- The AI Update with Kevin Davis
- Posts
- AI Update: Training AI To Write Like You, AI For Journalism, ChatGPT Custom Instructions
AI Update: Training AI To Write Like You, AI For Journalism, ChatGPT Custom Instructions
Today, we are talking about training ChatGPT, models for specific uses, and ChatGPT's Custom Instructions
Making AI Write Like You: A Guide to Train ChatGPT
Artificial Intelligence (AI) has made remarkable strides in recent years, with chatbots like OpenAI's ChatGPT capable of generating human-like text. However, one area where these AI models often fall short is in capturing a unique personal voice. Fortunately, there are ways to train ChatGPT to adapt your writing style, according to a recent article by Maxwell Timothy on MakeUseOf.
ChatGPT, while capable of composing thoughtful prose on any topic, often lacks a personal touch. Its phrasing can feel generic and impersonal. To make it a true writing companion, Timothy suggests using prompting techniques to train it to mimic your style, pacing, word choices, and tone.
The process starts with defining your writing style. For instance, you might be a detail-oriented author who uses a lively, engaging tone, provides detailed, accessible explanations, and uses analogies to explain key concepts. The challenge is to get ChatGPT to adopt this style.
Timothy presents two primary methods to teach ChatGPT to write like you: training it with samples of your work and using a Personal Preference Map (PPM).
Training with Samples
The first method involves feeding ChatGPT samples of your work that best capture your writing style. The AI model then studies the tone, word choice, sentence structure, pacing, explanation style, and other stylistic elements to mimic your unique voice. You might need to repeat this process with several samples to capture every aspect of your writing style.
Using a Personal Preference Map
The second method involves using a Personal Preference Map (PPM), a key-value list of preferences that ChatGPT can use to produce responses that better align with your preference. In this case, ChatGPT can extract a PPM from written samples to learn about your writing preferences and use it to replicate your writing style on demand.
While the PPM approach is slightly more complex, it offers enormous attention to detail and can get as detailed as possible, far more than anyone can easily discern at first glance. It also offers flexibility and portability, allowing you to tweak the writing style with precision and use the PPM with other AI chatbots.
ChatGPT's versatility means it can be taught to write in practically any style you want. With the right mix of training data, prompts, and feedback, you can transform this AI chatbot into your own writing doppelganger.
Genesis: AI's Mixed Reception in Newsrooms
In the ever-evolving world of technology, Google's latest venture into artificial intelligence has set off alarm bells among journalists. Codenamed "Genesis," this AI writing tool, capable of generating news articles, is being pitched as a personal assistant for journalists, automating the process of information gathering to free up time for other tasks.
But here's the rub: AI models like Genesis have a notorious tendency to hallucinate, churning out imaginary events as real facts. However, Google claims its news-writing tool has a leg up on the competition. Genesis can reportedly pull up details of current and past events, while other AI models like Bard and ChatGPT either avoid commenting on real-time events or simply refrain from answering due to a lack of information in their training data.
Google's strategy seems clear: harness the vast ocean of information, both ongoing and past events, to power Genesis and its information-gathering tools. But not everyone is on board with this brave new world of AI-assisted journalism.
The Wall Street Journal, The Washington Post, and News Corp are among the organizations that have been pitched the idea by Google. But the reception has been far from warm. The main bone of contention? Genesis could undermine the sheer effort and creative thinking that goes into reporting accurate and artful news stories.
As Google plunges into the fiery pit of AI, it's not just the journalistic fraternity that's up in arms. Protests like SAG-AFTRA and the writer's strike have already been fighting for better compensation, and more importantly, against AI snatching away their jobs.
It's clear that the news industry is at an inflection point. AI's influx has already made an impact. Outlets like Insider, The Times, and NPR are ready to experiment with AI writing tools. But the results have been mixed at best.CNET's foray into AI-written finance stories resulted in multiple inaccuracies and "very dumb errors." Gizmodo's AI bot also made serious errors, much to the chagrin of the newsroom staff.
So, will Google's Genesis stand out where others have failed? Can it change the newsroom landscape forever? The jury's still out. But one thing's for sure: Google's AI experiment is a wake-up call for journalists. The future of news writing might be more silicon than ink. And that's a future we need to grapple with, sooner rather than later.
ChatGPT Custom Instructions Example From Twitter…
The latest feature, custom instructions, is absolutely astounding!
You can simply use "Act as a..." and effortlessly receive the desired response from Chat GPT without revisiting the custom instructions settings.
I've attached the custom instruction I crafted for you to test:
— ibrahim uzun (@ibrahimuzn)
4:38 AM • Jul 21, 2023
Images Of The Day
PROMPT: a black and white photo of mountains under a large storm, in the style of luminous shadowing, bold use of impasto, whistlerian, flickr, richly colored skies, cryengine, sublime wilderness — ar 4:3
PROMPT: a woman sitting on top of an old white car with a cowboy hat, in the style of works progress administration (wpa), writer academia, national geographic photo, the san francisco renaissance., uniformly staged images, hasselblad 1600f, wood engraving — ar 15:16
Sincerely, How Did We Do With This Issue?I would really appreciate your feedback to make this newsletter better... |
That’s all for today. Hope you had a great weekend. By the time you’re reading this, I should be in the recovery room after getting my trained T-cells back into my body to fight cancer. I had 3 days of chemo last week, back-to-back. This helps suppress T-cell production so the new cells are not attacked by existing ones.
My fingers are crossed for a minor reaction to the infusion.
I really appreciate all of taking the time to read my newsletter. It helps me learn a great deal more about AI than simply consuming it. I am trying to stay ahead of the publishing schedule, but not sure what this week will bring.