Subscribe to GEN
Login to GEN
Language models are good at a small number of things, and very good at some of them: reading and summarising documents, finding patterns in large volumes of text, classifying what someone is asking for, and translating. GEN put those strengths to work on your own documents and data, using models that run on our own hardware or on yours, so none of it is handed to a cloud AI provider.
We are equally clear about what they are bad at. A model that answers confidently from the wrong material is worse than no model at all, so everything we build limits what the model answers from and how far its answers are trusted.
Imagine asking "what is the company policy on sick pay?" and getting the answer, with the paragraph it came from, instead of searching the handbook. Or asking how many sales Brian made between 1 November and 31 January, without pulling three months of reports and adding them up. With the right preparation, a language model can do both.
The preparation is the work. Your documents are converted and indexed; when a question arrives, the passages that matter are found first and given to the model with the question, so it answers from your material rather than from memory. The same approach gives staff a knowledge base that answers technical questions, and, with more tuning and testing, can answer customers through your website or a chatbot.
Large volumes of text or figures analysed for patterns, alongside conventional data science: documents that do or do not follow a required form, scripts and training material checked, and call recordings screened for fraud or script compliance.
Documents read, categorised and summarised at scale, with text extracted from scans and images by OCR, so that paper and PDFs become data other systems can use.
Websites, emails, documents and recordings translated quickly and at far lower cost than traditional translation, though a person should still check anything that will be published or relied upon.
Working out what an email, ticket or caller is asking for and sending it to the right place. We use it ourselves for the first analysis of new HelpDesk tickets.
A model's output becomes the input to something else. In an AI-driven phone menu, the caller's words become text, the model interprets them, and the result is filtered before it decides what the caller hears next.
Just because a task can be given to AI does not mean it should be, particularly where a customer's patience is at stake. Part of every project is deciding where a person, or plain code, does the job better.
You do not need a cloud AI provider to use language models, and for most business data there are good reasons not to. Documents, prompts and answers sent to a public service leave your control, and what happens to them next is set by someone else's terms.
GEN run a catalogue of models on our own hardware, including GAIN, our own platform, and they are integrated with our cloud hosting and virtualisation, so the systems that use them can sit alongside. Where the data must not leave your building, we install the models on your premises instead.
No. GEN run language models on our own hardware, and can install them on your premises instead. For many documents, such as HR records, contracts and customer data, that is the only arrangement that makes sense.
Yes. The documents are indexed so that, when a question is asked, the relevant passages are found first and given to the model with the question. The answer then comes from your material rather than from what the model picked up on the internet, and can point back to its source.
No, and anyone who says otherwise is selling something. Language models make mistakes. We reduce them by limiting what the model answers from, fine-tuning it on your material, and testing it against real questions, and where answers go to customers rather than staff, we add further checks and a route to a person.
The one that suits the job. Some models are better at arithmetic, some at documents, some at images, and some are better suited to tightly controlled chat. We choose from our own catalogue and from GAIN, our own platform, rather than tying you to a single vendor's model.
Yes. Our AI tool, under Tools on this site, lets you put questions to a language model running on our own platform. It is a real model, and it will sometimes be wrong, which is a useful lesson in itself.
Most useful AI projects start with a pile of documents nobody has time to read, or a question the business answers a hundred times a week. Tell us which it is for you, and where the data lives, and we will tell you what a model could do with it, and what it should not be trusted to do.
You can put questions to one of our models now, using the AI tool.