Posts

Mistral AI: Unchallenged in quantized forms

 Llama 2 was talk of the town for a good time, when Open AI launched GPT there are worries that it’s planning to monopolise its capabilities. Its actions in the senate hearing, added fuel. Meta’s decision to giveaway Llama 2 gave a breather. It performed fairly well across all the parameters Now, comes the new sensation Mistral AI, the one which changes the whole thought of being API based models to becoming one to one running on personal PCs. Of course, if we want to give the live knowledge going on around the world we still have to power with browsing capabilities. Currently Mistral AI in quantized versions is of size 4GB and can easily on 16GB RAM. This is slightly higher than Llama 2 13B quantized version. Vectorised Databases for specific uses, has become a trouble as being conducted on specific parameter. However, inter-operability of specific vectorised datasets is something to watch for over the period. Euphoria of GenAI on Personal Computers is yet to be experienced, every...

Hallucination: Uncorrected AI Brains

I have tried to train the AI with specific unstructured data sets to give additional powers to the AI to my specific use case.  Training the GenAI to the specific configurations, and additional vectorised databases - results on the platform are very frickle with extreme hallucinations. Using and preparing the structured databases for huge data sets is costly, time consuming and man power demanding.. The bigger challenge from the GenAI is the problem of hallucination along with the P Value or any creativity configuration. As it more creative, the more chance of the AI going for hallucinating over the additional data. This can be more evident in the instance of GPT from OpenAI. This challenge is optimised to an extent by controlling the P Value and other creativity inducing configurations. However, results aren't satisfactory. Optimising with the prompt is always an option to ground it from hallucinations but in my specific case grounding is to be done at a scale of converting whole ...

A deeper thought into Project Alpha and Tech

 There was sometime in 2019, I went on to Instagram and posted a video announcing Project Alpha. Which I did, to get committed to the public on creating it by 2022. Of course, I missed the deadlines. Let me get into what it is: I planned an ecosystem to target a large number entities which generally have a single model of transactions within their books. I planned that with BOTS and FLOWS. Now things really changed from what could potentially be a product, is now taken over with the GPT transformation over the past 3 years. It started in nuance stage in 2018 with it making strides with GPT-2. Now, with Chat GPT it aced itself to such a level, where a single person is able to create a better products. Encountering a new challenge On a fine day, someone asked me what's next? CA Exams through me out of my own world. I am no where near to experimentation. I started feeling struck and suffocated. Then I closed my eyes and took up a position in Big4. Here I am today, looking back into th...

Dive into Stochastic Modeling | Markov Chain

Image
A stochastic modeling is a tool which allows randomness within it's system. A  Markov chain  is a mathematical system that experiences transitions from one state to another. T he probability of transitioning to any particular state is dependent solely on the current state and time elapsed. A Markov chain is a stochastic process, but it differs from a general stochastic process in that a Markov chain must be "memory-less." That is, (the probability of) future actions are not dependent upon the steps that led up to the present state. This is called the Markov property. The initial probabilities for the creation of transition matrix are obtained based on the historical data with sequence(transitions and not the pure data). Since, is a stochastic process - the change/transition is a core assumption. Since, the process is dependent on the current state, the initial vector for N=N*A => N=N. Here, A is considered to be 1, in other words at the time of start the information p...

Do you know a share of partnership firm can't be transferred?

Unlike the shares of a public/private company, partner’s share in the firm cannot be transferred.  Why? The provisions of section 29 of the Indian Partnership Act, 1932 and section 42 of the Limited Liability Partnership Act, 2008 permit a partner to transfer its right to share profit/loss and to receive distribution from the firm.  However, such rights do not result in the assignee becoming a partner in the firm/ LLP and also does not entitle the assignee to participate in the management or conduct of various activities of the firm.  Since the intention is to acquire business, the acquirer would seek to participate in the management and such transfer of interest in sharing profit/loss etc. may not be relevant where the intention is transfer of ownership.  However, you can do it in another way! Hence, to obtain management control, the acquirer should become a partner in the Firm and the transferor/s should retire from the Firm.

Rationale: GST impact in case of Slump Sale/Demerger Scenario

 The GST Tax exemption is available for the transactions involved in the transfer of business undertaking as a whole, wherein all the assets and liabilities of the company are transferred by the target company to the acquirer. In the case, where any particular liability or asset not part of the transferred business/group of assets challenges the provision of exemption mentioned above. However, rationale logic can't be given for the same in the act, if any can be easily become another provision have to face the continuous loss of purpose by the parties in the transaction. Hence, this gives raise to the uncertainty in the Mergers and Acquisitions transactions. As due to the above, it is ordinarily suggested to get AAR before the transaction being entered into. Rationale: Question raised on whether the supply is of goods or services or not? Why? To resolve Sec.7 - Scope of Supply In Sri Ram Sahai vs. Commissioner of Sales Tax [(1963) 14 STC 275 (All)] held that ‘business’ is admitte...

How is RBI regulating BNPL industry?

The Reserve Bank of India (RBI) has issued various guidelines and regulations for the operation of buy now pay later (BNPL) providers in India. Specific measures that the RBI has taken to regulate the BNPL industry include: Issuing guidelines on the operation of BNPL schemes: The RBI has issued guidelines on the operation of BNPL schemes, which outline the requirements that BNPL providers must follow in order to offer their services to consumers. These guidelines cover areas such as customer onboarding, credit risk assessment, and pricing of BNPL products. Requiring BNPL providers to obtain authorization: BNPL providers in India are required to obtain authorization from the RBI in order to operate. This process involves submitting an application to the RBI, which is then reviewed and approved or rejected based on the RBI's evaluation of the provider's financial stability, business model, and other factors. Supervising BNPL providers: The RBI monitors the activities of BNPL prov...