJ default (right) with –p –p 9zrkar2 –p es868ts –p yypkxil
Midjourney web experience is now open to everyone | Hacker News
Locally Run AI Models
“How the hell Phi-3.5 is even possible? Phi-3.5-3.8B (Mini) somehow beats LLaMA-3.1-8B.. (trained only on 3.4T tokens) Phi-3.5-16×3.8B (MoE) somehow beats Gemini-Flash (trained only on 4.9T tokens) Phi-3.5-V-4.2B (Vision) somehow beats GPT-4o (trained on 500B tokens) how? lol
Open Source AI
Grok
“I got early (beta) access to Grok 2. Spent all night testing it. Here are 6 things it can do that ChatGPT cannot:
“xAI began rolling out early beta access for Grok 2, the new AI model that leverages real-time data from X I tested it and found 6 core features that it can do that ChatGPT can’t (in the thread below)
Publishing
Anthropic Lawsuit: Authors Sue Amazon-backed Company For Infringement
New Washington Post AI tool sifts massive data sets
OpenAI strikes search deal with Condé Nast
“We’re partnering with Condé Nast to deepen the integration of quality journalism into ChatGPT and our SearchGPT prototype.
“How The Washington Post’s AI tool “Hayatacker” empowers journalists 🚀📰 • Extracts stills, text & objects from videos • First story: Analyzed 745 political ads on immigration • AI organizes data, reporters verify & analyze AI enhancing, not replacing, journalism.” / X
Robotics and Embodiment
“Agibot unveiled 5 new commercial humanoid robots today & expects 300 deliveries this yr Raise A2, A2-W, A2-Max, Lingxi X1, Lingxi X1-W
https://twitter.com/tphuang/status/1825539760565 436484
“First, AGIBOT, a China-based robotics startup, unveiled a family of 5 advanced humanoid robots. The robots are specifically designed for diverse tasks — from household chores to industrial operations, directly challenging Elon and Tesla’s Optimus
Former Huawei ‘Genius Youth’ recruit launches humanoid robots to rival Tesla’s Optimus | South China Morning Post
“The new Astribot S1 has been unveiled by Chinese company Astribot, showcasing the bimanual wheeled robot performing many household tasks autonomously.
Watch this robot quickly install roof shingles | TechCrunch
Boston Dynamics
“Boston Dynamics posted a new video of its Atlas robot doing push-ups, showcasing advancements in dynamic movement control Can’t wait to see the Atlas vs. Figure vs. Optimus cage match
“Atlas doing a quick warm up before work.
Tesla
Tesla is hiring people to train its Optimus bot via motion capture – The Verge
Unitree
“Unitree, another high-performance robot manufacturer from China, showcased its new G1 humanoid robot. The updated bot is nearing “mass production” and comes at a price tag of only $16,000 This thing has insane whole-body control
Video: Unitree G1 humanoid robot ready to leap into mass production
“Unitree G1 mass production version, leap into the future! Over the past few months, Unitree G1 robot has been upgraded into a mass production version, with stronger performance, ultimate appearance, and being more in line with mass production requirements. We hope you like it.🥳
Science and Medicine
“We’ve added support for LaTeX rendering as a feature preview. Claude can now display mathematical equations and expressions in a consistent format.
“The hardest part of writing a social science paper, solved: “Claude, make this all nice in LaTeX”
“Deep learning to accelerate the solution of partial differential equations and other simulations.” / X
Video
Kling
“4. Here’s an example of text-to-video. The prompt was “Inside shot, close-up, a Chinese child is eating dumplings”
“9. This one is a classic. Prompt: “A Chinese man sitting at a table, eating noodles with chopsticks”
“images: @midjourney style: –p rixmhgc music: @suno_ai_ video: @Kling_ai upscale: @topazlabs soundfx: @elevenlabsio
“1. Here’s an example of Kling’s image-to-video. The prompt here was “clean the debris off the old woman”
“8. The prompt here was “Dog swimming, water ripples spreading”
“6. Kling AI’s Olympic creative trailer
“5. Kling AI is also really great at drone fly-through effects
“2. Another one of my favorite use cases is bringing old photos or memes to life
“You can access Kling AI here: https://t.co/wuW1msxrS6 1. Click “Sign Up” 2. Click “AI Videos” 3. Enter your creative ideas in “Prompt” or upload photos, then click “Generate” You’ll get enough free credits for ~6 videos per day.” / X
Luma
“The @LumaLabsAI dream machine v1.5 update seems promising > physics seem more realistic > movement looks less janky but damn, i REALLY wish the ecosystem would focus on image-to-video improvements (motion prediction, temporal coherence, etc) > text-to-video improvements (NLP” / X
“Luma Labs just released Dream Machine 1.5, a major upgrade to their current AI video generation model. The update boasts higher-quality text-to-video, smarter prompt understanding, and better image-to-video capabilities.
“Dream Machine 1.5 is here 🎉 Now with higher-quality text-to-video, smarter understanding of your prompts, custom text rendering, and improved image-to-video! Level up.
Luma Dream Machine
Runway
Runway Gen-3 Alpha Turbo: AI videos faster than you can type | VentureBeat
Leave a Reply