As a dyslexic individual, I have been testing assistive technologies most of my life. During high school, I would test different tools that the school system wanted to roll out. I also have ADHD and dysgraphia (a writing disorder). The technologies included the AlphaSmart portable keyboard to help with taking notes due to my dysgraphia, recording devices, and even pins that you could roll over a word in a book to have it read back the definition (which I didn’t particularly enjoy). This was all almost 30 years ago..
In recent years, there has been a new wave of assistive technologies that utilize artificial intelligence (AI) to help individuals with dyslexia and other learning disabilities. These include:
• LLAMA 2: This is an AI-powered tool that can be used for writing assistance, proofreading, and language translation. It can be especially helpful for those with dysgraphia and other writing disabilities.
• Chat GPT: This is an AI-powered tool that can be used for a variety of tasks, including writing assistance, research, and language translation. It can be especially helpful for those with ADHD who may struggle with focusing on one task for an extended period of time.
Here are some possible uses of AI and LLMs in assistive technology:
• Text-to-speech software: This software allows users to type in text and have it read back to them. It can be especially helpful for those with dyslexia who may struggle with reading written text. Natural reader is an example of ai voice generation. I use this to read books and websites. Even E books. This allows for very natural speech that makes it easy to understand.. There is also built in tools in Bing and Microsoft software. However I find it is not as good.
• Assistive writing software: This software provides features such as spell-checking, grammar-checking, and word prediction to help individuals with dysgraphia and other writing disabilities. One example of this is grammarly. However you can also use a LLM to do this also.
• Speech-to-text software: This software allows users to speak their thoughts and have them translated into written text. It can be especially helpful for those with ADHD who may struggle with writing by hand or typing. I do not use this that often however it is built into some Microsoft software.
Overall, AI and LLMs have the potential to revolutionize assistive technology for individuals with dyslexia and other learning disabilities. By providing personalized support and accommodations, these technologies can help level the playing field and provide equal access to education and employment opportunities.
Over the past few months, I have been experimenting with various AI tools such as ChatGPT on OpenAI, Bing Chat built into the Bing browser for quick checks. I have also used these tools to improve the readability and grammar of writing. Currently, I am having it check and rewrite this post to make it more suitable for a global audience.
While AI can generate entire blog posts or papers based on your specific requirements and information, ethical concerns arise regarding plagiarism and academic integrity, especially in a school environment.
After exploring closed-source and online models, I have started to experiment with open, local hosted models.
Here is an example where I had MetaAI’s free Llama 2 fix the first paragraph of this blog post….
Local hosting of generative AI models like LLAMA 2 can offer significant advantages, particularly when it comes to processing power. While these models require a large amount of computing resources to train and run, leveraging cutting-edge GPUs can greatly improve their performance. With my gaming system equipped with an RTX 4070, I have found that I can process AI models much faster than on other systems I have tested, thanks to the powerful CUDA cores. This local hosting approach not only reduces latency and improves responsiveness but also enables more complex and detailed AI-generated content.
One application where I use local AI models is in cleaning bash scripting. AI can help with loops and awks to get certain results and build code, although it is able to write full simple code, it’s important to fully check your self the code and build upon it. Additionally, the AI model is good at error checking your code, which helps me catch any potential issues before they become problems. I have been using Code LLAMA for this purpose and have found it to be very effective.
You may ask, ‘How can I start using AI LLMs?’ There are a couple of free options available. One option is OpenAI Chat GPT, which uses GPT 3.5, Bing Chat, or now called Co-Pilot (limited). Another option is to host your own LLMs.
I found the best way to self-host is by using LLM Studio. This keeps all my data from leaving my systems. It also allows you to pull specific, trained AI LLMs and narrow down what you want it to do. You can follow the setup details and download from the following webpage. However, as noted, you will need a powerful computer to host local LLMs.
See more here LM Studio – Discover, download, and run local LLMs
Dysclamer:
When using AI language models for business purposes, it’s important to consider the potential risks associated with sharing sensitive data. Many of these models use your inputs to train the AI, which means that any sensitive information you provide could be included in public data collections. If you are using AI for a company or with sensitive data, please be aware of this and take appropriate precautions. One option is to use an enterprise version of Bing Copilot, which does not share any data out to training and can help protect your company’s information.
Resorces:
OpenAI Chat GPT: ChatGPT (openai.com)
Meta Llama 2 : Llama 2 – Meta AI
LM Studio: LM Studio – Discover, download, and run local LLMs (Easy way to run local AI models)
The Hugging Face: Models (huggingface.co) (location of trained opensource models)
Bing Co-Pilot: Microsoft Copilot: Your everyday AI companion
Grammarly: Grammarly: Free AI Writing Assistance
Natural Reader: AI Voices – NaturalReader Home (naturalreaders.com)
Update 12/2/2024:
Since this post I have created an additional post on self hosting a web based llm interface. The example I had was hosting on ubuntu vm with ollama.
Host your own Local AI Server | ollama | open-ui [snap] | Ubuntu 22.04
#iworkfordell