Hello and welcome to Eye on AI.
It’s been a busy week for OpenAI. Coming off its product-packed Dev Day conference, the company followed up with an announcement that offers a window into the next stage of its data collection plans while simultaneously battling a DDoS attack that caused major disruptions to ChatGPT and other OpenAI tools. We also saw the very beginnings of an ecosystem for GPTs, custom ChatGPT-like models users can create for their own specific purposes.
OpenAI on Thursday announced Data Partnerships, a program for collaborating with third parties to build datasets for training AI models. A blog post introducing the program says the company is “interested in large-scale datasets that reflect human society and that are not already easily accessible online to the public today.”
“To ultimately make AGI that is safe and beneficial to all of humanity, we’d like AI models to deeply understand all subject matters, industries, cultures, and languages, which requires as broad a training dataset as possible. Including your content can make AI models more helpful to you by increasing their understanding of your domain," reads the blog post, adding that the company is “particularly looking for data that expresses human intention.”
It goes on to state that OpenAI is seeking partners to help it create an open-source dataset for training LLMs as well as private datasets for training proprietary AI models, including its own foundation, fine-tuned, and custom models. It’s not exactly clear what participating in this program looks like (there’s no mention of compensation, data ownership, etc.). But it’s clear the private dataset efforts will go beyond empowering AGI in theory and will directly enrich OpenAI’s own models, and therefore its bottom line. It’s also impossible to ignore the drumbeat of the call for data that isn’t “already easily accessible online to the public today,” seeing as OpenAI already scraped the internet as it pleased, and is now facing an onslaught of lawsuits as a result.
While this announcement was trickling into the world, Open AI was simultaneously fighting off a targeted DDoS attack that struck just two days after its first-ever in-person event, interrupting service for ChatGPT and other OpenAI tools for more than 24 hours. But despite this, users started gaining access to GPTs, building their own models, and sharing them with the world.
Kristina Martin, a marketer and educator, has already made three GPTs: one to aid K-12 educators in writing progress report comments, an assistant for grading MLA English composition papers, and a brand-building guide for solopreneurs. She told Eye on AI that she found the builder “relatively easy to use,” though she did have to actively test and refine the GPTs to get an outcome close to what she envisioned, and she never quite achieved the type of decision-tree line of questioning she was hoping to achieve with the brand-building model.
“They are fun to build, and I do think they can help with certain tasks,” she told Eye on AI. “However, those tasks can't be too complex. As a personal branding and content strategist, I would typically ask my clients a series of questions in a 1:1 interview and ask additional questions based on their responses. I think the GPT builder needs to be more sophisticated to handle those types of tasks.”
While OpenAI has yet to release the GPT Store, which it says will roll out later this month, users jumped at the opportunity to compile GPTs themselves and have already created various public lists and searchable GPT repositories. I discovered and experimented with a wide range of single-function GPTs doing everything from creating playlists (like PlaylistAI:Spotify) to emulating specific people (like Paul Graham and Elon Musk).
Most notably, I saw a large swath of purely playful GPTs that reminded me of the earlier days of the internet. For example, there’s FortuneTeller GPT (described as a “cryptic and mystical AI agent”), OppositeDay GPT (which answers inquiries with the opposite of the truth), RoastMeGPT (self-explanatory), and PessimismGPT (which had some interesting things to say about OpenAI). There’s even PirateSpeakGPT, which communicates exclusively in the style of a stereotypical pirate and elicits memories of when everyone discovered how to change their language to “Pirate” on Facebook.
These are far from Earth-shattering use cases, but they do show how GPTs are opening up new lanes for people to get creative on an internet that often feels as if it’s lost its magic. Not to mention, it is pretty amazing that people with zero technical expertise can create these models in minutes using only natural language—not a single line of code is required.
Now we just have to wait for the killer app, or in this case, the killer GPTs.
And with that, here’s the rest of this week’s AI news.
Reminder: Brainstorm AI is just weeks away!
Gain vital insights on how the most powerful and far-reaching technology of our time is changing businesses, transforming society, and impacting our future. Join us in San Francisco on Dec. 11–12 for Fortune’s third annual Brainstorm AI conference. Confirmed speakers include such A.I. luminaries as Google Assistant and Google Bard GM Sissie Hsiao, IBM Chief Privacy and Trust Officer Christina Montgomery, Walmart International SVP and CTO Sravana Karnati, Pfizer Chief Digital and Technology Officer Lidia Fonseca, and many more. Apply to attend today!
And with that, here’s the rest of this week’s AI news.
Sage Lazzaro
sage.lazzaro@consultant.fortune.com
sagelazzaro.com
