Although a number of legal tech companies use LLM consensus and multiple model strategies already, Pilot5 and its new adversarial Legal AI product specifically taps five frontier models with ‘one assigned to attack the emerging consensus and the lawyer receives a single recommendation’ to their query.
Moreover, the system also retains on record the strongest counterargument to any result.
This is how it works:
- Pilot5 builds a structured understanding of the matter:
- which side the lawyer represents,
- what the lawyer is trying to achieve,
- why the question is being asked,
- the relevant jurisdiction and legal context, deadlines and constraints,
- what is already known or decided,
- the facts or assumptions that could materially change the analysis.
- That context becomes the brief given to the panel of LLMs.
- Then, five AI models analyze the matter separately from different angles.
- One, the Contrarian, is assigned to attack the emerging consensus and to bring the authority that cuts against it.
- The lawyer can read each model’s first-round analysis as it lands, before the models have seen one another’s work.
- The models then critique each other’s analysis without knowing who wrote it, revise their positions under challenge.
- Finally it produces a final synthesis with one recommendation, the sources behind it, and the strongest counterargument preserved on the record.
They noted that a complex ‘deliberation’ typically costs about $7 and takes about fifteen minutes. But, there are also some more quick and economical options, such as:
- ‘Statutory Review – $0.50 – Statutory Review reads an employment agreement, such as a separation agreement, clause by clause against potentially applicable federal requirements and restrictions, including the OWBPA/ADEA, FCRA, DTSA and Speak Out Act, together with the law of the selected state (eleven states and the District of Columbia at launch). Each requirement is marked satisfied, partial, needs review or missing, with the relevant statutory or regulatory text quoted from an official source; where the text could not be retrieved, the review says so.
And also:
- Contract Understanding, up $0.20 – Contract Understanding identifies what each numbered clause in a contract does while keeping the source wording intact.’
They added that Pilot5 does not use client files to train models. A Zero Data Retention mode for deliberations and a data processing agreement are available on request, they added.
Wilfrid Rimbault and Thomas Wainstein, Co-CEOs of Pilot5, commented: ‘Our line to lawyers is simple: moot your case before you commit to it. A lawyer should not hear the other side’s best argument for the first time in court. Pilot5 Legal puts that argument in front of them before they commit, with the authority alongside it, for about seven dollars.’
Is this a big deal?
As noted, legal tech companies have been using a combination of multiple models to come to the best answer to a complex legal question for a while, including some companies using a ‘judge and jury’ system of LLMs where one model challenges the result of another model.
That said, Pilot5 is offering something standalone that is very distinct and highly structured. The idea of different models arguing things out so explicitly may appeal to lawyers who like that approach, and particularly where they are not looking to bring aboard one of the larger legal AI platforms, and prefer something lighter and more economical.
More about Pilot5 here.
—
Legal Innovators UK – Nov 4 and 5 / Legal Innovators New York – Nov 17 and 18
Cosmonauts have now opened Express Registration for Legal Innovators UK and New York. This means that if you are at a law firm or at an inhouse legal team you can now zoom directly to your tickets – which are complimentary as long as you qualify.
For London – Nov 4 and 5, at the Minster Building, see here:
And, for New York – Nov 17 and 18 – Midtown, please see here:
Discover more from Artificial Lawyer
Subscribe to get the latest posts sent to your email.