Adobe has announced the Firefly Video Model, an AI text-to-video generation tool that creates videos from written prompts ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 
 

Tuesday 15th October 2024 | Join Free | Archive

 
TechBlend Logo
 

The morning paper for busy AI & tech professionals

 
 
 

Good morning! This is your daily ☕️ TechBlend.

 

In today's blend:

 

🚀 NASA launches mission to explore the frozen frontier of Jupiter’s moon Europa

 

🚀 SpaceX's Starship test completes with a remarkable ‘chopstick' booster catch

 

🤖 Adobe unveils AI video generator trained on licensed content

 

🚗 Elon Musk wants to dominate robotaxis — first he needs to catch up to Waymo

 

🍏 Apple study exposes deep cracks in LLMs’ “reasoning” capabilities

 

🖼️ Zoom's custom AI avatar tool may come with risks

 

👓 Cheaper Apple Vision headset rumored to cost $2,000, arriving in 2026

 

🎁 + 8 other stories you might like

 

🛠️ + 7 tools and resources

 
 
 
 

🚀 NASA launches mission to explore the frozen frontier of Jupiter’s moon Europa

  • NASA's Europa Clipper mission successfully launched aboard a SpaceX Falcon Heavy rocket, beginning a $5.2 billion mission to explore Jupiter's moon Europa for potential signs of extraterrestrial life.
  •  
  • The spacecraft, carrying advanced scientific instruments, aims to study Europa's icy surface, investigate its subsurface ocean, and assess whether it has the necessary ingredients to support life today.
  •  
  • Europa Clipper will not land on the moon but will perform close flybys to gather data on its surface and subsurface structure, hoping to find evidence of liquid water and organic materials.
  •  
 

🚀 SpaceX's Starship test completes with a remarkable ‘chopstick' booster catch

  • SpaceX's Starship spacecraft successfully splashed down in the Indian Ocean an hour after its launch from South Texas, marking the second full Starship launch and return to Earth.
  •  
  • The Super Heavy booster made a historic landing back at the launchpad, where it was caught by SpaceX's launch tower using arms referred to as "chopsticks," demonstrating potential for reusability similar to the Falcon 9 rockets.
  •  
  • The Starship's heat shield was redesigned for this test, incorporating new tiles and an additional ablative layer after previous issues during reentry, and the FAA granted flight approval just a day before the launch.
  •  
 

🤖 Adobe unveils AI video generator trained on licensed content

  • Adobe has announced the Firefly Video Model, an AI text-to-video generation tool that creates videos from written prompts, focusing on being ethically safe by using licensed training data.
  •  
  • The Firefly Video Model is being marketed as the first video model intended to be commercially safe, although it is currently in beta testing and access is limited to a waiting list.
  •  
  • The model builds on Adobe's Firefly image synthesis techniques and aims to serve media professionals, promising seamless integration with traditional video content.
  •  
 

🚗 Elon Musk wants to dominate robotaxis — first he needs to catch up to Waymo

  • Elon Musk is set to unveil the "Tesla Robotaxi" at the "We Robot" event in Los Angeles, emphasizing Tesla's ambitions in the driverless car market, where it currently commands a stock valuation of $755 billion despite competing companies like General Motors and Ford selling more vehicles.
  •  
  • Tesla's Full Self-Driving software has shown significant limitations, requiring human intervention every 13 miles, while competitors like Waymo have been successfully offering fully driverless rides since 2020, significantly increasing their weekly ride numbers.
  •  
  • The disparity between Tesla and Waymo's progress raises questions about Tesla's ability to scale its self-driving technology effectively, particularly in terms of software development and operational infrastructure required for a robotaxi service.
  •  
 

🍏 Apple study exposes deep cracks in LLMs’ “reasoning” capabilities

  • A study by Apple researchers indicates that current large language models (LLMs) lack true logical reasoning capabilities, relying instead on probabilistic pattern matching, which leads to fragility in response to minor changes in input data.
  •  
  • The introduction of trivial changes to benchmark questions significantly decreased the performance of LLMs, with accuracy drops observed between 0.3% and 9.2%, and even more dramatic failures of 17.5% to 65.7% when irrelevant information was added to prompts.
  •  
  • These results reinforce previous research suggesting that LLMs do not perform formal reasoning but rather mimic reasoning through learned patterns from their training data, highlighting critical limitations in their understanding of underlying concepts.
  •  
 

🖼️ Zoom's custom AI avatar tool may come with risks

  • Zoom is set to launch a feature next year that allows users to create AI-animated, photorealistic avatars from video clips, enabling asynchronous communication and improved productivity.
  •  
  • The avatars will sync lip movements with user-generated audio based on typed scripts, but the feature raises concerns about deepfake risks and misuse.
  •  
  • Zoom plans to implement safeguards, including watermarking and authentication measures, to prevent malicious use of the technology, though specific details on these measures remain unclear.
  •  
 

👓 Cheaper Apple Vision headset rumored to cost $2,000, arriving in 2026

  • Apple is developing a successor to the Apple Vision Pro and a more affordable version expected to launch within the next two years.
  •  
  • The lower-end headset is anticipated to cost around $2,000 and may feature less powerful hardware and lower-resolution screens compared to the $3,499 Vision Pro.
  •  
  • Both headsets are projected to be released in 2026, with Apple anticipating higher sales for the lower-priced model.
  •  
 
 
 
⚡ Quick links
 
 
 

Asahi Linux's bespoke GPU driver is running Windows games on Apple Silicon Macs (2 minute read)

 

Two people communicate in dreams: Inception movie-styled sci-fi turned into reality (3 minute read)

 

This AI-Powered Invention Machine Automates Eureka Moments (10 minute read)

 

Deep Generative Models: A Guide for Transportation Research (75 minute read)

 

The quiet art of attention (3 minute read)

 

Hurricane Helene and the ‘Fuck It' Era of AI-Generated Slop (2 minute read)

 

Scientific found a way to make sound travel in only one direction (4 minute read)

 

OpenAI pursues public benefit structure to fend off hostile takeovers (5 minute read)

 
 
👨‍💻 Engineering & Resources
 
 
 

Realistic 4D Scene Generation (GitHub Repo)

 

OpenAI's Swarm (GitHub Repo)

 

zzz (GitHub Repo)

 

Document-level Translation Agent (GitHub Repo)

 

Text-to-Image Model on Public Data (18 minute read)

 

Fast 3D Compression (3 minute read)

 

Zyphra's Mamba 2 based model beats Mistral (8 minute read)

 
 
 

Not subscribed to ☕️ TechBlend yet?

Join Here
 

☕️ TechBlend

Have feedback? Send me a message.

You were sent this message because you subscribed to TechBlend (previously SMMRY).

 
Unsubscribe

Block A, Phoenix Yard, Upper Brown Street, Leicester, England, LE1 5AG, United Kingdom