rgupta
Regular
I assume sacking here may relates to some information leakage as well. So we can only know as much as information provided and there may be no further clues.
BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
FJ-215
Regular
equanimous
Norse clairvoyant shapeshifter goddess
Dissatisfied employees and dissatisfied managers can be a good source of improvement.
Two prerequisites are necessary:
1. Exemplary behavior or a desire for change
2. Constructive criticism
If neither is present, it is destructive criticism, which ultimately only damages morale.
I don't see exemplary behavior in Olivier. His criticism of management is the usual platitudes heard in every company—no matter how well or poorly the company is doing.
I firmly believe he's hoping for a job at NVIDIA.
But even at NVIDIA, not everything that glitters is gold.
Two prerequisites are necessary:
1. Exemplary behavior or a desire for change
2. Constructive criticism
If neither is present, it is destructive criticism, which ultimately only damages morale.
I don't see exemplary behavior in Olivier. His criticism of management is the usual platitudes heard in every company—no matter how well or poorly the company is doing.
I firmly believe he's hoping for a job at NVIDIA.
But even at NVIDIA, not everything that glitters is gold.
Frangipani
Top 20
I doubt THAT was what he was looking forward to sharing with the world this Friday... In fact, I don’t think he expected that phone call in the morning at all, it must have taken him by surprise, just before the start of his workday.
(He posted on LinkedIn about having been sacked shortly before 9 am his time).
While he seemed upbeat and enthusiastic in his first post late on Thursday night his time, presumably brimming with excitement to share the news about an important milestone after having put in a lot of work in recent weeks, his second post to me expresses utter disbelief and embitterment.
He evidently wasn’t prepared for the company’s decision at this point in time.
The reveal would likely have been about a new TENNs model or patent and not breaking any NDA with a customer?
Looks like it may have been SCORPIUS - a successor to PLEIADES - that a euphoric Olivier Coenen had planned to let his LinkedIn network know about on Friday…
Olivier Coenen - BrainChip | LinkedIn
I build neural networks from first principles. I blend mathematical physics, information… · Experience: BrainChip · Education: University of California, San Diego · Location: San Diego County · 500+ connections on LinkedIn. View Olivier Coenen’s profile on LinkedIn, a professional community of...
Last edited:
Apologies if posted already.
(Original post translated to English)
www.iis.fraunhofer.de
www.linkedin.com
(Original post translated to English)
Fünf Branchen, die von Spiking Neural Networks profitieren
Wir stellen fünf Branchen vor, in denen Spiking Neural Networks den Unterschied machen können.
🧠⚡️Sie sind die neuronalen Netze der nächsten Generation: Spiking Neural Networks (SNNs) – vom menschlichen Gehirn inspiriert, rasend schnell und extrem energieeffizient. Doch lassen sich diese… | Fraunhofer IIS
🧠⚡️Sie sind die neuronalen Netze der nächsten Generation: Spiking Neural Networks (SNNs) – vom menschlichen Gehirn inspiriert, rasend schnell und extrem energieeffizient. Doch lassen sich diese Vorteile auch in der Anwendung nutzen? Die Zeiten, in denen SNNs eine rein akademische Übung waren...
Boab
I wish I could paint like Vincent
Unfortunately they are using SENNA as their accelerator.
www.iis.fraunhofer.de
SENNA – Spiking Neural Network Inference Accelerator
SENNA ist ein neuromorpher Chip für die ultraschnelle und energieeffiziente Verarbeitung niedrigdimensionaler Zeitreihendaten in KI-Anwendungen.
Nice article. Brainchip gets a mention 
https://prateekvishwakarma.tech/blog/small-language-models-edge-computing-2025-breakthrough/
Extract - “
Small Language Models Are Revolutionizing Edge Computing: The 2025 AI Breakthrough Everyone’s Talking About
September 28, 2025
The artificial intelligence landscape is experiencing a paradigm shift that’s quietly revolutionizing how we think about computing power and accessibility. While tech giants have been racing to build ever-larger language models, a counter-movement is gaining unprecedented momentum: small language models (SLMs) running on edge devices.
This isn’t just another tech trend—it’s a fundamental reimagining of how AI can be deployed, accessed, and utilized across industries. With the SLM market projected to explode from $0.93 billion in 2025 to $5.45 billion by 2032, representing a staggering 28.7% compound annual growth rate, we’re witnessing the birth of truly democratized artificial intelligence.
What Are Small Language Models and Why Do They Matter?
Small language models represent a strategic pivot from the “bigger is better” mentality that has dominated AI development. Unlike their massive counterparts that require cloud infrastructure and enormous computational resources, SLMs are designed to deliver impressive performance while operating within the constraints of edge devices—smartphones, IoT sensors, autonomous vehicles, and embedded systems.
The magic lies in their efficiency. While a large language model might contain hundreds of billions of parameters and require gigabytes of memory, a well-designed SLM can achieve remarkable results with just a few billion parameters, fitting comfortably on consumer hardware.
Key Characteristics of Effective SLMs:
Edge computing represents a fundamental shift in how we process and analyze data. Instead of sending information to distant cloud servers, edge computing brings processing power directly to the source of data generation. This architectural change is particularly crucial for AI applications that demand:
Breaking Down the Barriers: Advantages of SLMs at the Edge
1. Privacy-First AI Processing
One of the most compelling advantages of edge-deployed SLMs is their ability to process sensitive data without ever leaving the user’s device. This “privacy by design” approach is particularly crucial for:
By eliminating the need to communicate with distant servers, edge-based SLMs can deliver near-instantaneous responses. This speed improvement is critical for applications like:
Traditional large language models require expensive cloud infrastructure that scales linearly with usage. SLMs deployed at the edge flip this model by:
4. Enhanced Reliability and Availability
Edge-based SLMs continue functioning even when internet connectivity is unreliable or unavailable, making them ideal for:
Smart Manufacturing and Industry 4.0
Manufacturing facilities are increasingly adopting edge-deployed SLMs for:
The healthcare sector is embracing SLMs for edge applications including:
The automotive industry is leveraging edge SLMs for:
Urban planners are deploying SLMs at the edge for:
Hardware Limitations and Optimization Strategies
Deploying SLMs on edge devices presents unique technical challenges:
Memory Constraints: Edge devices typically have limited RAM and storage capacity. Solutions include:
Several advanced techniques are making SLMs more practical for edge deployment:
Quantization: Reducing the precision of model weights from 32-bit floating point to 8-bit integers or even binary representations, dramatically reducing memory requirements and computation time.
Pruning: Systematically removing less important neural network connections, creating sparse models that maintain performance while requiring fewer resources.
Knowledge Distillation: Training smaller “student” models to replicate the behavior of larger “teacher” models, transferring knowledge while reducing computational requirements.
Architecture Optimization: Designing model architectures specifically optimized for edge deployment, such as MobileNets, EfficientNets, and custom transformer variants.
The Neuromorphic Computing Revolution
A particularly exciting development in edge AI is the emergence of neuromorphic computing architectures. These brain-inspired processors offer remarkable energy efficiency and processing capabilities perfectly suited for SLM deployment.
Leading Neuromorphic Platforms:
https://prateekvishwakarma.tech/blog/small-language-models-edge-computing-2025-breakthrough/
Extract - “
Small Language Models Are Revolutionizing Edge Computing: The 2025 AI Breakthrough Everyone’s Talking About
September 28, 2025
The artificial intelligence landscape is experiencing a paradigm shift that’s quietly revolutionizing how we think about computing power and accessibility. While tech giants have been racing to build ever-larger language models, a counter-movement is gaining unprecedented momentum: small language models (SLMs) running on edge devices.
This isn’t just another tech trend—it’s a fundamental reimagining of how AI can be deployed, accessed, and utilized across industries. With the SLM market projected to explode from $0.93 billion in 2025 to $5.45 billion by 2032, representing a staggering 28.7% compound annual growth rate, we’re witnessing the birth of truly democratized artificial intelligence.
What Are Small Language Models and Why Do They Matter?
Small language models represent a strategic pivot from the “bigger is better” mentality that has dominated AI development. Unlike their massive counterparts that require cloud infrastructure and enormous computational resources, SLMs are designed to deliver impressive performance while operating within the constraints of edge devices—smartphones, IoT sensors, autonomous vehicles, and embedded systems.
The magic lies in their efficiency. While a large language model might contain hundreds of billions of parameters and require gigabytes of memory, a well-designed SLM can achieve remarkable results with just a few billion parameters, fitting comfortably on consumer hardware.
Key Characteristics of Effective SLMs:
- Parameter efficiency: Typically ranging from 1B to 20B parameters
- Memory optimization: Designed to run on devices with limited RAM
- Task-specific training: Fine-tuned for particular use cases rather than general knowledge
- Local processing: No internet connection required for inference
- Energy conscious: Optimized for battery-powered devices
Edge computing represents a fundamental shift in how we process and analyze data. Instead of sending information to distant cloud servers, edge computing brings processing power directly to the source of data generation. This architectural change is particularly crucial for AI applications that demand:
- Ultra-low latency responses
- Enhanced privacy and security
- Reduced bandwidth consumption
- Improved reliability in disconnected environments
- Real-time decision making
Breaking Down the Barriers: Advantages of SLMs at the Edge
1. Privacy-First AI Processing
One of the most compelling advantages of edge-deployed SLMs is their ability to process sensitive data without ever leaving the user’s device. This “privacy by design” approach is particularly crucial for:
- Healthcare applications handling patient data
- Financial services processing transaction information
- Personal assistants managing private communications
- Corporate environments with strict data governance requirements
By eliminating the need to communicate with distant servers, edge-based SLMs can deliver near-instantaneous responses. This speed improvement is critical for applications like:
- Autonomous vehicles making split-second navigation decisions
- Industrial automation systems requiring real-time monitoring
- Interactive gaming experiences with AI-powered NPCs
- Voice assistants providing immediate responses
Traditional large language models require expensive cloud infrastructure that scales linearly with usage. SLMs deployed at the edge flip this model by:
- Eliminating ongoing cloud computing costs
- Reducing bandwidth expenses
- Enabling offline functionality
- Providing predictable operational expenses
4. Enhanced Reliability and Availability
Edge-based SLMs continue functioning even when internet connectivity is unreliable or unavailable, making them ideal for:
- Remote industrial facilities
- Maritime and aviation applications
- Emergency response systems
- Rural deployment scenarios
Smart Manufacturing and Industry 4.0
Manufacturing facilities are increasingly adopting edge-deployed SLMs for:
- Quality control automation using vision models
- Predictive maintenance systems analyzing sensor data
- Supply chain optimization with local decision-making
- Worker safety monitoring through real-time analysis
The healthcare sector is embracing SLMs for edge applications including:
- Wearable health monitors providing instant insights
- Medical imaging analysis in resource-constrained settings
- Emergency triage systems offering immediate assessments
- Medication management with personalized recommendations
The automotive industry is leveraging edge SLMs for:
- Advanced driver assistance systems (ADAS)
- In-vehicle conversational AI
- Fleet management optimization
- Autonomous vehicle decision-making
Urban planners are deploying SLMs at the edge for:
- Traffic optimization systems
- Environmental monitoring networks
- Public safety applications
- Energy grid management
Hardware Limitations and Optimization Strategies
Deploying SLMs on edge devices presents unique technical challenges:
Memory Constraints: Edge devices typically have limited RAM and storage capacity. Solutions include:
- Model quantization techniques reducing precision requirements
- Knowledge distillation transferring large model capabilities to smaller architectures
- Dynamic loading of model components based on current needs
- Hardware acceleration through specialized AI chips
- Neuromorphic computing architectures mimicking brain efficiency
- Optimized inference engines designed for specific hardware platforms
- Event-driven processing reducing idle power consumption
- Adaptive computation scaling based on task complexity
- Hardware-software co-design optimizing the entire stack
Several advanced techniques are making SLMs more practical for edge deployment:
Quantization: Reducing the precision of model weights from 32-bit floating point to 8-bit integers or even binary representations, dramatically reducing memory requirements and computation time.
Pruning: Systematically removing less important neural network connections, creating sparse models that maintain performance while requiring fewer resources.
Knowledge Distillation: Training smaller “student” models to replicate the behavior of larger “teacher” models, transferring knowledge while reducing computational requirements.
Architecture Optimization: Designing model architectures specifically optimized for edge deployment, such as MobileNets, EfficientNets, and custom transformer variants.
The Neuromorphic Computing Revolution
A particularly exciting development in edge AI is the emergence of neuromorphic computing architectures. These brain-inspired processors offer remarkable energy efficiency and processing capabilities perfectly suited for SLM deployment.
Leading Neuromorphic Platforms:
- Intel Loihi 3: Supporting up to 10 million neurons, ideal for robotics and sensory processing
- IBM NorthPole: Featuring 256 million synapses, excelling in image and video analysis
- BrainChip Akida 2: Enabling on-chip learning for consumer devices
Another week where shareholders are left standing in the rain without any explanation of what that whole show was about… plenty of fuel for sellers and shorters. In terms of irresponsibility, this one tops the list so far — congratulations! Instead of satisfying investors, you just keep giving them more and more reasons to be angry.
BTW… has anyone actually written an email and asked for clarification? After all, his two statements were anything but trivial, I’d say. This casts a poor light to the company..also from the perspective of a potential investor…
Imo
BTW… has anyone actually written an email and asked for clarification? After all, his two statements were anything but trivial, I’d say. This casts a poor light to the company..also from the perspective of a potential investor…
Imo
Last edited:
Bennysmadness
Regular
What show?
smoothsailing18
Regular
.
All just dribble mate, manipulating dribble not worth a pinch of salt.
Lol you sounding like a broken record mate.
If you're not happy, then please do us a favour and sell up. Plenty of us would scoop your lot of shares with no trouble.
This thing is going to the moon.
As always, not advice. DYOR yadda fada etc..
A powerful demonstration of Brainchip technology showcasing the capabilities of our State-Space Network Model, TENN. I’m proud to be the inventor of the dynamical network PLEIADES, developed… | Olivier Coenen
A powerful demonstration of Brainchip technology showcasing the capabilities of our State-Space Network Model, TENN. I’m proud to be the inventor of the dynamical network PLEIADES, developed independently of other SSMs, which later evolved into TENN through collaborative development with my...
Similar threads
