|
|
We present the world's first truly open source Ai...
Fact: for generative technology to be optimally impactful, language models must have ample wiggle room to predict and generalize.
Caveat: power without control is nothing. To place an LLM in the midst of business-critical infrastructure and grant it limitless ability to effect changes is to be willfully destructive.
|
Meet Tinc Tanc's Open Source: optimizing language model deployment without losing control.
With Open Source, a system layer - situated between the Ai layer and other infrastructure - continuously records all model/system/user interactions as they happen as well as all modifications to data or system state.
The output of the audit is in human-readable prose, meaning any user can review what is happening behind the scenes as they work on Tinc Tanc at the click of a button.
|
|
|
How Open Source Works
Tinc Tanc's Ai architecture is built on a technique we call process granulation, which decomposes every objective into the minutest building blocks and then reassembles them with generation. It means the natural state of the task is deliberately altered in order for it to conform to a mode that stays within guardrails no matter how much the language model alters it. If a building block is defective, it is defective everywhere it is used in generation. But there will never be technical bias.
One of the numerous "side-effects" of our technique is that it becomes possible to audit everything the model does in real time. What's more: the audit is entirely exclusive of the generative pipeline. That is, the language model has no input in it. Because the audit exists exclusively on the system layer.
Core Deliverables Of Open Source
- Validation receipts
Remember, every solution users create on Tinc Tanc is automatically bundled into modular knowledge packs that can easily be shared and monetized with third parties. Open source receipts are empirical proof which leave no ambiguity as to what those bundled knowledge packs contain. They can prove invaluable with product validation and procurement teams.
- Generic model training
Premium users can mass-download the open source scripts accumulated as they continuously create business solutions. These scripts, especially when they relate to the same subject matter, can be used to train Ai systems outside of Tinc Tanc. Think about a flow cytometry technique, or manual for connecting to a scientific instrument, or a novel accounting concept. These scenarios can be hypothesized out on Tinc Tanc for the specific purpose of gathering enough open source scripts that would enable any language model master those same SOPs.
- More precision; less hallucination
Many global corporations, especially in industries where technical precision is a must-have for financial data integrity and consumer safety reasons, already routinely generate precision-guided training scripts on Tinc Tanc for use in training their internal Ai systems. This approach not only enhances technical precision for models. It also helps to avoid model hallucination.
See simple illustration
Watch video
|
|
|