A couple of weeks in the past, I wrote here concerning the imminent launch by Josef of the beta model of Josef Q, its new product that makes use of superior AI to rework insurance policies and laws — similar to these pertaining to privateness, information safety, HR and procurement — into digital Q&A instruments, in order that staff or purchasers can simply get solutions to coverage and compliance questions.
As we speak, that launch arrives, as Josef makes the beta model accessible for anyone to try for free. The product will stay free all through its as-yet undetermined beta interval, after which Josef plans to start promoting the product.
An Australia-based firm whose main product is a no-code automation platform for authorized professionals, Josef has partnered with OpenAI to develop the instrument, which is presently utilizing GPT-4, the most recent model of Open AI’s giant language mannequin (LLM). Nonetheless, Josef is “remaining agnostic” about which LLM it makes use of and should experiment with different corporations’ fashions, Sam Flynn, cofounder and chief working officer, advised me.
The corporate sees Josef Q as notably helpful to companies, the place authorized, compliance, data safety, and HR groups can use it to supply staff with prompt solutions about company insurance policies and procedures.
It can be utilized by regulation corporations, each to reply inner worker questions and as a client-facing instrument to supply solutions to coverage and regulatory questions based mostly on the regulation agency’s information and experience.
Flynn mentioned the instrument may even be utilized by authorized assist organizations to supply customers with solutions to questions in areas similar to landlord-tenant regulation.
How It Works
In a quick demonstration yesterday, Flynn confirmed me that making a digital Q&A instrument utilizing Josef Q takes simply minutes and requires solely three steps.
First, the person uploads the coverage doc, similar to a privateness coverage or parental depart coverage, to be analyzed by the LLM. In Flynn’s demonstration, he used a parental depart coverage.
Second, the person trains the instrument by asking three questions. The questions ought to replicate those who others are prone to ask of the instrument. For the parental depart instance, one query is perhaps, “Am I eligible for parental depart?” Josef Q will counsel a solution, which the person can edit if want be.
As soon as the three questions have been requested and answered, the instrument is educated and able to launch.
Josef Q makes use of “human-in-the-loop” coaching and moderation. Even after the preliminary three-question coaching, the moderator can go in and see all of the questions which have been submitted and the solutions Josef Q has given, and refine the mannequin by correcting or modifying any of the solutions.
Additionally, the one who submits the query might help additional prepare the instrument by giving a thumbs up or down to point whether or not the reply was useful.
There isn’t a hazard of Josef Q “hallucinating” and making up solutions, Flynn advised me, as a result of the solutions it supplies will all the time be restricted to the coverage doc. Thus, if somebody have been to ask the parental depart Q&A the identify of the present U.S. president, it could reply, “I’m sorry, I don’t have a solution.”
A future model of the instrument will even embody confidence scores for every reply.
Throughout the beta interval, Josef is capping the dimensions of a doc that may be uploaded to eight,000 characters. As soon as the product is prepared for basic launch, customers will be capable of add any dimension doc and even a number of paperwork associated to a coverage.
See my prior submit for extra details about Josef Q.