Large Language Models or LLMs are a type of generative artificial intelligence.
Large language models are artificial neural networks that can process natural language and generate text. They are composed of millions or billions of parameters that are learned from a large corpus of text data. The parameters represent the weights and biases of the connections between the neurons in the network. The network takes a sequence of words or tokens as input and outputs another sequence of words or tokens as the most likely continuation. The network can also condition the output on some context, such as a prompt, a query, or a task.
Large language models are trained using a technique called self-supervised learning, which means that they learn from their own data without requiring any human annotations or labels.
The most common objective function for training large language models is called masked language modeling (MLM), which involves randomly masking some words or tokens in the input and asking the network to predict them. This way, the network learns to capture the syntactic and semantic relationships between words and tokens in the language.
Another common objective function is called next sentence prediction (NSP), which involves giving the network two sentences and asking it to predict whether they are consecutive or not. This way, the network learns to capture the coherence and structure of the text.
Large language models have many applications in natural language processing and beyond. Some of the most popular applications are:
· Text generation: Large language models can generate fluent and coherent text on various topics and domains, such as stories, summaries, headlines, captions, reviews, etc.
· Text completion: Large language models can complete partial or incomplete text, such as sentences, paragraphs, emails, essays, etc.
· Text summarization: Large language models can produce concise and informative summaries of longer texts, such as articles, reports, books, etc.
· Text paraphrasing: Large language models can restate the same meaning of a text in different words, such as synonyms, reordering, simplification, etc.
· Text translation: Large language models can translate text from one language to another, such as English to French, Spanish to Chinese, etc.
· Text classification: Large language models can assign labels or categories to text, such as sentiment, topic, genre, intent, etc.
· Text extraction: Large language models can extract specific information from text, such as names, dates, numbers, facts, etc.
· Text question answering: Large language models can answer questions based on text, such as trivia, facts, definitions, etc.
· Text dialogue: Large language models can engage in natural and conversational dialogue with humans or other agents, such as chatbots, assistants, games, etc.
LLMs have become available for general use. ChatGPT and Microsoft CoPilot are a couple of examples. LLMs are being deployed commercially, especially in areas such as law where language processing and developing information is a basic requirement.
Lawyers must be careful with the use of LLMs. In December 2023 the Judiciary published guidelines for the use of AI tools of which LLMs (also known as Generative AI) are one.
Chief Justice Dame Helen Winkelmann said, upon the release of the Guidelines:
“The use of new GenAI technology in the Aotearoa New Zealand legal context is increasing. These guidelines have been developed to help judicial officers, lawyers and non-lawyers appropriately use GenAI tools in court or tribunal proceedings.”
“GenAI chatbots are not a substitute for a qualified lawyer, but there is potential for GenAI – when used responsibly – to enhance access to justice by making legal knowledge and information more accessible to non-lawyers.”
Justice Paul Radich, Chair of the judiciary’s AI Advisory Group who led development of the guidelines said:
“These guidelines offer practical advice in plain language on what GenAI is, the risks inherent in its use, and situations in which its use may be appropriate or inappropriate. New Zealand is amongst the first to develop guidelines, although all jurisdictions are navigating the opportunities and challenges presented by GenAI technology.”
One of the problems with LLMs is that of hallucination. There have been instances where LLMs have been used to develop submissions for a Court where false, incorrect or non-existent citations of case have been offered. The best-known of these cases is the highly publicised the US District Court decision Mata v Avianca Case No. 22-cv- 1461 (PKC) (S.D.N.Y), where lawyers submitted non-existent judicial opinions with fake quotes and citations created by the AI tool ChatGPT.
In Mata the Court noted:
Many harms flow from the submission of fake opinions. The opposing party wastes time and money in exposing the deception. The Court’s time is taken from other important endeavors. The client may be deprived of arguments based on authentic judicial precedents. There is potential harm to the reputation of judges and courts whose names are falsely invoked as authors of the bogus opinions and to the reputation of a party attributed with fictional conduct. It promotes cynicism about the legal profession and the…judicial system. And a future litigant may be tempted to defy a judicial ruling by disingenuously claiming doubt about its authenticity.
A similar case arose in England. The case of Harber v Commissions for His Majesty’s Revenue and Customs [2023] UKFTT 1007 (TC) is the first reported UK decision to have found that cases cited by a litigant were not genuine judgments, but had been created by an AI system.
The case is a First-tier Tribunal Tax Chamber judgment, where a self-represented appellant taxpayer had provided the tribunal with the citations and summaries for nine First-tier Tribunal decisions which supported her position. While these citations and summaries had “some traits that [were] superficially consistent with actual judicial decisions”, they were ultimately found not to be genuine case citations.
In Harber, the respondent gave evidence that each of the cases provided by the self-represented appellant had been checked against the cases in the Tribunal data, using not only the appellants’ fictious party names and the year, but also extending the search to several years on either side – but were unable to find such cases.
The Tribunal itself also carried out checks for those cases in its own database and within other legal repositories.
The litigant in person accepted that it was possible that the cases had been generated by AI, stating that the cases were provided to her by “a friend in a solicitor’s office” whom she had asked to assist with her appeal; and had no alternative explanation as to why the cases could not be located on any available database of tribunal judgments.
The Tribunal ultimately concluded that the cited cases did not exist and had been generated by an AI system such as ChatGPT but accepted that the taxpayer was unaware that the AI cases were fabricated, and ultimately the incident made no difference to her unsuccessful appeal. However, the Tribunal made clear that citing invented judgments was not harmless, causing “the Tribunal and HMRC to waste time and public money, and this reduces the resources available to progress the cases of other court users who are waiting for their appeals to be determined”.
The Tribunal also highlighted other problems caused by citing hallucinated judgments; endorsing comments made in Mata v Avianca.
This highlights a difficulty with LLMs and Generative AI tools. As Jonas Andrulis Founder and Chief Executive Officer of Aleph Alpha, said in evidence before the House of Lords Communications and Digital Select Committee on 21 November 2023
“Based on our own research, we developed a method to visualise the patterns that the machine learned. They are not truth machines, which is also why they hallucinate, but they are pattern learners.”
And later
“The method of training the systems to arrive at large language models has nothing to do with truth; it is just to learn patterns of language and complete writing according to learned patterns. That is also why these models and their outputs are not consistent. They can contradict themselves, because they are not built as truth machines. As humans we care about truth, so we need to do something about this, because hallucinations are a problem for us. This is important for us to understand. Hallucinations are not against the learning method for the system, because the system is not built on truth.”
As long as these shortcomings are understood nevertheless LLMs have helpful applications in legal practice.
Some of the advantages are as follows:
1. Enhanced Efficiency: LLMs can significantly improve the efficiency of legal tasks. They can quickly analyze and process vast amounts of legal information, including case law, statutes, regulations, and legal documents. This can save lawyers valuable time in conducting legal research, drafting documents, and analyzing complex legal issues.
2. Cost Savings: By automating routine legal tasks, LLMs have the potential to reduce costs for legal services. Lawyers can leverage LLMs to handle repetitive and time-consuming tasks, allowing them to focus on higher-value work. This can lead to increased productivity and cost savings for both law firms and clients.
3. Improved Accuracy and Consistency: LLMs can provide accurate and consistent results in legal research and analysis. They can access a wide range of legal information and identify relevant precedents, statutes, and regulations with speed and precision. This can help ensure that legal arguments and advice are based on comprehensive and up-to-date information.
4. Access to Legal Knowledge: LLMs have the ability to process and understand vast amounts of legal knowledge, making it more accessible to lawyers. They can assist in retrieving relevant legal information, summarizing complex legal concepts, and providing insights into legal issues. This can empower lawyers to make more informed decisions and provide better legal services to their clients.
5. Augmented Decision-Making: LLMs can serve as valuable tools for lawyers in decision-making processes. They can analyze different legal scenarios, assess potential outcomes, and provide recommendations based on legal precedents and data. This can support lawyers in developing effective legal strategies and advising clients on the best course of action.
6. Scalability and Availability: LLMs can be easily scaled and made available to a wide range of legal professionals. Once developed, they can be deployed across different practice areas and jurisdictions, providing access to legal knowledge and expertise to a broader audience. This can help level the playing field and improve access to legal services.
It is important to note that while LLMs offer these benefits, they are not intended to replace human lawyers. LLMs are tools that can augment and support legal professionals in their work, but human judgment, expertise, and ethical considerations remain essential in the practice of law.
Lawyers can use language models for various tasks in their legal practice. Here are some examples:
1. Legal Research: Language models can assist lawyers in conducting legal research by quickly analyzing and processing vast amounts of legal information. They can help identify relevant case law, statutes, regulations, and legal precedents, saving time and providing comprehensive insights.
2. Document Drafting: Lawyers can leverage language models to draft legal documents such as contracts, pleadings, and briefs. Language models can provide suggestions, generate templates, and assist in structuring and formatting legal documents, improving efficiency and accuracy.
3. Due Diligence: Language models can aid lawyers in conducting due diligence for transactions, mergers, and acquisitions. They can analyze large volumes of documents, contracts, and agreements to identify potential risks, inconsistencies, or relevant information, streamlining the due diligence process.
4. Contract Analysis: Lawyers can use language models to analyze and review contracts. Language models can help identify key terms, clauses, and potential issues, ensuring compliance and mitigating risks in contract negotiations and management.
5. Legal Opinion and Case Strategy: Language models can provide lawyers with insights and analysis to support legal opinions and case strategies. They can assist in predicting case outcomes, assessing legal arguments, and identifying relevant precedents, helping lawyers make informed decisions and develop effective legal strategies.
6. Legal Compliance: Language models can assist lawyers in ensuring legal compliance by analyzing regulations, policies, and guidelines. They can help identify potential compliance issues, provide guidance on regulatory requirements, and support the development of compliance programs.
It is important to note that while language models can be valuable tools in these tasks, they should be used in conjunction with human expertise and judgment. Lawyers play a critical role in interpreting and applying legal principles, considering ethical considerations, and providing personalized advice to clients.
Although these advantages may seem obvious the reality is that LLMs will have a disruptive effect upon legal practice.
The duration of disruption caused by LLMs (Large Language Models) in the legal profession is uncertain and can vary depending on several factors. Here are some considerations:
1. Adoption and Integration: The speed and extent of LLM adoption and integration into legal practice will influence the duration of disruption. If LLMs are rapidly adopted and seamlessly integrated into existing workflows, the disruption may be relatively short-lived. However, if adoption is slower or if there are challenges in integrating LLMs effectively, the disruption may persist for a longer period.
2. Technological Advancements: The pace of technological advancements in LLMs will also impact the duration of disruption. As LLMs continue to evolve and improve, their capabilities may expand, leading to more significant disruptions in the legal profession. If LLMs become increasingly sophisticated and capable of performing complex legal tasks, the disruption may be more prolonged.
3. Ethical and Regulatory Considerations: Ethical and regulatory considerations surrounding the use of LLMs in legal practice may also affect the duration of disruption. As the legal profession grapples with questions of professional responsibility, accountability, and the ethical implications of relying on LLMs, it may take time to establish guidelines and regulations that govern their use. These considerations could prolong the disruption as the legal profession adapts to new ethical and regulatory frameworks.
4. Skill Adaptation and Evolution: Lawyers' ability to adapt and evolve their skills in response to LLMs will also impact the duration of disruption. Lawyers who can effectively leverage LLMs as tools to enhance their practice and complement their expertise may experience a shorter disruption. However, those who resist or struggle to adapt to the changing landscape may face a more prolonged disruption.
It is important to note that the disruption caused by LLMs may not be a linear process. There may be periods of rapid change followed by periods of stabilization as the legal profession adjusts to the new technology. Additionally, the extent of disruption may vary across different practice areas and jurisdictions.
Overall, it is challenging to predict the exact duration of disruption caused by LLMs in the legal profession. However, it is likely that the impact of LLMs will continue to be felt for a significant period as the technology evolves, adoption increases, and the legal profession adapts to the changing landscape.
Eventually, LLMs will push lawyers into highly specialized and nuanced roles. After fully mature LLMs arrive, the lawyer will continue to play a central role in legal practice, but only in non-routine legal tasks. These tasks will primarily involve value judgments, such as the development of precedent or its reversal, or the allocation of property and other scarce resources. This new mix of lawyer-machine labor, where machines primarily carry out routine legal tasks, and lawyers handle the non-routine, will give rise to a growing demand for lawyers who can exercise good judgment and empathize with the winners and losers of social change
Here are some potential roles that lawyers may have:
1. Legal Strategy and Counseling: Lawyers will continue to provide strategic advice and counseling to clients. LLMs may assist in providing legal research and analysis, but lawyers will use their expertise and experience to guide clients through complex legal issues, assess risks, and develop effective legal strategies.
2. Client Communication and Advocacy: Lawyers will still be responsible for communicating with clients, understanding their needs, and advocating on their behalf. LLMs may assist in drafting legal documents, but lawyers will ensure that the documents accurately reflect the client's goals and effectively present their arguments.
3. Ethical and Professional Responsibility: Lawyers have ethical and professional obligations to their clients, the legal system, and society. They will continue to uphold these responsibilities, ensuring that legal services are provided with integrity, confidentiality, and competence. LLMs cannot replace the ethical judgment and decision-making required in legal practice.
4. Negotiation and Dispute Resolution: Lawyers will continue to play a crucial role in negotiation and dispute resolution processes. LLMs may assist in analyzing legal issues and providing potential outcomes, but lawyers will use their negotiation skills and legal knowledge to advocate for their clients' interests and work towards favorable resolutions.
5. Specialized Legal Expertise: Lawyers with specialized expertise in specific areas of law will still be in demand. LLMs may provide general legal knowledge, but lawyers with deep knowledge and experience in niche areas will be sought after for their specialized advice and representation.
6. Legal Compliance and Risk Management: Lawyers will continue to help clients navigate complex legal frameworks, ensure compliance with laws and regulations, and manage legal risks. LLMs may assist in researching and interpreting laws, but lawyers will provide guidance on how to apply the law to specific situations and mitigate legal risks.
It is important to note that the arrival of fully mature LLMs may change the way lawyers work and the tasks they perform. Lawyers may rely more on LLMs for legal research and document drafting, allowing them to focus on higher-level tasks that require human judgment, creativity, and interpersonal skills. The exact roles of lawyers in legal practice will depend on the development and adoption of LLM technology, as well as the evolving needs of clients and the legal profession.
I guess LLMs (Masters of Laws) are mainly the product of the other LLMs (Large Language Models) these days Halfling! 😊