WHY DEEPSEEK LURE IS ANATHEMA

The Chinese AI startup DeepSeek caught a lot of people by surprise lately. Its new model, released on January 20, 2025, competes with models from leading American AI (Artificial Intelligence) companies such as OpenAI and Meta despite being smaller, more efficient and much cheaper to both train and run.

The furore over DeepSeek could likely have been predicted by the much talked about and widely accepted management theory — the theory of disruptive innovation. After all, disruptive innovation is all about low-cost alternatives that aren’t cutting-edge but perform adequately for many users. This, it seems, is exactly how DeepSeek has created the shockwave that has challenged some of the assumptions of the American AI industry and sent tech and energy stocks tumbling as a result.

In Alphabet Incorporated, the parent company of Google, stocks plunged by more than 8% recently, marking the greatest percentage fall since October 2023. The decline, for Google parent Alphabet, erased as much as USD 211 billion in market value. According to reports, the decrease is the company's biggest one-day value decline. In midday trading, Alphabet shares finished at USD 193, down 7.6% at USD 190.70.

Chipmaking giant Nvidia had a similar outcome, after the release of China's ground-breaking AI model DeepSeek R1 - which completely turned the AI industry on its head.

Nobel laureate Demis Hassabis, the head of Google’s AI research lab DeepMind, argued that DeepSeek misled people by reporting it was made at a fraction of the cost of Google’s DeepMind and OpenAI.

In an interview Hassabis, who oversees Google’s AI division, explained that the Chinese start-up "only reported the cost of the final training round, which is just a small part of the total cost."

DeepSeek claimed to have spent USD 5.6 million on computing costs to train its model using older Nvidia chips. 

The key to  DeepSeek’s working is Test Time Compute. Test time compute operates through two powerful mechanisms that fundamentally change how language models approach problem-solving. The first mechanism involves refining the proposal distribution, where models iteratively improve their answers through guided self-revision. During this process, the model generates a sequence of revisions, with each attempt building on insights from previous ones. This sequential approach is particularly effective when the base model has a reasonable initial understanding but needs refinement to reach the correct answer. Research has shown that by allowing models to dynamically modify their output distribution based on previous attempts, they can achieve improvements up to 4 times in efficiency compared to standard parallel sampling approaches.

The second key mechanism focuses on optimizing verifier search through Process Reward Models (PRMs). Unlike traditional output verification that only judges final answers, PRMs evaluate the correctness of each intermediate step in a solution. These dense, step-wise reward signals enable sophisticated tree search algorithms like beam search and lookahead search to explore multiple solution paths simultaneously. The effectiveness of these search strategies varies with problem difficulty – beam search, which maintains multiple candidate solutions at each step, often outperforms simpler approaches on harder problems but can lead to over-optimization on easier ones. Meanwhile, lookahead search, which simulates future steps to evaluate current decisions, helps prevent the model from getting stuck in local optima but requires more computational resources.

                             The combination of these mechanisms creates a powerful synergy.

While refining the proposal distribution helps the model generate better initial solutions, the verifier search ensures these improvements are systematic and well-directed. The ideal balance between these approaches depends critically on the problem's difficulty level. For easier problems, putting more emphasis on sequential revisions often yields better results, while harder problems benefit from more extensive verifier-guided search. Advanced implementations can dynamically adjust this balance based on the model's confidence and early performance indicators.

For DeepSeek, it has been found that the algorithms explode during distillation to arrive at the right, extremely precise solutions !!

For Lian Wenfeng (below), the founder of DeepSeek, it has been a matter of redemption. He first made his mark in China’s investment world in the late 2010s, co-founding a hedge fund that used artificial intelligence models. In 2023 he poured money into Artificial Intelligence and assembled a team to build China’s answer to Silicon Valley frontrunner OpenAI.

His 2010 thesis at Zhejiang University took on a topic of interest – Improving intelligent tracking algorithms for surveillance cameras. He took off from there.

“Providing cloud services is not our main goal. Our aim is still to achieve AGI (Artificial General Intelligence – AGI is a type of AI that possesses broad cognitive abilities similar to humans, capable of learning and performing a wide range of tasks across different domains, whereas the present other AIs usually specialize in one area). Everyone has their own unique journey and brings their own ideas with them. There is no need for a push” – commented Wenfeng when asked if DeepSeek were extracting codes from OpenAI.

This has furiously unsettled the workings of the industry bigwigs. Microsoft and OpenAI are investigating whether DeepSeek has used OpenAI’s API to integrate OpenAI’s artificial intelligence model into DeepSeek’s own model (API - Application Programming Interface – is a way for two systems to communicate with each other. APIs are used to share data and access resources).

OpenAI has found evidence linking DeepSeek to the use of distillation – a technique that developers use to train AI models by extracting data from larger, more accomplished ones.

This is an efficient way to train smaller models at a very low cost.

Using OpenAI’s API to distil the outputs to build rival models is a violation of OpenAI’s terms of service.

What is Artificial Intelligence(?). According to the father of Artificial Intelligence John McCarthy, it is the science and engineering of making intelligent machines, especially intelligent computer programmes.

Artificial Intelligence is a way of making a computer, a computer-controlled robot, or a software which thinks intelligently in the same manner the intelligent humans think.

AIs are skilled to find out how human brain thinks, how humans learn, decide, and work while trying to solve a problem and then use these outcomes to frame the basis of intelligent softwares and systems.

Gert-Jan Oskam is a brilliant example (!). Oskam (picture below)  had been paralyzed at his legs for more than a decade. It happened after he suffered a spinal cord injury during a bicycle accident. Thanks to artificial intelligence technology, he can now walk naturally, take on difficult terrain and even climb stairs - a freedom that he did not have before.

How it all happened. The path breaking research on AI combined the spinal implant with the new technology called brain-computer interface. This was implanted above the part of the brain that controlled leg movement. The interface had the ability to decode brain recordings in real time. That allowed the interface designed by researchers at France's Atomic Energy Commission (CEA) to work out how the patient wanted to move his legs at any moment in real time.

The modus operandi. Uninjured nerves send signals to muscles. Those signals make muscles move. When humans are paralyzed, or have paralysis, they can’t move certain parts of their body. There is a problem with the nervous system in the body’s command and communication system. Normally the nervous system sends signals from the brain throughout the body, telling what to do.

If something damages the nervous system, messages can’t get through to muscles.

When AI was introduced into the system of Oskam it created Artificial Neural Networks (ANNs) to help him.

Who invented Artificial Neural Network – the veritable life saver(?). The inventor of the first neurocomputer, Dr. Robert Hecht-Nielsen, defines a neural network as a computing system made up of a number of simple, interconnected processing elements, which process information by their dynamic response to external inputs.

Artificial Neural Networks (ANNs) have the ability to think, re-think and improve upon the responses it has already created based on the external factors!! It will go on changing and improving the network to yield optimum results.

The logic of ANN  is based on the belief that working of human brain can be understood by making the right connections and can be imitated using silicon and wires as living neurons and dendrites.

The human brain is composed of 100 billion nerve cells called neurons. They are connected to other thousand cells by axons (picture above). Stimuli from external environment or inputs from sensory organs are accepted by dendrites. These inputs create electric impulses which quickly travel through the neural networks. A neuron can then send (or does not send) the message to other neuron to handle the issue at hand.

                              So, ANN as such, is not at all a bad idea. Its wrongful usage is !!

The tragedy of DeepSeek is that it is on a price war with OpenAI. DeepSeek is offering the same, or rather better, services at a hugely reduced price . Since it is an open-source platform any user can use it whereas in OpenAI platform the services are to be bought at a ridiculously high price.

DeepSeek has done nothing to enhance the usability of medical research – especially in the area of cancer treatment. Using artificial neural networks we can actually reverse the onset of cancer by altering the chromosome structure in a human DNA.

Take this brilliant example. We know that every human possesses 23 pairs of chromosomes that are foundations of our DNA curled in a double helix structure.

The amusing demeanour of 22nd chromosome(!). It is one of the smallest and arguably one of the most widely discussed of all human chromosomes. In case of blood cancer (leukaemia) the 22nd chromosome breaks away and curls towards the 9th chromosome and sits with the 9th! The DNA structure gets disturbed and the sitting point gets weak. This creates enough ground for the enzymes to attack the point which leads to profuse multiplication of white blood cells in the bone marrow.

This new chromosome is called Philadelphia chromosome and it contains the fused gene BCR-ABL. This gene is the ABL gene of chromosome 9 juxtaposed onto the breakpoint cluster region of BCR gene of chromosome 22. This fusion creates the hybrid protein tyrosine kinase – often termed as the brutal protein – the cause of cancer.

Artificial Neural Networks can play a big role to stem the rot. Every human being does not suffer from cancer. The breakage of chromosome 22 and 9 happens when in specific cases the area of chromosome 9 is by nature weak for a target cancer human (nicknamed Ollie, let us assume) as compared to another human (nicknamed Pepe) whose chromosome 9 is relatively stronger in the DNA structure.

                                                        What Ollie can do and how ANN can help?

Ollie can deposit regular DNA samples to a Lab where it is fed to an AI controlled ANN system and the neural computer can actually predict the probability of Ollie’s 9th getting raided by the 22nd. The degree of probability (45% or 90%) will help the doctor to start medication even before cancer has stepped in by prescribing medicines which would stop chromosome 22 to sit alongside 9 and create that brutal protein, spelling doom. 

So all of artificial intelligence (ANNs) is not bad, but the arrogant stance of DeepSeek is disappointing!

There are concerns. Advancements in AI for lethal weapon development can cause many unfortunate situations. Battles and wars would be more menacing and human mortality would increase manifold.

Humans would be plucked off from real world. We might as well live in a bubble, in a virtual world surrounded by machines and artilleries never knowing who would my next neighbour be and how I work to survive in my own bubble.

Machines would be taking lot of human jobs. Many of the human jobs would be taken over by robots. Human creativity would be at its lowest. The man-machine friendship would die and one would not be surprised if robots created by humans start annihilating humans. Sam Altman, CEO of OpenAI and creator of ChatGPT is optimistic when he says AI will create new jobs but some old ones may die fast.

By the year 2119, future will be way different than we can see and imagine currently. With such great advancements in technology - automated cars, space travels and what not it would be interesting to note how humanity would respond under these troubled circumstances. Constantly interacting with machines would make us machines(?), how our children would adapt – they would have fewer interactions with educators and classmates, there are instances of young people becoming violent under the influence of excruciatingly painful contents, whether marriages would be between two machines(?). Will humans would still be humble and polite or be cruel and ravenous?

                                                                      It is a basket of worries !!

Time will tell if the future of our children is dark or bright. It would depend on how we handle AI. After all, there is a bright side to every dark one !

On the contrary, times ahead could be much better than what we have now, the times we never imagined would come to us!!

Disclaimer: The names Ollie and Pepe are imaginary and do not resemble any person(s) dead or alive.