GPT-3 — the highly acclaimed, supercomputer-driven auto text generator that has been turning heads for more than a year now — is readying for an upgrade.
Turns out, the next iteration — GPT-4 — will sport 100 trillion parameters, or 500 times more parameters than GPT-3, according to Alberto Romero, a writer for Towards Data Science.
“The sheer size of such a neural network could entail qualitative leaps from GPT-3 we can only imagine,” observes Romero.
That should be something, given that hordes of AI computer scientists have been swooning over the raw power of the software’s current version since its inception.
In a phrase, GPT-3 has spawned hundreds of advanced AI apps — including dozens upon dozens of AI-generated writing programs that are bringing useful — if limited — automated writing to the masses.
One caveat: We’ll probably have to wait a few more years to be gobsmacked by GPT-4 — the current release schedule for the upgrade, according to Romero.
(For an in-depth look at GPT-3, check out: “GPT-3 and AI Writing: Stunning, if Imperfect,” by Joe Dysart.)
In other AI-generated writing news:
*In-Depth Review: ShortlyAI: SEO info site Yo Motherboard reports the ShortlyAI auto-writer is plenty adept at generating sentences and paragraphs that are well written and organic-sounding.
But unfortunately, the site’s reviewer mostly uncovered disappointments in the software, including:
~ShortlyAI sometimes generates content that is factually incorrect
~The tool sometimes generates content that comes off as gibberish
~ShortlyAI seems to sometimes ‘borrow’ content from Wikipedia without attribution
-The tool’s AI sometimes misunderstands what you want it to generate
Probably most telling about the software was a review ShortlyAI was asked to write about itself — which is included in its entirety on Yo Motherboard.
Essentially: ShortlyAI completely misinterpreted what was requested and instead wrote a highly generalized, 500-word piece about artificial intelligence.
*With AI Writing, Sometimes Smaller May Be Better: On some writing tasks, auto-writing engines less powerful than GPT-3 can actually deliver superior results, according to a new study.
Specifically, Google researchers discovered that auto-writing engine FLAN — which uses 137 billion parameters as opposed to GPT-3’s 175 billion — was better at summarizing some long stories as well as answering some questions with text.
The findings could be good news for researchers and others who are looking to leverage the power of GPT-3 — but can’t afford the training and supercomputing costs needed to make it fly.
*Microsoft Releases New AI Personalized News Feed: Microsoft has rolled-out its own version of an AI-curated news feed — free for all users.
Dubbed Microsoft Start, the new service is designed to source news from 1,000+ publishers.
And it promises to generate a personalized news feed for you, featuring timely updates when and where you want them.
Ideally, the new service will grow increasingly more refined at providing the kind of reading you’re looking for by continually studying and analyzing what you’re clicking on — and then adjusting the headlines it provides accordingly.
*New Firm Promises Easy Access to AI-Generated Writing: Canadian firm Cohere is promising to deliver AI writing and related services to everyday businesses in a simple and uncomplicated way.
The firm — which includes researchers from Google Brain — just snared $40 million in new funding.
“Cohere’s deep technical expertise makes them the perfect team to bring NLP to the global marketplace,” says Mike Volpi, co-founder of Index Ventures — which helped put the funding together.
Volpi adds: “We look forward to working with Cohere as they democratize access to one of the most important technologies of our time.”
*AI Writing Training Firm Snags $50 Million: Writing coach firm NoRedInk has pulled in significant new funding to further its quest to nurture better writing in students.
The company uses AI-driven adaptive learning to analyze a student’s writing strengths and weaknesses — and then create personalized lessons designed to make the student a better writer.
Observes Natasha Mascarenhas, a writer for TechCrunch: “To date, more than 10 billion exercises have been completed on NoRedInk’s practice engine.”
*AI Text Analytics Firm Snapped Up: InMoment has acquired Lexalytics, a firm that “transforms structured data, text, and unstructured data into actionable intelligence,” according to Joyce Wells, a writer for KM World.
Says Jeff Caitlin, Lexalytics CE: “Lexalytics is all about helping customers discover more meaningful intelligence in structured and unstructured data sources.”
Adds Wells: “Brands such as Hootsuite, Transcom Worldwide, Altair, Kaplan, and Biogen rely on Lexalytics to uncover actionable insights from customer, employee, and marketplace interactions.”
*Legal Document Automation Firm Acquired: Clio — maker of a software suite for law firms — has purchased Lawyaw, a legal document automation firm.
Lawyaw “enables law firms to convert their existing legal documents into reusable templates and then provides a tool for editing and populating those templates and generating finished documents in Microsoft Word,” according to Bob Ambrogi, a writer for Lawsites.
Clio’s tools are used everyday by law firms for client interactions, case building, billing and more.
*Round-up: AI-Driven Content Management Tools: Digital Marketing Stream offers an interesting rundown of AI-based tools for writers and other content creators in this piece.
Among the curiosities detailed:
~Atomic Reach, a tool — that among other things — can rewrite content using words and phrases that are preferred by a specific reader
~Hey Orca, a drag-and-drop platform that takes the drudgery out of scheduling and publishing social media content
~Magnolia, a content management analysis tool that presents curated content to readers — based on their past consumption behavior
AI Big Picture: How We’ll Use AI in 20 Years’ Time: Kai-Fu Lee: Stop by this 30-minute podcast for an eye-opening look at what AI might bring us in 20 years’ time.
The show features an interview with Kai-Fu Lee, a highly recognized authority in AI.
Lee scored big as an author a few years back with his AI bestseller, “AI Superpowers: China, Silicon Valley, and the New World Order.”
This time around, Lee discusses key insights from his latest book, “AI 2041: Ten Visions of Our Future,” released Sept 14.
The tome is a collection of fictional short stories about AI’s anticipated future — extrapolated from what Lee knows about the current state of AI — that he wrote with science fiction writer Chen Qiufan.
Share a Link: Please consider sharing a link to https://RobotWritersAI.com from your blog, social media post, publication or emails. More links leading to RobotWritersAI.com helps everyone interested in AI-generated writing.
–Joe Dysart is editor of RobotWritersAI.com and a tech journalist with 20+ years experience. His work has appeared in 150+ publications, including The New York Times and the Financial Times of London.