Warp-Speed Wednesdays

Your Must-Read Weekly Tech Updates: Apple Vision Pro developers tools, Sam Altman's World Tour Recap, Self-driving to AGI, & Wearables will be Health Data Industry

Click that web link here to get the full galactic experience.

---

Apple’s Vision Pro

I am sure you all have heard about Apple’s newest transformative product — Vision Pro VR/AR glasses. You can learn about the features from the link in the title.

Let's discuss the history of social media and why this is the next moment in online social experience.

The iPhone 1 launched 2007.

• Facebook - 2006

• Twitter - 2006

• Instagram - 2010

• YouTube - 2005

• Pandora - 2005

No coincidence! The iPhone brought the computer on the go, specifically video, and music. Apple is looking to recreate this iPhone moment. Why?

The number 1 resource of the new economy.

DATA!

Health, consumer sentiment, financial data, etc. People should really start to see Apple as the next-century health-data company. They will know a lot about your well-being.

Apple scans your face to create an avatar of you for FaceTime and even uses it for audio reconstruction specifically for you.

But, how can you take your piece of the pie from this moment?

VisionPro 1 is a developer's tool to create new social media platforms for the next decade. Tools and information here.

An ex-Apple employee that worked on the Vision Pro provides insider details on how you can develop applications for the device here.

---

---

Sam Altman’s world tour comes to an end…

Here is the highlight reel:

1. Altman is worried that they may have already released the AI beast onto the internet. I don’t think so and neither does Ilya sound very terrified that this has already happened. The near future is a different question…

2. Ilya Sutskever exclaims superintelligent AI is the ultimate challenge for Humanity in recent interview at Tel Aviv University. It is inevitable and part of our evolution arc.

3. OpenAI's moat is the technology needed to create Superintelligence.

4. Sam Altman says OpenAI won’t go public now because he may have to make ‘a very strange decision’ that investors will disagree with

   “In terms of decision-making, Altman said he wants full autonomy from OpenAI shareholders, saying in Abu Dhabi that ‘the chance that we have to make a very strange decision someday is nontrivial.’ Altman didn’t clarify what a ‘very strange decision’ could look like, but added, ‘I don’t want to be sued by…public market, Wall Street, etc.’”

5. Intelligence as a “fundamental property of matter”

   Q: "After doing AI for so long, what have you learned about humans?" Sam Altman: "I grew up implicitly thinking that intelligence was this, like really special human thing and kind of somewhat magical. And I now think that it's sort of a fundamental property of matter..."

6. OpenAI, Google DeepMind and Anthropic just granted the UK government priority access to their AI models for research and safety purposes.

7. “Honestly, I feel so bad about the advice I gave while running YC I’ve been thinking about deleting my entire blog… Make something people want.” - Altman

8. Not a part of his tour but here he gives advice for young people (even not young people):

   1. Compound yourself

   2. Have almost too much self-belief

   3. Learn to think independently

   4. Get good at "sales"

   5. Make it easy to take risks

   6. Focus

   7. Work hard

   8. Be bold

   9. Be willful

AI

1. AI creates photorealistic ultrasound images. Expect live video reconstruction soon.

   

2. Orca: Progressive Learning from Complex Explanation Traces of GPT-4

   Recent AI advancements aim to improve smaller models through imitation learning, utilizing large foundation models (LFMs). Despite challenges, such as insufficient imitation signals and homogenous training data, Orca, a new 13-billion parameter model, excels. Guided by teacher assistance from ChatGPT, it learns from GPT-4's rich signals, using a broad range of carefully selected imitation data.

   Orca outperforms current instruction-tuned models, matching ChatGPT on complex benchmarks, and demonstrating competitive results in professional exams. This emphasizes the potential of learning from step-by-step explanations, whether human or AI-generated, in enhancing AI capabilities.

Robotics & Transportation

1. Brett Adcock, CEO of Figure Robotics, leads Acher Aviation to fully electric aviation with 10-min flights from Newark to New York City

   

2. Tesla FSD is continuously trying to progress toward safer and fully autonomous driving nation-wide by converting its software stack to an end-to-end neural network

3. Tesla’s FSD is still not autonomous and has led to crashes and fatalities

   “The number of deaths and serious injuries associated with Autopilot also has grown significantly, the data shows. When authorities first released a partial accounting of accidents involving Autopilot in June 2022, they counted only three deaths definitively linked to the technology. The most recent data includes at least 17 fatal incidents, 11 of them since last May, and five serious injuries.”

   Tesla FSD is at the frontier of AI research, and these crashes and deaths further send the signal that a regulatory body should be set in place to ensure standards and metrics for AI safety. Currently, there is a lot of bias in the FSD crash data from Tesla and the critics which makes it difficult for the consumer to make an educated decision.

4. Yann LeCun, Head of Meta AI, believes self-driving is the best application to get to AGI

   “I do not believe we can get anywhere close to human-level AI (even cat-level AI) without

   (1) learning world models from sensory inputs like video

   (2) an architecture that can reason and plan”

   He is suggesting neural nets trained on images and video (images stitched together) will produce world models and reasoning capabilities (more about LeCun below).

   This is in line with what other AI experts, such as Andrej Karapthy, have said in the past.

   What do you think?

   “As an example, here's a generated video (this is not real!) from our AI's world model showing it is capable of reasoning and imagining about overtaking a bus.” — Alex Kendell, CEO of Wayve AI

   

   James Douma & Scott Walker agree and so do I. Timestamp: 1:11:00

5. Robots upskilling using videos of humans doing tasks

   Imagine if robots could watch our daily actions, learn from them, and then recreate those actions seamlessly.

   These researchers are training their robots by showing them videos of people doing everyday stuff, right from videos. The robots learn to anticipate where and how people are likely to interact within a scene, whether it's reaching for a cup of coffee or picking up a book.

   

   Now, here's the cool part. They are not just training the robots to mimic people but using the newfound knowledge to integrate with four different learning styles that robots can use to understand their world better. It's like giving our robots a well-rounded education!

   Vision-Robotics Bridge (VRB): Uses computer vision techniques (how computers see and understand images) with the physical actions of robots. This unique blend is already showing promise across a wide range of tasks and environments.

   So, the future looks bright, and our robots will continue to learn from videos of humans performing tasks. Think of it as downloading YouTube into a robotic fleet, letting them learn everything from cooking to carpentry. Exciting times are indeed ahead!

6. I-JEPA: LeCun’s Vision finally releases his vision for computers that reason like Humans

   • It can be used for many different applications without fine tuning and is highly scalable.

   • Achieves SOTA by training 632M parameter visual transformer model using 16 A100 GPUs in under <72 hrs. Other methods take 10 times more GPU-hours.

   • I-JEPA captures common-sense knowledge through self-supervised learning from unlabeled data.

   

   • The model predicts missing information at a high level of abstraction, avoiding generative model limitations.

7. My AGI Prediction

   Let's take a moment to emphasize that a significant majority, around 80-90%, of current AI tools are all built on a common ground, which is the Transformer neural network. Essentially, they're tweaking the same architectural framework (I-JEPA has a more significant change), tailoring it with minor adjustments and the specific data they are trained on.

   Now, let's take this a step further: I foresee a future where we'll have a cornerstone model for Robotics, acting as the bedrock on which advanced robotic technologies will be developed.

   In fact, I predict this model will be the very foundation for AGI, or Artificial General Intelligence. Picture this as the blueprint that could potentially steer us toward a world where AI can rival human intelligence.

Healthcare

1. Artificial limbs are and will change lives

   

2. A washable wearable biometric device using printed electronics technology provides real-time data for telemedicine.

   Biometric sensors, utilizing electrical signals to monitor muscle activity, heart rate, and more, are evolving from traditional cumbersome designs to innovative printed solutions.

   

   These printed sensors, made with conductive inks by Butler Technologies, adhere easily to garments and medical braces, eliminating the need for messy gels or complicated wires.

   They're flexible, washable, and ideal for telemedicine, transmitting real-time data to doctors via Wi-Fi or Bluetooth. To streamline the data flow, Loft collaborated to ensure the sensor signals could be read and displayed on a webpage.

   Despite challenges in applying these sensors to fully stitched garments, clever engineering solutions allowed accurate muscle activity data collection.

   This revolutionary, comfortable technology can enhance remote healthcare, track muscle recovery post-injury, and even be incorporated into everyday or sports apparel. The future of printed biometric sensors, bridging the gap between hardware and software, is here.