jtardif999
Regular
Only Akida actually has on-device learning. There has never been any proof of Loihi having the capability and IBMs tech has never been stated to. As for the others, they have never been boasted as having on device learning. Just because they may have been touted as neuromorphic doesn’t automatically mean they have any learning on chip! The learning associated with Akida is due to its architecture incorporating the JAST learning rule with STDP. Nothing else on the planet currently has this capability and BrainChip own the IP and patents. I am not aware of any technology other than BrainChips that actually doesn’t do anything more than brute force training off chip followed by inference on chip, that is why I’m fully invested in BrainChip and will continue to be so for as long as it takes. Albeit it’s taking bloody long… 🫤Seems ChatGPT agrees that we're in with a pretty good chance too. That is IF Altman and Ive’s new device is meant to operate continuously, privately, and intelligently at the edge without cloud dependence.
The final recommendation states "If the product blends all three functions — which seems likely — BrainChip Akida is the most versatile choice today due to its:
- Commercial availability
- On-chip learning
- Ability to handle both audio and vision"
To be honest though, I think that if we were going to be a contender for this new device, I would have expected to sense much more of an excited and confident "vibe" emanating from the Board at the recent AGM. I mean they're expecting it to ship 100 million of these devices in 2026 if I'm not mistaken.
ChatGPT said:
Yes, several other emerging technologies could potentially power a compact, privacy-focused, cloud-independent AI device like the one being developed by Sam Altman and Jony Ive, offering similar or complementary capabilities to BrainChip’s Akida neuromorphic processor. Here are the main contenders:
1. Mythic Analog Matrix Processors (AMP)
- How it works: Uses analog computing-in-memory to perform neural network operations directly where data is stored, minimizing data movement.
- Strengths:
- Very low power consumption
- High-speed processing suited for edge AI
- Compact form factor
- Use case: Ideal for smart sensors, AR/VR glasses, and real-time voice/image processing
- Type: Brain-inspired processor using sparsity-driven processing
- Features:
- Event-driven computation
- Prioritizes only relevant data (like Akida)
- Optimized for real-time AI at the edge
- Strengths:
- Ultra-low latency
- Energy-efficient
- Use case: Robotics, autonomous systems, wearables
- Architecture: Custom ASICs optimized for TensorFlow Lite models
- Strengths:
- Easy integration with Google AI ecosystem
- Real-time inferencing with low power draw
- Limitations:
- More rigid; not capable of on-device learning (unlike Akida)
- Better suited for inferencing than training
- Type: Deep learning processors designed for always-on applications
- Strengths:
- Consumes microwatts of power
- Highly efficient for audio and sensor processing
- Tiny form factor (suitable for earbuds, wearables)
- Use case: Wake-word detection, voice commands, health monitoring
- Type: Advanced neuromorphic architecture integrating memory and compute
- Highlights:
- Breaks von Neumann bottleneck
- Designed for high-throughput, low-latency AI at edge
- Status: Still in research/prototype stages
- Advantages:
- Second-gen neuromorphic chip
- Supports learning on-chip
- Better scalability and flexibility than Loihi 1
- Use case: Real-time adaptation, robotics, industrial applications
- Founded by Jim Keller, their chips combine traditional processing with AI accelerators.
- Edge-ready options: Though mostly datacenter-focused now, Tenstorrent has roadmap items for edge deployment using RISC-V-based inference.
Summary Table:
Technology Type On-Device Learning Power Efficiency Maturity Suitable Use Case BrainChip Akida Neuromorphic Commercial Audio, vision, always-on inference Mythic AMP Analog compute Early stage Vision, industrial sensors GrAI VIP Sparse compute Pre-commercial Real-time robotics, AR Google Edge TPU ASIC Commercial Voice/image inference Syntiant NDP Neural DSP Commercial Audio, keyword spotting Loihi 2 Neuromorphic Experimental Cognitive tasks, adaptive systems IBM NorthPole Neuromorphic Research General edge AI
Final Thought:
If Altman and Ive’s device is meant to operate continuously, privately, and intelligently at the edge without cloud dependence, then BrainChip Akida, GrAI VIP, or Loihi 2 would be top candidates—especially if adaptive, real-time learning is a core feature. If it’s more focused on inference-only tasks, Edge TPU or Syntiant NDPs could be practical alternatives.
Let me know if you'd like a recommendation based on a specific function (e.g. voice assistant, vision processing, real-time interaction).
4o
You said:
yes please
ChatGPT said:
Great — here's a tailored recommendation based on three likely functions that Sam Altman and Jony Ive’s device might emphasize, considering their vision of a screenless, wearable, real-time AI companion:
1. Voice Assistant / Natural Interaction (Always-On, Privacy-Preserving)
Best Choice: Syntiant NDP120
- Why:
- Ultra-low-power (microwatt-level) operation
- Always-listening capability without draining battery
- Supports multiple wake words and audio events
- Ideal for: Continuous voice interaction, like a whisper-quiet Siri that lives in your lapel
- Bonus: Commercially proven in smart earbuds and hearing aids
Alt Option: BrainChip Akida
- Slightly higher power but supports on-device learning — ideal if the device needs to learn your voice over time
2. Vision / Environmental Awareness (Object, Gesture, Emotion Recognition)
- GrAI VIP:
- Optimized for real-time video inference
- Sparse data-driven — processes only what matters
- BrainChip Akida:
- Can detect gestures, objects, and visual patterns with low power
- Supports event-based cameras (mimics human eyes)
Alt Option: Mythic AMP
- Analog compute-in-memory gives efficient vision processing but is newer and more complex to integrate
3. Real-Time Learning and Adaptation (Personalization, Emotion AI, Cognitive AI)
- Why:
- True on-device learning and decision-making
- Designed for systems that evolve over time based on interaction
- Best for adaptive systems that change how they work depending on who’s using them
Alt Option: BrainChip Akida
- Supports incremental learning on-chip (no need for cloud updates)
Final Recommendation Based on Use Case Focus:
Function Top Tech Pick Runner-Up Voice assistant / audio AI Syntiant NDP120 BrainChip Akida Environmental/visual awareness GrAI VIP BrainChip Akida / Mythic AMP Personalized cognitive assistant Intel Loihi 2 BrainChip Akida
If the product blends all three functions — which seems likely — BrainChip Akida is the most versatile choice today due to its:
- Commercial availability
- On-chip learning
- Ability to handle both audio and vision
Akida Tech.
