Wednesday 21 August 2024

3 key insights from the release of "Simple Bench - Basic Reasoning" LLM benchmark













Background











Many of the best and brightest minds in AI have said that current benchmarks are not fit for purpose and are losing their utility. Popular benchmarks such as MMLU, HumanEval, DROP, GSM8K, HellaSwag and others have recently been "saturated" by the most intelligent models and many people are calling for better benchmarks to help propel the industry further forward. A number of benchmarks show models crushing human performance even when it's obvious that model intelligence and reasoning capability isn't quite at human level performance across the board. Those with a vested interested in the current progress of AI would like a better understanding of how model intelligence and reasoning capability is progressing with each new model release. It would be incredibly helpful if we could more precisely gauge model progress relative to human level performance.

In comes Philip of AI Explained to the potential rescue. Yesterday he released Simple Bench which is a basic reasoning benchmark. He created it because he couldn't find a good enough reasoning benchmark where the questions, phrased in english, could be easily and correctly answered by normal people but current frontier models might struggle with answering them due to their limited reasoning capabilities at present. 

Philip has previously gone into detail about the problems he's found within benchmarks such as MMLU.


Insight #1 - Anthropic has the edge


Notice that GPT-4 Turbo-Preview (which was recently replaced by the newer GPT-4o) is sitting in 2nd place whereas GPT-4o is only sitting in 7th place with a difference of 10% in it's basic reasoning capability. In their May 13, 2024 blog post, OpenAI market GPT-4o by saying it "matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models." Based on Philip's Simple Bench scores this doesn't appear to be the case. He says that he's almost certain that OpenAI made GPT-4o a lighter model with fewer parameters (making it cheaper and quicker to run). That better price-to-performance ratio though has trade-offs in that it appears to have lost some reasoning capability.

The exception to this paradigm (where newer models being cheaper and faster but less capable with reasoning) is Anthropic's latest models. Claude 3 Opus is the biggest model from Anthropic and is sitting in 3rd place where as Claude 3.5 Sonnet (which is not the biggest model and therefore faster and cheaper) is sitting in 1st place and has slightly better reasoning performance. Philip says this is "unambiguous evidence that Anthropic has the secrete sauce currently with LLMs" given how it is able to push speed, cost and intelligence all in the right directions without an apparent trade-off. The Simple Bench score that Claude 3.5 Opus receives when released will be very telling and may further cement Anthropic's innovation edge in the LLM space.

My key takeaway


If you're deciding to buy and rollout a chatbot within your organisation or are beginning to invest in an LLM API to build products on, I'd pay particular attention to whether Anthropic does in fact continue to have the edge when Claude 3.5 Opus is released. I'd also keep tabs on OpenAI's next model release and see whether they reverse their current trend of sacrificing the model's reasoning capability for improvements in cost and speed.

Insight #2 - Models hallucinate but they also engage in sophistry


Current models can often pick up on important facts and even sometimes inform us of the importance of those facts in answering a the question but models aren't always able to link those facts together properly in order to come up with the correct answer. Its as if models can recall the facts but not always reason about them effectively. Take this example that Philip of AI Explained has hand crafted and then given to various models to answer and observes the results:

Some of the results:
  • Gemini 1.5 Pro is able to make the connection that "the bald man in the mirror is John" but then still gets the final answer wrong by saying that "John should apologise to the bald man" even though it's himself and thus one does not need to apologise to oneself
  • Claude 3.5 Sonnet says "The key realization here is that the 'bald man' John sees in the mirror is actually John himself. He's looking at his own reflection, and the lightbulb has hit him on the head." This is a good result however Claude then decides to eliminate an answer saying "C) is incorrect because someone else (the bald man) did get hurt"  
Philip says he "sees these illogicalities all the time when testing Simple Bench" and goes on to say "models have snippets of good reasoning but often can't piece them together.". What seems to be happening is that "the right weights of the model are activated to at least say something about the situation but not factor it in or analyse it or think about it." Philip goes onto say that the paper Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought  Prompting highlights that "models favour fluency and coherency over factuality" and interprets this as "models don't quite weigh information properly and therefore don't always have the right level of rationality that we'd expect."

My key takeaway


This isn't hallucination per se but it's another form of "sophistry" where the model can unintentionally deceive. The model is confidently recalling true and relevant facts but then sometimes follows this on by making blatant reasoning errors that a normal human probably wouldn't make themselves. When building products in various domains where accuracy and user trust is paramount these reasoning errors will need to be mitigated against. 

Insight #3 - Slight variations in wording strongly affect performance



Philip explains that the above paper describes the scenario where "slight variations in the wording of the questions cause quite dramatic changes in performance across all models. Slight changes in wording triggers slightly different weights. It almost goes without saying that the more the models are truly reasoning the less the difference there should be in the wording of the question."

Given this sensitivity to specific wording, he goes onto describe the next potential paradigm with LLMs where when an LLM is given a prompt is can effectively rewrite it to be more optimal. The LLM can effectively "actively search for a new prompt to use on itself, a sort of active inference in a way" which produces better results than the user's original prompt.

My key takeaway


  • Iterating on prompts by changing the wording matters and this should be built into the product development and improvement process your team operate within. Evals really help here as they allow for more systematic iteration on prompts since they allow for repeated performance evaluation across a range of outcomes (not just a select few) that matter most to the users of your product. The workflow for your team could be:
    • Write a prompt
    • Run the eval system and record it's overall score
    • Iterate on that prompt
    • Re-run the eval system
    • If the overall score improves adopt the prompt change, if not continue to iterate on it or move onto other work if you're unable to squeeze out more performance
  • It's hard to know for sure but will AI labs train future models to be more resilient to these wording variations. If Philip is right this seems likely given that for a model to "truly reason" differences in wording should not produce wildly difference levels of performance. In the meantime how should a product be built? There are a plethora of prompt rewrite libraries to choose from but these feel like more of a plaster solution than an actual solution itself.

Thursday 18 July 2024

The new AI reality: what's happening now and what's next?

 (This was re-posted from a blog post I recently wrote on Orbital Witness's Tech Blog)

---

Annual keynote to law firm real estate partners














Context

A year ago Orbital Witness held our annual event where I spoke about “Generative AI: Opportunities and risks for property transactions”. Earlier this month, we continued this tradition by hosting ‘The AI Edge: Real-world Lessons for Real Estate Lawyers’. Taking place at Google HQ, we brought together real estate partners from some of the leading law firms in the UK to share the latest developments in Generative AI and how it is currently revolutionising real estate legal.

I gave the keynote presentation which started by setting the scene for where we are on the innovation S-Curve. I then delved into a range of important aspects of Generative AI for real estate lawyers such as model intelligence, context window size, model cost & speed, proprietary vs open-weight models, AI Agents, multi-modality and use cases in real estate legal. During the discussion, I was also able to contextualise for the audience the trajectory of Orbital Copilot, our own AI legal assistant, as we continue to innovate the product and as Generative AI advances.

This visual sums up the incredible achievement of what’s now possible at the bleeding edge of Generative AI when combining large language models (LLMs) within an AI Agent framework and focusing on a specific practice area, real estate legal, in order to provide turnkey solutions to customers:













Presentation

Here is the full 40 minute video of the keynote:

Slide Deck

Here is the complete slide deck I presented for my keynote with references below:

Slide 6: https://www.tooltester.com/en/blog/chatgpt-statistics

Slide 11: https://www.ben-evans.com/presentations

Slide 19: https://public.flourish.studio/visualisation/18163738

Slide 24: https://www.linkedin.com/feed/update/urn:li:activity:7183400918934016001

Slide 25: https://twitter.com/AIExplainedYT/status/1793561610730320338

Slide 27: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4389233

Slide 28: https://www.youtube.com/watch?v=PeSNEXKxarU

Slide 29: https://mmmu-benchmark.github.io

Slide 32: https://x.com/LibertyRPF/status/1658497036080017408

Slide 42: https://www.linkedin.com/feed/update/urn:li:activity:7183501457684365314

Slide 42: https://www.youtube.com/watch?v=DQacCB9tDaw

Slide 44: https://preshing.com/20120208/a-look-back-at-single-threaded-cpu-performance

Slide 47: https://ai.meta.com/blog/meta-llama-3

Slide 48: https://x.com/maximelabonne/status/1790519226677026831

Slide 52: https://tech.orbitalwitness.com/posts/2023-06-27-genai-opportunities-and-risks-for-property-transactions

Slide 53: https://tech.orbitalwitness.com/posts/2024-01-10-we-built-an-ai-agent-that-thinks-like-a-real-estate-lawyer

Slide 55: https://lilianweng.github.io/posts/2023-06-23-agent

Slide 56: https://sierra.ai

Slide 58: https://www.paulweiss.com/resources/podcasts/waking-up-with-ai/waking-up-with-ai-list/2024/april/ep-6-autonomous-ai-agents-are-a-hot-topic-for-2024

Slide 69: https://www.ben-evans.com/benedictevans/2024/4/19/looking-for-ai-use-cases

Saturday 1 June 2024

Teaching my eleven year old’s class about AI

(This was re-posted from a blog post I recently wrote on Orbital Witness's Tech Blog)

---





















What happens when you’re asked to teach 30 eleven year olds about AI? I did just that at my son’s school last month. Here’s what I learnt.

Background

My son came home a few months ago and said “Daddy, you need to come to my class at school very, very soon and teach us everything about AI.”

That’s clearly no small feat trying to unpack the complexity of artificial intelligence (AI) for an audience of eleven year olds but I was up for the challenge. I wanted to strike the right balance between being informative about what powers AI and how it works but also show them the creative ways AI is being applied in the world right now (because that’s the really fun stuff). I set out to create a visual presentation that would achieve that along with providing ample time for the myriad of questions they would inevitably pop up throughout the lesson.

Presentation

Here’s the 37 slide deck that I used for around 30 children. I was allotted an hour but due to all the questions, from both students and teachers, we ended up spending an hour and half in total getting into some of the nitty gritty of AI:

Some observations

From the ethics around AI, to what it means for the future of work and creativity, here’s some of the themes that came out of our time together:

  1. The kids I taught had a real thirst for a deeper understanding of AI. They really wanted to know how it worked, what it could help them accomplish, what the risks and ethical considerations were with using it. They even wanted to know why exactly AI doesn’t always say naughty things they sometimes ask it to say.
  2. I was pleasantly surprised how philosophical some kids were evidenced by questions such as “given that you’ve been coding for years and you’re now coding AI, did you ever think that maybe someone is actually coding you…?”
  3. There was a bit of an undercurrent of what they do career-wise if AI is going to end up being able to do anything. If they’re interested in music making or art or coding and AI ends up being good at those, is it worth still pursuing those interests…?

Overall it was a fascinating lesson not only for them to learn about AI but also for me to see how they perceive the latest developments in this technology. This tidal wave of innovation is underway and it’s already impacting a future generation of creators and builders who will enter the workforce in a decade from now. And – if you end up using this deck to educate your own children, let me know. I’d love to hear what you learn from the experience.

Wednesday 31 January 2024

We built an AI Assistant, Orbital Copilot, that thinks like a real estate lawyer!

 It uses the latest agentic GenAI concepts and is accelerating lawyers' work by 70%













(This was re-posted from a blog post I recently wrote on Orbital Witness's Tech Blog just after we launched Orbital Copilot)

Executive Summary

  • Innovative Partnership: Orbital Witness and Bryan Cave Leighton Paisner (BCLP) collaborate to bring Orbital Copilot to the forefront of real estate legal practice
  • Evolution of AI in LegalTech: The transition from classical machine learning (ML) to Large Language Models (LLMs) and now the rise of AI Agents transforms what is possible with Generative AI
  • Orbital Copilot’s Capabilities: Offers unparalleled analysis, review, and reporting for real estate legal documents, giving lawyers up to 70% in time savings for lease reporting and other tasks
  • Global Expansion: Following the UK success, Orbital Copilot will expand to the U.S. and other sophisticated real estate markets
  • Customer Cohort: Prestigious law firms are among the first adopters, demonstrating trust and confidence in Orbital Copilot

Introduction


Two months ago we announced our partnership with Bryan Cave Leighton Paisner (BCLP). It was one of many significant steps that has enabled us to bring our AI Assistant, Orbital Copilot, to the forefront of real estate legal practice and deliver a significant impact for our customers.

The genesis of Orbital Copilot dates to mid-2022, born out of our data scientists’ exploratory work with Generative AI, harnessing the power of Google’s BERT and T5 language models. The landscape suddenly shifted with OpenAI’s introduction of ChatGPT in November 2022 and the subsequent release of GPT-4. These advancements fueled our research and development (R&D) of cutting-edge internal tools, quickly proving indispensable to our legal engineers. Our presentation in June 2023, “Generative AI: Opportunities and Risks for Property Transactions,” was a turning point. Our law firms immediately recognised the potential of the GenAI tools we had developed, eagerly requesting access as soon as they were available. Responding to this demand, our team rapidly advanced from a Closed Alpha of Orbital Copilot, meticulously refined with invaluable feedback from a select group of early adopters.

This rapid evolution from an R&D concept to a practical tool for real estate lawyers epitomises our agility, innovation and relentless commitment to delivering genuinely valuable products to our customers. Orbital Copilot today stands as the first product of its kind in the industry: a real estate, domain-specific AI Agent that offers unparalleled analysis, review, and reporting capabilities across multiple real estate legal documents.

In this blog post, we will uncover the unique aspects that make Orbital Copilot a groundbreaking product. We’ll clarify why it’s more than a rebranded ChatGPT and delve into its potential to revolutionise real estate legal practice.

The evolution of AI in LegalTech


  • Classical Machine Learning (ML): In the early versions of LegalTech, NLP-based solutions primarily relied on classical supervised ML techniques. This involved collecting extensive labelled datasets and training various supervised ML models for text classification and question answering. At Orbital Witness, this was our initial approach, leading to the creation of some of the industry’s most accurate ML models for classifying real estate legal text.
  • Large Language Models (LLMs): The advent of LLMs in 2023 marked a paradigm shift. The traditional dependency on vast collections of labelled data for tasks like classification and question answering began to fade. Systems could be built with LLM APIs (such as OpenAI’s GPT-3.5 or Anthropic’s Claude or Google’s Gemini Pro) where they are given a portion of legal text, the “context”, along with specific instructions, the “prompt”, and the underlying LLM generates relevant responses. These systems can be advanced further by incorporating techniques such as retrieval-augmented generation (RAG) which enhances their ability to manage and interpret extensive context from multiple, lengthy documents. This is the most common type of system that many companies are currently building. Such a system is good for simplistic tasks but has several limitations when applied to the type of work lawyers typically do when performing due diligence. As the LegalTech AI landscape evolves, we will continue to see many more products built that summarise documents or answer specific one-off questions. The technology to perform these types of simplistic tasks is becoming well understood. The real long-term value is a dynamic AI assistant built on the AI Agent architectural pattern.


  • AI Agents: Highlighted by OpenAI’s CEO, Sam Altman, at OpenAI’s DevDay in November 2023 (referenced in this Financial Times article), the future of AI Agents was brought into the spotlight. Altman’s announcement of customisable “GPTs” and the prospective “GPT Store” (a marketplace for ChatGPT-based chatbots) signaled a new era. While current “GPTs” are relatively basic AI Agents, they hint at a transformative trend: AI-powered ‘agents’ capable of autonomously performing tasks and radically improving what is now possible with LLM-based applications. In LegalTech, imagine an AI Agent that, from a single query like “What is the rent for this property?”, could pinpoint relevant details across multiple documents, such as leases and deeds of variation, and then logically deduce the answer, mimicking a lawyer’s analytical process.

What defines an AI Agent?


In exploring the concept of AI Agents, Lilian Weng’s insightful blog post stands out, where she defines them as “LLM Powered Autonomous Agents”. Weng emphasises that the core of an AI Agent is the LLM, functioning as its ‘brain’, complemented by three critical components: Planning, Memory, and Tool Use.


















This architecture, when expertly constructed, harbours the potential to transform the legal field. It transcends the more simplistic notion of being merely a thin wrapper around GPT-4 or a “ChatGPT for lawyers”. An AI Agent can be equipped with specialised tools designed for intricate real estate legal tasks such as looking up specific data in a land registry or determining how rent provisions might be varied by another document. The “LLM brain” skillfully determines the optimal use of these tools to execute tasks with precision and depth, as directed by a legal professional. Moreover, the AI Agent possesses the ability to reason about its generated outputs. This ability enables a real estate legal AI Agent to decide between several potential outcomes:

  • Continue to delve deeper into the legal documents at it’s disposal to find a more fitting answer
  • Request additional real estate documents or data, it feels are missing, to more comprehensively answer the question
  • Decide it has a complete answer and present a valid response back to the lawyer who initiated the instruction

The Significance of AI Agents


The concept of AI Agents has been gaining substantial attention, particularly highlighted by OpenAI’s CEO, Sam Altman (see this Financial Times article). Altman underscores the significance of these AI agents in the overall AI landscape, with OpenAI’s upcoming GPT Store being a testament to their commitment to this platform shift. Ethan Mollick further explores this idea in his post Almost an Agent: What GPTs can do, where he states:

“In their reveal of GPTs, OpenAI clearly indicated that this was just the start… GPTs can be easily integrated into with other systems, such as your email, a travel site, or corporate payment software. You can start to see the birth of true agents as a result. It is easy to design GPTs that can, for example, handle expense reports. It would have permission to look through all your credit card data and emails for likely expenses, write up a report in the right format, submit it to the appropriate authorities, and monitor your bank account to ensure payment. And you can imagine even more ambitious autonomous agents that are given a goal … and carry that out in whatever way they see fit.”

Olivia Moore, a Consumer Partner at venture capital firm a16z, further echoed the potential of AI Agents in a recent tweet at the end of 2023:


The Impact of a Real Estate Legal AI Agent: Orbital Copilot


Orbital Copilot, our innovative AI Agent at Orbital Witness, is revolutionising the way real estate legal work is conducted. It closely emulates the tasks of real estate lawyers, who often wade through extensive legal documents to perform due diligence for their clients. Here’s how Orbital Copilot is transforming the field:

  • Comprehensive Document Analysis: It can digest hundreds of pages of intricate legal text across numerous PDF documents whether typed, written manuscript or both
  • Diverse Question Resolution: From straightforward questions like “What is the date of the lease?” to more complex queries such as “How has the service charge varied?”, Orbital Copilot handles them all
  • Contextual Understanding: It tracks down definitions within documents to enhance understanding and reasoning
  • Thorough Information Gathering: Whether it’s following the breadcrumb trail across one or several documents, it ensures all necessary information is collated
  • Supplementary Research: It seeks out additional legal information to refine its understanding of the lawyer’s initial instructions. This could be a proprietary legal knowledge base or data from HM Land Registry
  • Targeted Summarisation: Orbital Copilot can summarise entire documents or specific sections across multiple documents
  • Language Simplification: It adeptly rephrases complex legal jargon into layman’s terms for client comprehensibility
  • Trusted Referencing: Orbital Copilot indicates the parts of the PDF documents it consulted, facilitating direct navigation to the primary evidence supporting its answers so lawyers can immediately see and trust where an answer came from
  • Transparent Reasoning: Like consulting a junior lawyer to ask how they came to their conclusion, it transparently reveals its thought process and how it arrived at specific conclusions so a lawyer can focus on the legal nature of the problem and not some “black box” technology

Specifically tailored for real estate legal tasks, Orbital Copilot’s combination of features yields remarkable time savings for lawyers. Considering the busy schedules of legal professionals, often billing in six-minute increments, Orbital Copilot’s efficiency is a game-changer. Our thorough testing with top-tier UK law firms, involving real client work, revealed that Orbital Copilot can reduce the time for a comprehensive lease report by up to 70%. Given that a single property’s lease report can take 2-10+ hours depending on complexity, this translates to substantial financial savings per property for law firms and their clients. Given the regularity of lease reports in real estate law, the cumulative efficiency and cost savings are substantial.

A Glimpse into Orbital Copilot’s Functionality


Let’s take a closer look at how Orbital Copilot operates in practice. Consider this scenario where two key PDF documents are uploaded:

  • Lease dated 06-06-2008
  • Deed of variation dated 31-03-2016

In this instance, the deed modifies several aspects of the lease, including the rent. When prompted with the query “What is the rent and how has it been varied?”, Orbital Copilot leaps into action. It begins by understanding the question’s context and the documents at hand. Then, it meticulously searches and reads the pertinent sections in both documents. Finally, Orbital Copilot analyses its findings, formulates a response, and presents it for review.


Another illustration of Orbital Copilot’s capabilities is shown in the processing of a short form lease report, which includes 10 targeted questions. It’s important to note that Orbital Copilot is equipped with a variety of pre-configured lease reports, ranging from basic to highly detailed enquiries about the legal documents. The next video demonstrates the types of questions posed in the short form lease report. It also showcases how users can easily reference specific parts of the underlying documents for additional context or to validate Orbital Copilot’s responses:


Engineering challenges


At Orbital Witness, leveraging the most advanced LLMs like GPT-4 is essential to meet the high standards required for legal document analysis. However, this approach presents several engineering challenges:

  • Cost Management: Utilising state-of-the-art LLMs for thorough analysis of extensive legal documents, often running into hundreds of pages, is crucial for achieving the accuracy our lawyer clients depend on. However, the use of such advanced technology incurs significant costs. Although we anticipate a decrease in expenses as Nvidia ramps up GPU production and AI labs enhance LLM efficiency, the current challenge lies in optimising our LLM usage to maintain a balance between cost-effectiveness and high-quality output.
  • Resource Availability: The global shortage of Nvidia GPUs, coupled with the soaring demand for LLM functionalities, has compelled LLM providers to impose caps on the number of tokens (akin to words) processed per minute through their APIs. This limitation affects our capacity to onboard new customers and influences the execution speed of tasks within Orbital Copilot. While we expect this issue to diminish as GPU availability increases and LLMs become more efficient, it remains a significant short-term constraint that requires careful management.
  • Ensuring Reliability: Many LLM providers, despite their technological prowess, are relatively new to managing complex, fault-tolerant services on a global scale. This inexperience can lead to occasional service fragility, manifesting as uptime issues and performance degradation. Such challenges directly impact our operations, necessitating continuous vigilance and adaptability to maintain uninterrupted service quality.

The Future of Orbital Copilot


The trajectory of Orbital Copilot is set to reach remarkable milestones. Our recent collaboration with Bryan Cave Leighton Paisner (BCLP) is a testament to this. Through our “global design partnership,” we’re extending our lease reporting capabilities, initially honed in the UK, to BCLP’s real estate practices in both the U.K. and the U.S. This marks Orbital Witness’ ambitious leap from a UK-centric operation to a transatlantic presence in 2024. Our existing clients are already expressing eagerness to harness Orbital Copilot’s benefits on a global scale.

At our core, we are a product-centric company, deeply invested in understanding and addressing our customers’ needs. This customer-first approach drives our product development, guiding us in crafting a roadmap that tackles their most pressing challenges. Currently, we are channelling our energies into developing some groundbreaking features, slated for release in the first and second quarters of 2024. These upcoming enhancements are poised to further revolutionise the landscape of real estate legal technology, strengthening Orbital Copilot’s position as a trailblazer in the field.

Launching with an Esteemed Customer Cohort


Emerging from a highly successful private closed beta in the final quarter of 2023, we at Orbital Witness have quickly transitioned to welcoming our first batch of paying customers. This group, having been on our eagerly anticipated waitlist, represents a diverse array of prestigious companies. Their readiness to adopt Orbital Copilot’s AI Agent speaks volumes about their commitment to embracing the cutting-edge of Generative AI in real estate legal work. We are immensely proud and excited to collaborate closely with these industry leaders, each a prominent name in their respective fields:

  • BCLP: Global law firm with 31 offices worldwide and clients who represent 35% of the Fortune 500
  • Clifford Chance: One of the world’s largest law firms, with significant depth and range of resources across five continents
  • Charles Russell Speechlys: International law firm headquartered in London with offices across the UK, Europe, Asia and the Middle East
  • Macfarlanes: A distinctive London-based law firm with a unique combination of services built and shaped around their clients’ needs
  • Ropes and Gray: Global team with 13 offices on three continents and named “Law Firm of the Year” by The American Lawyer in 2022 and ranked number one on The American Lawyer’s A-List of elite firms
  • Walker Morris: Independent law firm with a first-class international reputation
  • Thomson Snell and Passmore: The oldest law firm in operation tracing back to the late 16th century
  • Thompson, Smith and Puxon: Established in 1879, TSP has grown to be one of the leading law firms in Essex
  • Able UK: Market leader in wind energy & marine decommissioning along with being a significant land developer and port & vehicle storage operator

Customer Testimonials


“Orbital Copilot is next generation legal technology and is helping us continue to focus our Real Estate lawyers’ time on the areas that are most valuable to clients. Our IT strategy has always been to seek out the best technology tools for our needs and our collaboration with Orbital Witness is a key plank of our Real Estate AI plan.”

        – Matt Taylor, Partner @ Clifford Chance

“BCLP was an early adopter of Orbital Witness’ products, and we are thrilled to collaborate with the company on an AI solution that will provide marked benefits to our clients by accelerating lease reporting and enhancing real estate due diligence. We’re also excited to see so many of our lawyers deeply engaged in the development of leading-edge technology and experiencing the potential of generative AI firsthand.”

        – Samant Narula, BCLP’s U.K. Head of Real Estate

“The potential of Orbital Copilot is stunning. A time saving AI tool which will increase our efficiency. Icing on the cake of Orbital Witness’ services.”

        – Clive Gotley, Head of Legal @ Gridserve

“Copilot is the kind of tool that every lawyer wishes they had. It enables you to review documents quickly and efficiently, whilst also allowing you to check and verify the information, ensuring that the end result is the one that you want.”

        – Amy Shuttleworth, Associate @ Charles Russell Speechlys

Conclusion: The Future is Here with AI Agents


  • AI Agents - A LegalTech Revolution: The future of LegalTech is being reshaped by AI Agents. Their advanced capabilities and adaptability make them indispensable tools to begin integrating into modern legal practices
  • Orbital Copilot - Leading the Charge: Orbital Witness has pioneered the development of the world’s premier AI Agent tailored for real estate legal work. Orbital Copilot is not just an innovation; it’s a game-changer, already enhancing due diligence and reporting processes by an impressive 70%
  • Customer Acclaim: The response from our customers has been overwhelmingly positive. The efficiency and precision Orbital Copilot brings to their client work have made it an essential component of their legal toolkit
  • Join the Early Adopters: For those eager to be part of this transformative journey, we have a limited number of spots in our early adopters cohort. Interest has been high, so we encourage you to sign up quickly to secure your place as more slots become available

As we move forward, Orbital Copilot continues to set new benchmarks in the realm of real estate legal technology. Stay tuned for more updates and innovations as we navigate this exciting frontier.

Wednesday 1 November 2023

My team has officially shipped something very exciting today!

Back in June 2023 I gave a presentation titled "Generative AI: Opportunities and risks for property transactions" in which I demoed an internal AI Assistant project that my team and I had been building ever since ChatGPT launched to the world in November 2022 and it became abundantly obvious this new Generative AI technology had the potential to transform the legal industry. We received a tremendous amount of engagement off the back of my presentation and much of it was from customers wanting to get their hands on this AI Assistant that we built so they could try it out for themselves. Because our initial AI Assistant was only an internal tool, we need to pull it out of our backend application and productionise it for law firms and other companies so they could begin trialing it and giving us feedback. 

Fast forward a few months once we had ironed out a number of challenging technical problems with building a bleeding edge AI system built on top of the latest large language models (LLMs), our own customised prompt engineering, a retrieval-augmented generation (RAG) system and our own proprietary legal specific document OCR and structuring technology all the while controlling for the eye-watering costs of some of the more powerful LLMs like OpenAI's GPT-4. 

Today BCLP and Orbital Witness have publicly released the details of our "global design partnership to test, optimize and deploy Orbital Witness’ lease reporting capabilities in its latest generative artificial intelligence (AI) solution, Orbital Copilot, across BCLP’s U.K. and U.S. real estate practices." I'm proud to say that the AI Assistant I showcased at my presentation back in June 2023 has now taken on a life of it's own as Orbital Copilot and has a ton more functionality, improved accuracy, ability to work with any legal documents and is initially focused on lease reporting for lawyers and legal professionals as we build this out for BCLP and others. The press release goes on to say: "Orbital Copilot is the first real estate-sector-specific AI assistant created for real estate lawyers that is capable of analyzing and reporting on any real estate document. By leveraging Orbital Witness’ market-leading capabilities in real estate-specific AI and the deep sector knowledge and experience of BCLP’s global real estate practice, this first-in-market collaboration seeks to use this cutting-edge technology to enhance BLCP’s reporting on complex commercial leases, providing accelerated insights to their clients."

I'm also quoted in the press release alongside Samant Narula, BCLP’s U.K. head of real estate: 

Andrew Thompson, chief technology officer at Orbital Witness, says, “A key part of our strategy with Orbital Copilot is to develop real estate-domain-specific AI that enables us to deliver solutions for property professionals that are incomparable in terms of capability, quality and reliability. This collaboration and pilot program with BCLP allows us to demonstrate and test that value at a firm that shares our ambitions and vision for this technology.”

BCLP has one of the largest legal real estate practises in the country, so they're a brilliant law firm to partner with for the ground breaking Generative AI product called Orbital Copilot that we've just built and launched to the world.