Did ChatGPT Really Cure a Dog’s Cancer? It’s Complicated

OpenAI co-founder Greg Brockman amplified a broadly shared story over the weekend a few canine handled with a customized mRNA most cancers vaccine developed with assist from ChatGPT, drawing consideration throughout tech and AI communities.

The case centres on Rosie, a seven-year-old Shar Pei owned by Australian AI marketing consultant Paul Conyngham.

Based on posts circulating on-line, Rosie had been given solely months to stay earlier than receiving the experimental therapy, which Conyngham mentioned was developed with help from the AI chatbot.

“Again in 2022, I observed unusual lumps on her head,” Conyngham wrote in a November 2024 thread documenting the journey from the start. “What the vet deemed as ‘simply warts’ ended up being late-stage most cancers.” Vets estimated Rosie had between one and 6 months left and instructed Conyngham there was nothing extra they may do.

]]>

The account unfold quickly after Brockman shared it along with his tons of of 1000’s of followers, prompting protection throughout a number of know-how retailers.

Whereas the therapy itself seems real, the function ChatGPT is credited with in creating the vaccine has been debated, with some researchers questioning how a lot of the method might realistically be dealt with by a big language mannequin.

Pushing forward

Conyngham mentioned he did not surrender on Rosie. As an alternative, he determined to construct a analysis pipeline out of shopper AI instruments. He began with ChatGPT, utilizing it to design a plan of assault.

The mannequin instructed him he wanted genomic sequencing, one pattern of wholesome tissue and one from the tumor, and pointed him towards particular establishments and tools.

“Probably the most ironic factor is that in a earlier chat session with ChatGPT, it mentioned that I ought to try to achieve out to Elita or Dr. Martin and that I ought to use an Illumina machine,” he wrote on the time.

So he adopted that lead.

A director at UNSW linked him to Dr. Martin Smith, head of the Ramaciotti Centre for Genomics, who agreed to sequence Rosie’s genome for round $3,000.

Ten days. Thirty-times depth in wholesome tissue, 60-times in tumor: the upper cross charge wanted to isolate the mutations driving the most cancers. The Centre returned 320 gigabytes of uncooked information.

Genomic data is expressed in strings of the letters A, T, C, and G, so consultants basically ended up with a stack of 700,000 double-sided pages filled with solely these 4 letters, the College of New South Wales reported in June of final 12 months. That was Rosie’s genome, her organic fingerprint.

He then targeted on c-KIT, a protein well-documented within the printed literature on mast cell tumors in canines.

Utilizing Google’s AlphaFold, he modeled Rosie’s model of the protein and in contrast it in opposition to the wholesome baseline. It seemed incorrect, mutated in ways in which matched what the literature predicted. He then looked for present compounds that may assault c-KIT or proteins just like it, and located one: a drug already in use within the U.S. to deal with a distinct most cancers in people.

“We took her tumour, sequenced the DNA, we transformed it from tissue to information, and we used that to search out the issue in her DNA after which develop a treatment based mostly on that,” Conyngham instructed the Australian At the moment Present on Saturday. “ChatGPT assisted all through that total course of.”

AI’s true function

Even so, there’s a giant hole between ChatGPT discovering a treatment for most cancers and ChatGPT helping in analysis.

Conyngham finally linked with Prof. Palli Thordarson, Director of the us RNA Institute. “Prof. @martinalexsmith carried out the DNA/RNA sequencing to transform Rosie’s tissue into uncooked information,” Conyngham posted. “Prof @PalliThordarson assembled the mRNA vaccine,” he added in one other tweet.

Thordarson confirmed this in his personal thread: “Proud with @UNSWRNA to have been concerned & making the mRNA-LNP for Rosie,” he wrote on X on Sunday. “The intersection of RNA know-how, genomic & AI poses a possibility to vary the way in which we do medication and make entry extra equitable.”

However Dr. Smith wasn’t a person behind a ChatGPT display screen. He was a professor operating a college RNA institute, doing what his lab was constructed to do.

And when Conyngham recognized the ultimate vaccine assemble—the precise molecular blueprint that will be encoded into the mRNA—he revealed which instrument designed it. Not AlphaFold. Not ChatGPT. “The ultimate vaccine assemble for Rose was designed by Grok.”

That mentioned, he acknowledged in a separate put up that “Gemini did a ton of the heavy lifting too.”

ChatGPT was used to sift by scientific papers and establish researchers who would possibly be capable of assist. The chatbot pointed to the Ramaciotti Centre and instructed sequencing tools suited to the duty, functioning largely as a instrument for navigating the analysis literature. That function might be helpful, however it differs from designing a vaccine or performing scientific evaluation.

AlphaFold, a deep-learning system from Google DeepMind, predicts three-dimensional protein constructions from amino acid sequences. It’s not the primary mannequin educated on organic information: different open-source initiatives like Ankh or AlphaGenome work on related premises.

Conyngham used Alphafold to mannequin Rosie’s c-KIT protein. The rendering carried a confidence rating of 54.55, which UNSW structural biologist Dr. Kate Michie publicly described as low.

She famous that AlphaFold “can get stuff incorrect” and that vital lab work is required to validate any output. Dr. Smith, the us genomics director, confirmed publicly in the identical thread that AlphaFold was not, in reality, used for the mRNA vaccine design in any respect.

Dr. Thordarson was cautious in regards to the framing, too.

“This may increasingly not have cured Rosie,” he wrote on X. “Purchased time for positive, sure, however a few of the tumours did not reply.”

His crew is now checking whether or not these tumors mutated in another way, which might clarify why elements of the therapy labored and others didn’t. The vaccine additionally didn’t work in isolation.

“The therapy required co-administration of a checkpoint inhibitor,” Thordarson famous, “more likely to be with all customized most cancers vaccines.”

The usage of AI for most cancers therapy has not at all times been a historical past of success.

In 2017, inside IBM paperwork revealed that Watson for Oncology, marketed as a system that would advocate most cancers remedies higher than human oncologists, was producing what its personal engineers flagged as “unsafe and incorrect” suggestions.

MD Anderson Most cancers Heart deserted the undertaking after spending $62 million on it. IBM bought off Watson Well being in its entirety in 2022.

The Rosie case doesn’t fall into the class of AI failures. Nobody was harmed, the underlying science is established, and the researchers concerned have acknowledged credentials.

The mRNA platform itself is supported by scientific analysis. The unease lies extra in how the story has been framed. When AI instruments obtain credit score for work carried out by scientists and analysis establishments, it could possibly blur public understanding of what the know-how truly does.

The researchers who carried out the sequencing, produced the vaccine, and managed the protection protocols threat fading into the background.

The episode affords a reminder that AI can help with duties equivalent to navigating scientific literature, however it stays removed from changing the experience and infrastructure required to design and produce medical remedies.